Login| Sign Up| Help| Contact|

Patent Searching and Data


Title:
METHOD FOR CALCULATING A SEVERITY INDEX IN TRAUMATIC FRACTURES AND CORRESPONDING SYSTEM IMPLEMENTING THE METHOD
Document Type and Number:
WIPO Patent Application WO/2021/250710
Kind Code:
A1
Abstract:
The present invention relates to a method for calculating a severity index in fractures of a humerus of an individual, from a plurality of section images Sk of said humerus, where k=1,...,N with N which is a positive integer, captured by means of a diagnostic imaging technique, characterized in that it comprises the following steps: A. normalizing the grey levels of each section image Sk of said plurality of images Sk, so that each section image Sk has the same grey level; B. starting from said normalized images in said step A., segmenting each section image Sk, in a significative region, comprising at least one bone structure, wherein at each pixel p is assigned a value 1, and in a non significative region, wherein at each pixel p is assigned value 0, so that each section image Sk is a binary image; C. starting from said section images Sk segmented in said step B., identifying at least one connected component corresponding to said humerus, or a shoulder blade, in said at least one bone structure comprised in each segmented section image Sk, and labelling said at least one connected component with a respective label, so that said humerus is labelled with a first label and said shoulder blade is labelled with a second label which is different from said first label; D. identifying a plurality of fragments of the head of said humerus, labelling each fragment of said plurality of fragments with a respective label and generating a real model (0) of said humerus from said plurality of fragments; E. given a reference model (M) of the integral bone structure of said humerus, recomposing and aligning said plurality of fragments, identified in said step D., in said real model (0) according to said reference model (M); and F. calculating a severity index of said fracture of said humerus starting from said plurality of fragments recomposed and aligned in said step E. The present invention also relates to a system which implements said method.

Inventors:
FIORENTINO FABRIZIO (IT)
PIETROLUONGO LIVIA RENATA (IT)
RUSSO RAFFAELE (IT)
RICCIO DANIEL (JM)
ROSSI SILVIA (IT)
Application Number:
PCT/IT2021/050173
Publication Date:
December 16, 2021
Filing Date:
June 08, 2021
Export Citation:
Click for automatic bibliography generation   Help
Assignee:
E LISA S R L (IT)
International Classes:
G06T7/00; A61B5/00; A61B5/103; A61B34/10
Foreign References:
US20070191741A12007-08-16
US20180055644A12018-03-01
Other References:
RAFFAELE RUSSO ET AL: "A new classification of impacted proximal humerus fractures based on the morpho-volumetric evaluation of humeral head bone loss with a 3D model", JOURNAL OF SHOULDER AND ELBOW SURGERY, vol. 29, no. 10, October 2020 (2020-10-01), AMSTERDAM, NL, pages e374 - e385, XP055768856, ISSN: 1058-2746, DOI: 10.1016/j.jse.2020.02.022
RUSSO RAFFAELE ET AL: "A morphovolumetric study of head malposition in proximal humeral fractures based on 3-dimensional computed tomography scans: the control volume theory", JOURNAL OF SHOULDER AND ELBOW SURGERY, MOSBY, AMSTERDAM, NL, vol. 27, no. 5, 5 February 2018 (2018-02-05), pages 940 - 949, XP085382731, ISSN: 1058-2746, DOI: 10.1016/J.JSE.2017.12.004
Attorney, Agent or Firm:
CATALDI, Silvia et al. (IT)
Download PDF:
Claims:
CLAIMS

1. Method for calculating a severity index in fractures of a humerus of an individual, from a plurality of section images Sk of said humerus, where k=l,...,N with N which is a positive integer, captured by means of a diagnostic imaging technique, characterized in that it comprises the following steps:

A. normalizing the grey levels of each section image Sk of said plurality of images Sk, so that each section image Sk has the same grey level;

B. starting from said normalized images in said step A., segmenting each section image Sk , in a significative region, comprising at least one bone structure, wherein at each pixel p is assigned a value 1, and in a non- significative region, wherein at each pixel p is assigned value 0, so that each section image Sk is a binary image;

C. starting from said section images Sk segmented in said step B., identifying at least one connected component corresponding to said humerus, or a shoulder blade, in said at least one bone structure comprised in each segmented section image Sk, and labelling said at least one connected component with a respective label, so that said humerus is labelled with a first label and said shoulder blade is labelled with a second label which is different from said first label;

D. identifying a plurality of fragments of the head of said humerus, labelling each fragment of said plurality of fragments with a respective label and generating a real model (0) of said humerus from said plurality of fragments;

E. given a reference model (M) of the integral bone structure of said humerus, recomposing and aligning said plurality of fragments, identified in said step D., in said real model (0) according to said reference model (M); and

F. calculating a severity index of said fracture of said humerus starting from said plurality of fragments recomposed and aligned in said step E.

2. Method according to the preceding claim, characterized in that said step F comprises a sub step FI for:

- calculating a number of fragments of each bone region of said humerus, defining the number of different labels, 1k, where k=l,...,N with N which is a positive integer, associated with the voxels of the respective bone region;

- determining if the same label Ik is associated to the voxels of two distinct bone regions, or not;

- determining the volume of each bone region of said real model (0);

- determining the displacement of the lid of said humerus from a first position to a second position, different from said first position, wherein said first position is the position of said lid before said step E, while said second position is the position of said lid after said step E;

- calculating the number of fragments of said lid; and - identifying the fragment of said lid having the maximum volume between said fragments.

3. Method according to the preceding claim, characterized in that said step F comprises, after said sub step FI, a sub step F2 for calculating said severity index with respect to a volume of said humerus, wherein said volume of said humerus is the volume comprised between a plane α, arranged in correspondence of the anatomical neck of said humerus, and a plane b, substantially parallel to said plane α, and arranged in correspondence of the surgical neck of said humerus, and wherein said volume of said humerus is subdivided, by means of a plane λ, into a medial zone and a lateral zone which is opposite to said medial zone, said severity index being calculated by means of the following formula:

IS = ISM + ISL + ISC where ISM is the severity index of said medial zone, ISL is the severity index of said lateral zone and ISC is the severity index of said lid of said humerus.

4. Method according to the preceding claim, characterized in that said severity index of said medial zone ISM is calculated by means of the following formula:

ISM — PM + f (ΔCM) + f( BLm) where pM is a value associated with the respective fracture configuration of said medial zone with respect to said plane α and/or said plane b, f(ΔCM) is a value associated with the comminution of the respective fracture configuration of said medial zone and f(BLM) is a value associated with the bone loss of said medial zone.

5. Method according to anyone of the claims 3 or 4, characterized in that said severity index of said lateral zone ISL is calculated by means of the following formula:

LSM — PM + f (ΔCM) + f( BLm) where pL is a value associated with the respective fracture configuration of said lateral zone with respect to said plane α and/or said plane β, f(ΔCL) is a value associated with the comminution of the respective fracture configuration of said lateral zone and f(BLL) is a value associated with the bone loss of said lateral zone.

6. Method according to anyone of the claims 3-5, characterized in that said severity index of lid of said humerus ISC is calculated by means of the following formula:

ISC = Pc + f (n) + f(HF) , where pc is a value associated with each position of said lid, f(n) is a value associated with the number of fragments and bone loss of said lid and /(HF) is a value associated with the size of the largest fragment of said lid.

7. Method according to anyone of the preceding claims, characterized in that said step A comprises the following sub steps:

A1. given V the set of pixels p contained in the entire volume of said section images Sk , rescaling the values in V in the range [0, 4096] by means of a min-max transformation and calculating the average mV, the standard deviation sV and the value gmax, V of the grey tone of each pixel p, such that gmax, V > mV;

A2. given, for each section image Sk, the previous section image Sk-1 and the subsequent section image Sk+ 1 . calculating, for each pixel p in the section image Sk , a weight w as the geometrical average of the grey tone values that said pixel p assumes in said section images Sk-1 r Sk e Sk+1, rescaling said weights w in the range [0,1] and replacing the grey tone of pixel p with the value (1 + w) p;

A3. for each section image Sk normalized within said volume V, calculating the average mS, the standard deviation sS and the value gmax,S of the grey tone of each pixel p, and recalculating the grey tone of each pixel p in the section image Sk by means of the formula p' = (p — gmax,S)/sS . sV + gmax, V ;

A4. applying an enhancement process to each section image Sk in order to highlight the bone density of the bone regions comprised in the respective section image Sk; and

A5. rescaling, by means of a min-max transformation, the grey tones of the pixels p comprised in the section image Sk in the range [0, 4096].

8. Method according to the preceding claim, characterized in that said enhancement process comprises the following steps: i) calculating a smooth version SMk of Sk by applying a median filter; ii) calculating a coefficient cs = mS + sS, where mS is the average of the pixel values in SMk and sS is the standard deviation of the pixel values in SMk; iii) given a neighborhood 3D Ip centered in pixel p, calculating a factor w from the ratio between the number of pixels in Ip with a value greater than p and the number of pixels in Ip with a value less than p; and iv) recalculating the grey tone value of pixel p come p' = e(q/cs') . (1+ log2(2 — e — w)), where q is the pixel in SMk homologous to p in Sk .

9 . Method according to anyone of the preceding claims, characterized in that said step B comprises the following sub steps:

B1. given a section image Sk, rescaling the grey tones of the pixels p comprised in said section image Sk in a range [0, Gmax] ;

B2. given a threshold value th, calculating the foreground entropy Ef and the background entropy Eb, according to the following formulas: where H(gi) is the total number of times that the grey tone gi appears in the image section Sk, and P(gt) is the probability that the grey tone gi appears in the image section Sk .

B3. combining foreground and background entropies in an amount defining the difference between the adjacent values of , calculating the value and defining defining the difference between adjacent values of and determining a threshold thk for the segmentation of the image section Sk as the value

10. Method according to anyone of the preceding claims, characterized in that said step C comprises the following sub steps: C1. detecting the connected components of each image section Sk;

C2. extracting a plurality of morphological features for each connected component detected in said sub step Cl; and

C3. assigning each connected component to a respective class and labelling each connected component with a respective label, so that said humerus is labelled with a first label, said shoulder blade is labelled with a second label which is different from said first label, and a further bone region different from said humerus and said shoulder blade is labelled with a third label which is different from said first label and said second label.

11. Method according to the preceding claim, characterized in that said plurality of morphological features extracted in said substep C2 comprises the following morphological features:

- a first real value ekj equal to the ratio of the distance between the foci and the major axis of the ellipse that approximates said respective connected component;

- a second real value skj equal to the ratio between the area of the respective connected component and the area of the convex envelope containing said respective connected component;

- a third real value akj equal to one minus the ratio between the minor axis and the major axis of the ellipse approximating said respective connected component;

- a fourth real value Akj equal to the area of the respective connected component; and

- a fifth real value rkj equal to min(l- where e represents the mean square error with which an ellipse approximates the perimeter of the respective connected component.

12. Method according to the preceding claim, characterized in that said each connected component is labelled, in said substep C3, as said shoulder blade if as said humerus if or said further bone region, in the remaining cases.

13. Method according to anyone of the preceding claims, characterized in that said step D comprises the following sub steps:

D1 . calculating a bone density weight w for each pixel labelled as humerus, by means of said first label, in each section image Sk}

D2 . labelling said at least one connected component;

D3 . calculating a compatibility coefficient between the labels of said sub step D2; and

D4. applying a region growing algorithm to detect fragments .

14. Method according to anyone of the preceding claims, characterized in that said step E comprise the following sub steps: E1. extracting a set of features for each fragment;

E2. recomposing said plurality of fragments by means of said reference model (M);

E3. coupling fragments that are compatible with a fracture rhyme on the basis of a similar rigid transformation; and

E4. reassembling said plurality of fragments according to said sub steps E3 and E4.

15. System (S) for calculating a severity index in fractures of a humerus of an individual, from a plurality of section images Sk of said humerus, captured by means of a diagnostic imaging technique, comprising: a processing unit (SI) for performing steps A-F of the method according to any of claims 1-14, and displaying means (S2), connected to said processing unit (SI), for displaying at least said plurality of section images (I) of said humerus.

16. Computer program comprising instructions which, when the program is executed by a computer, cause the processor to perform the method steps A-F according to anyone of claims 1-14.

17. Computer-readable storage medium comprising instructions which, when executed by a computer, cause the processor to perform the method steps A-F according to anyone of claims 1-14.

Description:
METHOD FOR CALCULATING A SEVERITY INDEX IN TRAUMATIC FRACTURES AND CORRESPONDING SYSTEM IMPLEMENTING THE METHOD

The present invention relates to a method for calculating a severity index in traumatic fractures, in particular in the proximal third of the humerus of an individual.

The present invention concerns also a system for the implementation of this method. Field of the invention

More specifically, the invention relates to a method of the aforementioned type, studied and implemented in particular to define, by means of data from medical imaging instruments processed by a computer, a severity index of a proximal humerus fracture useful for classifying, in a precise manner, the fracture itself, according to an automatic multifactorial analysis of the parameters that characterize the traumatic event. In particular, the present invention allows supporting the diagnostic phase of traumatic fractures of the proximal third of the humerus of a person, having available in an automatic way all the information necessary to carry out an appropriate therapeutic choice for the benefit of the person involved in a trauma.

In the following, the description will be addressed to traumatic fractures of the proximal third of the humerus, but it is clear that it should not be considered limited to this specific use.

Prior art As is known, the treatment of traumatic fractures of the proximal third of the humerus is a method in continuous development and presents several criticalities.

The complexity resides in the low ability to properly figure out either by a radiographic point of view by means of a CT - Computer Tomography - the multiplicity of factors and parameters characterizing a specific traumatic event, such as, for example, the precise location of fragmented parts, their recognition or their specific interest.

Therefore, many proximal humerus treatments produce unsatisfactory results, with serious consequences on the life quality of the examined patients.

Currently, in the context of the treating of the humerus fractures, two classification techniques of the proximal humerus fractures are known, which play a key role in the signaling of clinical and epidemiological data, allowing a uniform comparison of similar conditions.

Among the most widespread classification techniques in the trauma field, there is the Neer classification, which dates back to the 70s of the last century.

In particular, in this classification, the 4 fragments of proximal humerus fractures are analyzed, indicating the limits of the surgery treatment in the cases in which the displacement of such fragments is less than 1 cm and 45° of rotation.

Another widely used classification is the AO - "Arbeitsgemeinschaft fur Osteosynthesefragen" classification -, which is based, also, on the association of the groups of fragments between them and it provides multiple subgroups.

However, a limitation of these known solutions consists in the fact that their analysis strongly depends on the evaluation subjectivity of the medical staff who performs it, since objective evaluation parameters are still missing. Further classifications of fractures are known, which, instead of supporting the importance of the number of fragments, push the observer to look at the etiopathogenesis of fractures, namely to try to associate the fragments with the displacements caused by the vector actions of the various muscles that are attached to the humeral head.

However, despite the efforts made over the past fifty years, the attempts to produce classification tools to support the diagnosis and the subsequent therapeutic indication were modest.

In fact, the possibility of improving the diagnostic quality through the CT and MR study - Magnetic Resonance - has been contrasted by scientific evidence that equally demonstrates poor reliability of the analysis made by medical personnel, even with 2D and 3D CT studies.

The literature also reports another type of classification obtained through the CT evaluation of the humeral calcar and in particular for 4-part fractures, in which it is demonstrated how the evaluation of the type of this study can help the orthopedist to improve the anatomopathological understanding of the displacements of the fragments in relation to the forces that determined them.

Scope of the invention

In light of the above, it is, therefore, an object of the present invention to overcome the aforementioned disadvantages by providing a method for calculating a severity index in traumatic fractures of the proximal third of the humerus of a specific individual, and a relative implementation system of the method.

Another object of the invention is to provide a method that allows analyzing automatically the severity of traumatic fractures of the proximal third of the humerus, providing all the information necessary to carry out a therapeutic choice as objectively as possible.

A further object of the present invention is to provide the tools necessary for carrying out the method and the apparatuses that carry out this method.

Object of the invention

It is therefore specific object of the present invention a method for calculating a severity index in fractures of a humerus of an individual, from a plurality of section images S k of said humerus, where k=l,...,N with N which is a positive integer, captured by means of a diagnostic imaging technique, characterized in that it comprises the following steps: A. normalizing the grey levels of each section image S k of said plurality of images S k , so that each section image S k has the same grey level; B. starting from said normalized images in said step A., segmenting each section image S k , in a significative region, comprising at least one bone structure, wherein at each pixel p is assigned a value 1, and in a non-significative region, wherein at each pixel p is assigned value 0, so that each section image S k is a binary image; C. starting from said section images S k segmented in said step B., identifying at least one connected component corresponding to said humerus, or a shoulder blade, in said at least one bone structure comprised in each segmented section image S k , and labelling said at least one connected component with a respective label, so that said humerus is labelled with a first label and said shoulder blade is labelled with a second label which is different from said first label; D. identifying a plurality of fragments of the head of said humerus, labelling each fragment of said plurality of fragments with a respective label and generating a real model of said humerus from said plurality of fragments; E. given a reference model of the integral bone structure of said humerus, recomposing and aligning said plurality of fragments, identified in said step D., in said real model according to said reference model; and F. calculating a severity index of said fracture of said humerus starting from said plurality of fragments recomposed and aligned in said step E.

Further according to the invention, the step F of the method may comprise a sub step FI for: calculating a number of fragments of each bone region of said humerus, defining the number of different labels, Ik, where k=l,...,N with N which is a positive integer, associated with the voxels of the respective bone region; determining i f the same label I k is associated to the voxels of two distinct bone regions, or not; determining the volume of each bone region of said real model; determining the displacement of the lid of said humerus from a first position to a second position, different from said first position, wherein said first position is the position of said lid before said step E, while said second position is the position of said lid after said step E; calculating the number of fragments of said lid; and identifying the fragment of said lid having the maximum volume between said fragments.

Still according to the invention, the step F of the method may comprise, after said sub step FI, a sub step F2 for calculating said severity index with respect to a volume of said humerus, wherein said volume of said humerus is the volume comprised between a plane α, arranged in correspondence of the anatomical neck of said humerus, and a plane β, substantially parallel to said plane α, and arranged in correspondence of the surgical neck of said humerus, and wherein said volume of said humerus is subdivided, by means of a plane λ, into a medial zone and a lateral zone which is opposite to said medial zone, said severity index being calculated by means of the following formula:

IS = IS M + IS L + IS C where IS M is the severity index of said medial zone, IS L is the severity index of said lateral zone and IS C is the severity index of said lid of said humerus. Advantageously according to the invention, the severity index of said medial zone IS M is calculated by means of the following formula: where p M is a value associated with the respective fracture configuration of said medial zone with respect to said plane α and/or said plane β, f(ΔC M ) is a value associated with the comminution of the respective fracture configuration of said medial zone and f(B LM ) is a value associated with the bone loss of said medial zone.

Conveniently according to the invention, the severity index of said lateral zone IS L is calculated by means of the following formula:

IS L = p L + f(ΔC L ) + f(B LL ) where p L is a value associated with the respective fracture configuration of said lateral zone with respect to said plane α and/or said plane b, f(ΔC L ) is a value associated with the comminution of the respective fracture configuration of said lateral zone and f(B LL ) is a value associated with the bone loss of said lateral zone.

Further according to the invention, the severity index of lid of said humerus IS C is calculated by means of the following formula: IS C = Pc + f (n)+ f(H F ) where p c is a value associated with each position of said lid, f (n) is a value associated with the number of fragments and bone loss of said lid and is a value associated with the size of the largest fragment of said lid.

Always according to the invention, said step A comprises the following sub steps: A1. given V the set of pixels p contained in the entire volume of said section images S k , rescaling the values in V in the range [0, 4096] by means of a min-max transformation and calculating the average mV, the standard deviation sV and the value gmax,V of the grey tone of each pixel p, such that gmax, V > mV; A2.given, for each section image S k , the previous section image S k-1 and the subsequent section image S k+lr calculating, for each pixel p in the section image S k , a weight w as the geometrical average of the grey tone values that said pixel p assumes in said section images S k-lr S k e S k+lr rescaling said weights w in the range [0,1] and replacing the grey tone of pixel p with the value (l+ w)-p; A3, for each section image S k normalized within said volume V, calculating the average mS, the standard deviation sS and the value gmax, S of the grey tone of each pixel p, and recalculating the grey tone of each pixel p in the section image S k by means of the formula p' = (p — gmax, S)/sS . sV + gmax, V; A4. Applying an enhancement process to each section image S k in order to highlight the bone density of the bone regions comprised in the respective section image S k ; and A5. rescaling, by means of a min-max transformation, the grey tones of the pixels p comprised in the section image S k in the range [0, 4096].

Still, according to the invention, the enhancement process may comprise the following steps: i) calculating a smooth version S Mk of S k by applying a median filter; ii) calculating a coefficient cs = mS + sS, where mS is the average of the pixel values in S Mk and sS is the standard deviation of the pixel values in S Mk ; iii) given a neighborhood 3D I p centered in pixel p, calculating a factor w from the ratio between the number of pixels in I p with a value greater than p and the number of pixels in I p with a value less than p; and iv) recalculating the grey tone value of pixel p come p' = e(qZcs ' ) (1 + log 2 (2 — e — w)), where q is the pixel in S Mk homologous to p in S k .

Advantageously according to the invention, the step B may comprise the following sub steps: B1. given a section image S k , rescaling the grey tones of the pixels p comprised in said section image S k in a range [0, Gmax] ; B2. given a threshold value t h , calculating the foreground entropy E f and the background entropy E b , according to the following formulas: where H(gi) is the total number of times that the grey tone g t appears in the image section S k , and P(g t) is the probability that the grey tone gi appears in the image section S k; B3. combining foreground and background entropies in an amount defining the difference between the adjacent values of E tot as calculating the value m E = and defining defining the th difference between adjacent values of and determining a threshold th k for the segmentation of the image section S k as the value further according to the invention, the step C of the method may comprise the following sub steps: Cl. detecting the connected components of each image section S k ; C2. extracting a plurality of morphological features for each connected component detected in said sub step Cl; and C3. assigning each connected component to a respective class and labelling each connected component with a respective label, so that said humerus is labelled with a first label, said shoulder blade is labelled with a second label which is different from said first label, and a further bone region different from said humerus and said shoulder blade is labelled with a third label which is different from said first label and said second label. Conveniently according to the invention, the plurality of morphological features extracted in said substep C2 comprises the following morphological features: a first real value e kj equal to the ratio of the distance between the foci and the major axis of the ellipse that approximates said respective connected component; a second real value s kj equal to the ratio between the area of the respective connected component and the area of the convex envelope containing said respective connected component; a third real value a kj equal to one minus the ratio between the minor axis and the major axis of the ellipse approximating said respective connected component; a fourth real value A kj equal to the area of the respective connected component; and a fifth real value r kj equal to min(l- where 8 represents the mean square error with which an ellipse approximates the perimeter of the respective connected component.

Always according to the invention, each connected component is labelled, in said substep C3, as said shoulder blade if , as said humerus if or said further bone region, in the remaining cases.

Advantageously according to the invention, the step D of the method may comprise the following sub steps: D1. calculating a bone density weight w for each pixel labelled as humerus, by means of said first label, in each section image S k; D2. labelling said at least one connected component; D3. calculating a compatibility coefficient between the labels of said sub step D2; and D4. applying a region-growing algorithm to detect fragments.

Further according to the invention, the step E of the method may comprise the following sub steps: E1. extracting a set of features for each fragment; E2. recomposing said plurality of fragments by means of said reference model; E3. coupling fragments that are compatible with a fracture rhyme on the basis of a similar rigid transformation; and E4. reassembling said plurality of fragments according to said sub steps E3 and E4.

It is also object of the present invention a system for calculating a severity index in fractures of a humerus of an individual, from a plurality of section images S k of said humerus, captured by means of a diagnostic imaging technique, comprising: a processing unit for performing steps A-F of the method, and displaying means, connected to said processing unit, for displaying at least said plurality of section images (I) of said humerus.

It is further object of the present invention a computer program comprising instructions which, when the program is executed by a computer, cause the processor to perform the method steps A-F.

It is also object of the present invention a computer-readable storage medium comprising instructions which, when executed by a computer, cause the processor to perform the method steps A-F.

Brief description of the drawings The present invention will be now described, for illustrative but not limitative purposes, according to its preferred embodiments, with particular reference to the figures of the enclosed drawings, wherein: figure 1 shows a flowchart of an embodiment of a method for calculating a severity index in traumatic fractures of the proximal third of the humerus of one individual, according to the present invention; figure 2a shows a side view of a control volume of the humerus; figure 2b shows a side view of the control volume of figure 2a defined by means of a α and a β plane; figure 3a shows, in a perspective view, a subdivision, by means of a δ plane, of the control volume of figure 2, in a anterior and a posterior zone; figure 3b shows, in a perspective view, a subdivision, by means of a plane λ, of the control volume of figure 2, in a lateral region and a medial region; figure 4 shows, in a front perspective view, a subdivision, by means of said plane d and said plane λ, of the control volume of figure 2a, in a anterior lateral zone and a posterior lateral zone and in an anterior medial zone and a posterior medial zone; figure 5 shows a schematic view of images obtained before and after a pre-processing step of the method according to the present invention; figure 6 shows a schematic view of a previous image I and a subsequent image II after the application of a segmentation step, and the relative graphs, of the method according to the present invention; figure 7 shows, in a perspective view, a graphic representation of the fracture conformations of the medial zone of the humerus; figure 8 shows, in a perspective view, a graphical representation of the fracture conformations of the lateral region of the humerus; figure 9 shows, in perspective view, a graphical representation of the conformations of the position of a humerus lid; and figure 10 shows a schematic view of a system for implementing the method for calculating a severity index in traumatic fractures of the proximal third of the humerus of an individual, according to the present invention.

In the various figures, similar parts will be indicated by the same reference numbers.

Detailed description

With reference to figure 1, the method for calculating a severity index in traumatic fractures of the proximal third of the humerus of an individual, according to the present invention, comprises the following steps:

A. pre-processing of a volume of images produced by a CT - Computed Tomography exam;

B. segmentation of the pre-processed image volume;

C. humerus and scapula identification;

D. humeral head fragments identification;

E. recomposition and alignment of the humeral head fragments with respect to a reference model M; and

F. estimate of a severity index. In particular, in the embodiment disclosed, each of the steps A. - F. mentioned above is performed fully automatically, for example, with the aid of a computer aided system - Computer-Aided System or CAS.

However, in further embodiments of the present invention, each of the steps A. - F. may provide the possibility of a correction manual intervention by an operator, through a suitable interface.

A description of the various steps is now provided.

Step A.: pre-processing of a volume of images produced by a CT exam

Currently, CT imaging equipment exhibits high variability in terms of image quality.

In particular, such inhomogeneity depends on the different generations of computed tomographs in use, as well as on the different configurations of acquisition parameters typically adopted by the single radio - diagnostician .

Therefore, this requires the need to standardize the features of the images, by means of said step of pre-processing of a plurality or a volume of images previously acquired by the CT technique.

In particular, the input data of said step A. comprise a volume of CT images, namely images of sections - or slices - of the object under examination obtained by Computed Tomography applied to an individual.

In the present embodiment, such images are CT images. However, in further embodiments of the present invention, it is possible to use images obtained by various diagnostic imaging techniques. This step A. allows correcting the levels, or gray tones of each slice of the volume of the input CT images, in order to reduce the variability of the levels or gray shades both between one slice and another and within of the same slice.

The output data of step A. include, however, a normalized volume of images CT, namely images that disregards not as much as possible by the acquisition conditions, i.e., that have features as objective as possible.

Step A comprises the following sub-steps:

A1. indicating with V the set of pixels contained in the entire volume of images, the values in V are in the range [0, 2 12 ], in accordance with the depth of pixels (12 bits) provided by the CT devices acquisition. The values in V are rescaled in the range [0, 4096] by means of a known min-max transformation and on V are calculated the following parameters: the average mV, the standard deviation sV and the value of the gray tone gmax,V, such that gmax,V > mV, with the maximum number of occurrences in V, where occurrence means the number of times a value is present in V ;

A2. it is considered, for each slice S k , where k = 1,..., N, where N is a positive integer, the previous slice S k-1 and the next slice S k+1 , in the sequence of volume images; for each pixel p in the slice S k , a weight w equivalent to the geometric mean of the values of gray tones that it assumes in the slices S k-1 , S k and S k+1 is calculated; the weights are then rescaled to the interval [0, 1] and the gray tone of the pixel p is replaced with the value (1 + w) - p;

A3, for each slice S k normalized within the volume V, the same parameters are calculated: average mS, standard deviation sS and the value of gray tone gmax,S, limited to the set of pixels in the slice S k ; the gray tone of each pixel p in the slice S k is recalculated according to the formula: p' = (p — gmax,S)/sS . sV + gmax,V ;

A4. an enhancement process (or improvement process) is applied to each slice S k aimed at highlighting the bone density of the structures present therein; such enhancement process comprises the following steps: i) computation of a smooth version S Mk of S k by applying a median filter, in which this smooth version has a kernel of size 4x4; ii) calculation of a coefficient S Mk of S k , where mS and sS are respectively the average and the standard deviation of the pixel values in S Mk ; iii) for each pixel p in S k a 3D neighborhood I p is considered, centered in p and of size (12, 12, 5), within which the factor w given by the ratio between the number of pixels in I p with a value greater than p, and the number of pixels in I p with a value less than p; iv) the value of the gray tone of the pixel p is recalculated as p'= e(q/cs) · (1+ log 2 (2 — e — w)), where q is the pixel in S Mk the homologue of p in S k ;

A5. the gray tones in the slice S k are rescaled to the interval [0, 4096], by means of a min-max transformation .

Referring to figure 5, the difference between some slices, present in the row I is noted, which show portions of the humeral head of an individual before the pre-processing step A., and the respective slices, present in the row II, obtained after the pre-processing step A.

Step B : segmentation of the volume of the pre- processed images .

The Hounsfield (HU) scale, also called the CT number, is a unit scale adopted to quantitatively describe radiodensity.

Currently, the known solutions for the segmentation and extraction of bone structures from a volume of CT images, allow mapping the gray tones of the pixels in a slice in the corresponding HU index, by means of a linear transformation, and to select as pixels of interest those whose value is contained in the interval [300, 900], generally referred to as the range of variation of bone density.

However, due to the great variability between the volumes of images produced by different tomographs and, sometimes, between slices of the same volume, a single threshold operation on the entire volume produces highly noisy and poor-quality results.

Therefore, a manual intervention by the operator is necessary, in order to refine the segmentation obtained previously in an automatic way.

The method according to the present invention, on the other hand, provides for a local segmentation step of the volume of images leaving the pre-processing step A., in which the optimal threshold for the segmentation of the bone structures in each image is automatically determined.

The input data of segmentation step B. comprise the volume of normalized CT images, i.e., the output data of step A.

The segmentation step B. allows partitioning each slice in significant regions, resulting in an automatic way the optimal threshold for segmenting or dividing, the pixels corresponding to the bone structures from the additional pixels of the image.

More specifically, given a slice S k , by segmentation of bone structures contained therein, it is meant a calculation of a threshold, such that all pixels p with gray tone greater than threshold th k are set to 1, and considered pixels of foreground, namely bone, while the pixels with a gray tone lower than the threshold th k are set to 0, and therefore considered as background pixels, that is, of irrelevant content.

Therefore, the segmentation step produces a binary mask, where only the pixels corresponding to the bone structures in S k are set to 1.

As mentioned, in fact, the concept underpinning the segmentation process is that of automatically find that value th k for the slice S k , which divides the pixels into two sets, i.e., foreground and background, for which it results that the corresponding internal entropy is maximum.

The output data of the step B. comprise a volume of segmented or binary CT images, i.e., images having only two possible values for each pixel, 0 for non-relevant content or background, and 1 for the relevant or foreground content.

Step B includes the following sub-steps:

B1. Given a slice S k , the gray tones are rescaled in an interval [0, Gmax] , in order to improve the efficiency and effectiveness of the algorithm. In the specific case, Gmax is set to 512;

B2. H(g i ) is defined as the total number of times the gray g i tone appears in the slice S k .

It is also defined P(g i ) as the probability that the gray tone g^ appears in the slice S k , that is as H(g i ) divided by the total number of pixels within the slice s k .

Given a threshold value th, the entropy of the foreground and the entropy of the background are calculated, according to the following formulas:

B3. the entropy of the foreground and the background are combined in a single quantity the difference between the adjacent values E tot is defined as the value is calculated and is defined. Again, the difference between adjacent values of is defined; and the threshold th k for segmentation of the slice S k is determined as the value

Referring to figure 6, it is seen a slice I relative to the humeral head of an individual and a slice II after the segmentation step described above, and a graph of the corresponding quantities Ey, E b and E tot , and the variation of th in [0, Gmax] .

Step C : identification of humerus and scapula

The bone structure extracted from a volume of CT images can include several elements including, for example, humerus, clavicle, scapula, ribs, thoracic vertebrae, and or sternum.

However, only the scapula and, in particular the humerus, are of real interest in planning a reconstruction surgery.

The present method comprises a step for the detection of the scapula and humerus, so as to be able to exclude by subsequent steps all the elements of the previously segmented bone structure, which are not of interest.

The input data of the step C comprise the volume of segmented CT images, namely the output data of the step

B.

As mentioned, the scapula and humerus identification step allows identifying these bone elements within each image of the segmented CT image volume. The output data of the step C comprise the volume of the CT images, wherein in each image the humerus and the scapula have been identified.

Step C comprises the following sub-steps:

C1. detecting the connected components in the single slices and label the connected components along with an axial scan of the volume of segmented CT images;

C2. extracting the morphological features of the connected components; and

C3. assigning each connected component to one of the classes of the scapula, humerus, or other.

In sub-step Cl., the CT volume is processed slice by slice in the axial direction, from top to bottom. In particular, an algorithm for labeling the 8-connected components is applied to each slice S k .

This known algorithm produces as a result a matrix of the same size as S k , in which all the pixels belonging to the same connected component c kj are marked by the same label value l kj .

Furthermore, the system maintains a different set of labels L, which is global with respect to the CT volume and which initially contains all and only the labels present in the slice S 1 .

The algorithm proceeds slice by slice, trying to assign the labels already present in L to the objects under examination in the current slice, on the basis of a compatibility criterion.

In particular, if there are connected components in the current slice, for which a label cannot be found in L, a new label is generated for each of them and it is added to the global set of labels L.

The procedure for assigning labels on the current slice S k is as follows.

It is considered the generic connected component c kj and it is indicated with the set of labels, or labels, L that in the slice Λ k-1 have been assigned at the points c kj of the previous step.

If Λ k-1 isempty, a new label is generated, inserted in L, and assigned to the component c kj .

If Λ k-1 contains only one label, such label is assigned to c kj .

If Λ k-1 contains multiple labels, the one with the most overlap with c kj is assigned to c kj .

At the end of this process, L represents the set of labels assigned to the different objects in the bone structure extracted during the segmentation step.

As anticipated, this type of labeling has the sole purpose of identifying the scapula and the humerus. In fact, such labeling cannot be considered as the labeling of the individual fragments produced by fracture of the humeral head, because if several fragments adjacent to one another have points of contact, such fragments are necessarily labeled with the same label.

In sub-step C2., for each connected component c kj in each slice S k , the following features are extracted: a. Eccentricity e kj , that is a real value equal to the ratio of the distance between the focuses and the major axis of the ellipse approximating the connected component; b. Solidity s kj , that is a real value calculated as the ratio between the area of the connected component and the area of the convex hull, that contains it; c. Anomaly a kj , that is a real value equal to one minus the ratio between the minor axis and the major axis of the ellipse approximating the connected component; d. Area A kj , that is a real value equal to the area of the connected component; and e. Regularity r kj , i.e. a real value equal to min(l- where 8 represents the error root mean square with which an ellipse approximates the perimeter of the connected component.

In sub-step C3., the connected components of all slices in the CT volume are examined individually.

In particular, on the basis of the aforementioned features, each connected component is assigned to the class: a. Scapula, if b. humerus, if r kj >0.4; and c. other, in the remaining cases.

Whenever a connected component c kj is assigned to the scapula class, the probability of its corresponding label in L of being selected as the scapula object is increased.

Similarly, when a connected component is assigned to the humerus class, the probability of its corresponding label in L being selected as a humerus object is increased. Finally, objects, whose labels L have the highest probability for the respective classes, are selected as the scapula and the humerus.

The displaced fracture of the humerus head produces fragments, which, due to their morphological features, may not have been assigned to the humerus class.

The method according to the present invention, being known the typical arrangement of such fragments near the surgical neck, identifies an approximation of the latter as a reference point and assigns to the humerus class all those fragments, not classified as scapula, whose distance from it is less than a threshold dth.

The algorithm for the search of the approximation of the surgical neck operates only on the connected components c kj having as a label in L the one assigned to the humerus class in the previous step.

In particular, the algorithm implements the following steps: a) each slice S k of the CT scan volume is examined along the axial direction, and, if it contains a connected component c kj labeled as the humerus, it is placed H(k)=A kj, where H is a histogram with a number of bins equal to the number of the components labeled as humerus and A kj represents the area of the connected component. The values in H are, then, rescaled in the interval [0,1] by means of a min-max transformation; b) indicated with href the value with the highest number of occurrences in H, a new histogram H' is constructed, such that H(k)'represents the number of c kj in the previous slices (S1,..., Sk-1), for which it is true that c) the slice S kcg which approximates the surgical neck is identified as

Finally, indicated with c kcgj the component labeled as humerus in the slice S kCg , the threshold dth is calculated as the diameter of that component.

All the fragments, in the CT volume not labeled as scapula and having at least one voxel, i.e., a volumetric pixel, at a distance from c kcgj less than dth, are labeled as the humerus.

Step D : identification of the fragments of the humerus head .

The fracture of the humeral head produces fragments which number, size, and dislocation constitute an essential information for the calculation of the severity index.

The prior art solutions provide an extremely small number of methods for the detection of bone fragments from the volumes of CT images.

The simple application of a search algorithm for the connected components in a 3D space, in fact, is not sufficient, since in many cases the fragments generated by a fracture are not perfectly disjoint, but have one or more points of contact.

The step D of the method proposed solves this problem by means of labeling of the connecting components and the first slice S k and propagation of such labels along the axial direction of the CT volume or creation of new labels on the basis of a superposition criterion of the same connected components.

A compatibility coefficient is then associated with the generated labels, while a weight proportional to bone density is associated with the individual pixels.

Finally, a region growing algorithm aggregates the different pixels of the volume into different sets, corresponding to the fragments, on the basis of the previously calculated information.

The input data of said step D. comprise the volume of the CT images, wherein in each image humerus and the scapula have been identified, namely the output data of the step C.

As mentioned, step D allows identifying the fragments of the humeral head generated by a fracture, even if such fragments are not perfectly disjoint but have one or more point contact between them.

Furthermore, in step D, a real model 0 of the fully reconstructed humerus is generated (since it is fragmented into several parts).

The output data of said step D. comprise the 3D fragments of the humerus head identified in the previously labeled CT volume.

The step D comprises the following sub-steps:

D1. calculation of a bone density weight w for each pixel labeled as humerus in each slice S k;

D2. labeling of the connected components;

D3. calculation of a compatibility coefficient between labels; and

D4. weighted region growing algorithm application for fragment labeling. In step D1, a weight proportional to its bone density is assigned to each pixel.

Considering a slice S k , the complement is calculated, where gmax is the maximum gray tone value in S k .

For each pixel p in S k , a coefficient C pk proportional to its gradient along the dominant direction is calculated.

Similarly, a coefficient is calculated for the same pixel in . The weight wp associated with the pixel is therefore calculated as

In sub-step D2, the CT volume is processed slice by slice in the axial direction from the top to the bottom.

An algorithm for labeling the 8-connected components is applied to each slice S k .

This known algorithm produces as a result of a matrix of the same size as S k , in which all the pixels belonging to the same connected component c kj are marked by the same label value l kj .

Furthermore, the system maintains a different set L of labels, which is global with respect to the CT volume and which initially contains all and only the labels present in the slice S- L .

The algorithm proceeds slice by slice trying to assign the labels already present in L to the objects under examination in the current slice, on the basis of a compatibility criterion.

If there are connected components in the current slice, for which a label cannot be found in L, a new label is generated for each of them and it is added to the global collection L.

The procedure for assigning labels on the current slice S k is as follows.

Consider the generic connected component c kj and indicate with Λ k-1 the set of labels of L that have been assigned in the slice S k-1 at the points c kj of the previous step.

If includes zero or a label, a new label is generated, inserted into L, and assigned to the component c k,j·

If Λ k-1 contains two or more labels, for each label lj in Λ k-1 the intersection region between R kj and c kj and the set of pixels in S k-1 with label lj is determined.

Furthermore, the Dice index d between R kj and the corresponding component connected in S k-1 is calculated, in order to determine a threshold th δ = 0.25 (1— δ).

If the percentage of pixels in R kj with label lj is greater than th δ , a new label is generated, inserted in L and assigned to the component to the pixels present in

R k,j·

Otherwise, these pixels take the label of the closest already labeled pixel. The result is the set of labels L and a CT volume VL, in which all points are labeled with a label of L.

In the third sub-step D3, a compatibility indexbetween the different labels within the CT volume identified in substep D2 above is calculated to.

This index, together with the weight calculated in the sub-step D1, is a fundamental part of the region growing algorithm for identifying the fragments.

The compatibility index indicates the probability that pixels labeled with two different labels can be merged within the same fragment.

There are denoted by C and D, two square matrices of dimensions |L|x|L|, where |L| represents the cardinality of the set of labels identified in sub-step D2.

The algorithm proceeds in three steps: a) the CT volume is processed slice by slice in the axial direction from top to bottom.

For each pair of labels i and j in S k , C(i,j) is set with the value where A k i , A k,j , and d respectively represent the area of the connected components labeled with iandjand d is the distance between the two components.

Similarly, D(i,j) is set with the value 1/d. Also, for each pair of labels i in S k and j in S k+1 , it is set with the value 1/d. b) the CT volume is processed slice by slice in the sagittal direction.

For each slice S k , it individuates all the connected components it contains.

For each connected component c kj in S k , be A kj the set of labels contained in c kj .

For each pair (i,j) in A kj the position is increased by an amount equal to c) the CT volume is processed slice by slice in a coronal direction.

For each slice S k , it individuates all the connected components it contains.

For each connected component c k j in S k , be A k j the set of labels contained in c kj . For each pair (i,j) in A kj , the position C(i,j) is increased by an amount equal to

In the fourth step D4, it is applied a region growing algorithm to locate 3D fragments in the TC labeled volume.

The result is a new set of labels L F , each of which corresponds to a different fragment in a new labeled volume V LF .

The points of the volume already labeled are indicated with V LF r and the points of the volume not yet labeled are indicated with V NLF . Initially, V LF = 0 and V NLF =V L .

The algorithm operates by iterating two steps: a) selection of the starting seed; and b) expansion of the label.

Finally, it proceeds to label all the points of the volume not labeled with the previous two steps.

Initially, in step a, at each iteration k the algorithm chooses a seed from which starting, that is a label l k L, not yet selected in the previous iterations, from which starting the expansion process.

The seed l k is selected as that label in L, for which the sum of the compatibility indexes with all the other labels is maximum. That is, like that label that has the greatest probability of expansion. Once selected l k :

- l k is eliminated from L;

- all points in V L labeled with l k are inserted into a set P lk ;

- a new label l j in L F is generated corresponding to a new fragment;

- a threshold value th j is calculated equal to the average of the compatibility indexes in C of l k with all the other labels.

Subsequently, in step b, the label expansion algorithm exploits the information generated in steps D1 and D4,a, i.e., the weights wp, the threshold th j and the set of points P lk .

In particular, for each point p in it is considered its eight neighbors q in V NLF one at a time.

If the quantity is greater than th j , q is added to P lk .

After the visit of his eight-neighbors list, p is eliminated from P ik . The algorithm ends when the set P lk is empty. If L is not empty, go back to step D 4,a.

Step E : recomposition and alignment of the fragments of the humeral head with respect to the reference model M.

The known solutions do not solve the problem of the recomposition of 3D volumes starting from a set of its parts, as they are based solely on the correspondence between the fracture lines of the different fragments.

These known solutions, then, have the further disadvantage of requiring an alignment process with respect to a reference volume, when it is desired to project structural elements of interest on the volume under examination, such as those necessary for the calculation of the control volume and the severity index.

As can be seen in particular from figures 2a and 2b, the control volume of the humerus is the volume between the lid and the diaphysis and, more precisely, between the α and β planes. In particular, this control volume comprises the tuberosity and the calcar.

Step E solves the problem in the recomposition of a set of fragments and their alignment with respect to the reference model M, in a concomitant way.

The step E input data comprises the 3D humeral head fragments found in the CT volume, i.e., the step D output data.

In particular, as will be better described in the following, in step E, two types of regions are considered for the single fragment, namely the intact zones, and the fracture zones. As a result, starting from these regions a set of characteristic points is identified, used, then, for a dual matching process, both at the local level, between pairs of fragments, and globally, between more fragments groups, with the objective of maximizing the consistency of the re-composition with respect to the reference model M.

Therefore, step E is employed to align the real model 0 with the reference model M.

The output data of step E comprise the different fragments of the humeral head recomposed according to the reference model M and aligned with it with respect to the scale, the position, and the orientation in the space.

Step E comprises four sub-steps:

E1. extraction of a set of features for each fragment;

E2. preliminary recomposition guided by the reference model M;

E3. matching of fractures between pairs of fragments; and

E4. matching between groups of fragments based on intact zones and fracture zones.

In sub-step E1, given a fragment F i , a set of reference points, considered as characteristic points, is identified within this fragment.

Similarly, a set of characteristic points is identified in the reference model M.

In the specific case, the algorithm applied for the determination of the characteristic points is based on the ISS - Intrinsic Shape Signatures descriptor.

The known ISS algorithm considers for each 3D point its support region, of which it calculates the covariance matrix and marks it as a reference point if the difference in magnitude between the first two most significant eigenvalues is maximum. For each of the reference points P j , a features vector V Pi is then calculated, which describes its local geometric properties.

In sub-step E2, each fragment is placed in correspondence with the region most similar to it in the reference model M, through a process of matching local between the features of the fragment and those of the reference model M.

Specifically, for each pair of points P j G F i and M a correspondence is established if the distance between the corresponding feature vectors V Pi and V Qi is less than a predetermined threshold.

The set of corresponding points thus identified is then reduced. In particular, for each pair of points identified on the fragment F i , the Euclidean distance is measured, and if the corresponding pair of points identified on the reference model M is at a distance greater than a predetermined threshold, the pair is eliminated from the set of potential matches.

In sub-step E3, the algorithm for the matching of fractures between a generic pair of fragments f i and f j has the purpose of coupling fragments that are compatible with respect to the fracture gap according to a rigid affine transformation T ij .

For this purpose, the algorithm characterizes the rime of fracture, namely the exact interruption point of the bone, of each fragment f on the basis of sets of characteristic curves C i . Given two fragments andf and f j, with their respective sets C i and C j , the algorithm solves a Largest Common Point-Set (LCP) problem, in order to find the transformation F ij , which minimizes the distance between f i and f j . If the distance between the fragments, with respect to the relative fracture lines, is less than a predetermined threshold, they are considered compatible with the transformation found.

In sub-step E4, the method carries out the global reassembly of the fragments, taking into account both the matching with the reference model M, and the compatibility between the different pairs of fragments.

In particular, the algorithm represents the fragments as a graph, in which a node is represented by a pair (f i , T i ) r where fi is the i-th fragment, while T i is a transformation matrix that maps the corresponding fragment in its final position.

An arc characterized by a transformation T ij is inserted between two nodes of the graph if the two fragments are compatible, according to the criterion described in the previous step. A global compatibility index between two fragments is defined as the sum of two different components, that is, one relating to the compatibility of the fracture lines and one relating to the correspondence with the reference model

Thus, there is where l is a weight factor that depends on the quality of the reference model M.

Finally, the algorithm solves a multi-parametric search problem of the subgraph of the graph previously constructed, and which maximizes the sum of the global compatibility indices between the nodes.

Phase F : estimate of the severity index (IS) .

The recomposition process reassembles the different fragments extracted from the CT volume by means of steps A-D, in the 3D model of the humerus, or in the real model

0.

Furthermore, the operation and realignment with respect to the reference model M, performed in conjunction with the recomposition process, matches each part of the real model 0 with its counterpart in the reference model M.

The term part refers to a bony region of the humerus. This part is whole, namely, it is free of fragments, in the absence of a fracture, or it can include one or more fragments in the presence of a fracture.

Given the direct correspondence between the real model 0 and the reference model M, the annotations made on the reference model M are directly reproducible on the model real 0.

In particular, considering the reference model M annotated with the following parameters:

- planes α,β,δ, λ;

- the greater tuberosity GT;

- the lesser tuberosity PT;

- the anterior calcar CA;

- the posterior calcar CP;

- the diaphysis; and

- the lid; it is possible to identify the same parameters on the real model 0.

This allows all measurements defined on the reference model M to be carried out directly on the real model 0 for the evaluation of the control volume.

Furthermore, since the different fragments of the real model 0 are labeled independently, the method is capable of automatically calculating the weights associated with the codes shown in the tables adopted for the classification.

The input data of step F includes the different fragments of the humeral head recomposed and aligned according to the reference model M, i.e., the output data of step E.

Step F allows calculating the severity index of a fracture of the proximal third of the humerus of a specific subject.

The data output of the step F includes one or more values of the severity index.

Step F includes two sub-steps FI and F2, which respectively have the objective of extracting all the measures necessary for estimating the severity index and calculating the latter on the basis of the formulas and the weight tables illustrated below in the description.

Downstream the alignment process, the annotations relating to the margins of the individual parts can be directly mapped from the reference model M (previously annotated) to the real model 0. Since each voxel in the real model 0 is associated with the label of one of the fragments identified in step D, in sub-step F1, the system is able to automatically calculate the following information:

- the number of fragments of each part - the system counts the number of different labels l k where k = 1,..., N with N being a positive integer, associated with the voxels of each part.

- the links with the other parts - if the same label l k is associated with the voxels of two different parts, a link is detected between these parts, as the fragment identification algorithm has not identified any fracture line between them.

- the volume of each part on the real model 0 - the count of the number of voxels of the real model 0 associated with the single part and size in mm of the single voxel allow calculating this information directly.

- the displacement of the lid (neutral, varus, valgus, posterior and anterior dislocation) - the inverse spatial transformation to that applied for the relocation of the lid and the original position of the latter with respect to that occupied after the realignment, and to the glenoid cavity, or to the joint cavity of the bone, allow the classification of the displacement.

- number of lid fragments - this information is obtained by counting the number of labelsl k of different fragments associated with the lid.

- the lid fragment with maximum volume - the labell k associated with the largest number of lid voxels represents the fragment with the maximum volume. Sub-step F2 evaluates the severity index. As will be better described in the following, it is calculated by assigning weights to different factors and parameters that characterize the fracture, such as the fractured segments in the medial or lateral zone of the control volume, or the position of the humeral head and its fragmentation (head split), or the comminution and the bone loss of the parts.

The segments considered for the formulation of the severity index are those enclosed by the control volume, that is the calcar in the medial zone, the greater tuberosity GT, and the lesser tuberosity PT in the lateral zone and the lid.

For each of them fractures the forms are identified. Each factor was assigned a numerical value (a weight) assessed on the basis of the severity.

The concept of severity can be linked to three different aspects: complexity of the fracture in anatomical-pathological terms, or the link between the parts (anatomical-pathological severity), risk of necrosis of the humeral head (biological severity), and difficulty in surgical reconstruction (severity mechanics).

The overall value of the severity index is given by the weighted combination of the various factors or parameters that make up the index itself. The weighted values of the various factors or parameters taken into consideration in the calculation are given below by way of example but not by way of limitation.

The severity index for fractures of the proximal third of the humerus can be considered as a single value or also as the sum of three separate components as reported in the following formula: IS — IS M + IS L + IS c where IS M is the medial severity index, IS L is the lateral severity index, IS C is the lid severity index . In the rest of the description, all the components of the index will be specified in detail.

The comminution

As it is known, the comminution of a bony part is related to the energy of trauma and it was used as a measure of the seriousness of the joint injury.

A greater energy input involves a greater number of fragments and also of a smaller size. The energy absorbed by the bone during an impact is released when the bone breaks, so the more energy the bone can absorb and the greater the release in the event of a fracture, the greater the comminution.

The bone subjected to a high-speed impact will absorb more energy than a low-speed one. This phenomenon explains why rapid loading injuries at high speeds involve greater energy and cause greater comminution of the fracture and related displacement of the fragments.

Therefore, the parameter of the comminution turns out to be very important for the purposes of assessing the severity of a fracture of the proximal third of the humerus.

The solutions of the known technique define a classification of the comminution of a bone part as reported in the following table 1:

In the following description of the severity index, reference is made to this definition of the level of comminution of a bone part.

Medial zone severity index (IS M )

The calcar occupies the medial area of the control volume. It is an important hinge point and the more intact its surface, the more stable the lid is, so it is relevant both from the mechanical and the vascularity point of view.

In particular, as can be seen from figure 4, the plane d divides the medial zone formed by the calcar into two parts, which have a different role: the posterior calcar CP and the anterior calcar CA.

The posterior calcar CP is of greater relevance because it is located posteriorly and therefore closer to the posterior circumflex artery, so the circulation is greater. In particular, the posterior calcar CP, from a surgical point of view, is not very visible due to its position. Therefore, the surgical approach is more difficult because typically the surgical approach window is anterior or lateral.

The anterior calcar CA is of less relevance because it is more visible from a surgical point of view. The anterior calcar CA is close to the anterior arcuate artery but being more visible its management is easier.

In particular, the value associated with each configuration of the medial zone is that indicated in Table 2 (or links table of the medial zone of the control volume). Figure 7 shows a graphic representation of the fracture conformations of the medial zone.

Table 2

In addition, to each medial conformation, it is associated a second value ΔC M . This second value ΔC M depends on the additional level of comminution that is automatically found in the case examined. In particular, reference is made to the difference (ΔC M ) between the detected comminution C R and the medial reference comminution C MR if .

In particular, the value associated with the additional comminution of the medial zone is that indicated in Table 3 (or table of comminution additional of the medial zone of the control volume).

Table 3

In addition, a third value B LM is added to the evaluation of the severity index of the medial zone.

This third value B LM depends on the medial bone loss.

In particular, the value associated with medial zone bone loss is that indicated in Table 4 (or medial zone bone loss table of the control volume).

Table 4

Therefore, the sum of the three values described above allows the identification of the severity index of the medial zone (IS M) , as per the formula below:

Lateral zone severity index ( IS L ) As can be seen from Figure 4, the l plane divides the control volume into the lateral and medial zones.

The lateral area of the control volume is occupied by the lesser tuberosity PT and the greater tuberosity GT. These two tubercles are the seat of the insertions of the rotator cuff muscles, thus playing an important role in the movement and stabilization of the glenohumeral joint.

The greater tuberosity GT is important because it is attached to the rotator cuff. If the fragment is large, the vascular crisis is less, as is the weight of the tendon damage of the periosteum.

The more it is multi-fragmented, the more it is de- vascularized, it is less stable and it is more difficult to heal. For this reason, the greater tuberosity GT has greater importance than the lesser tuberosity PT, but less than in the calcar and the lid.

The lesser tuberosity PT has less relevance because it is difficult for it to necrosis. However, the lesser tuberosity PT is still an important part as it is located near the subscapularis tendon that attaches to it. This makes it important for a possible reconstruction.

In particular, the value associated with each configuration of the lateral zone is that indicated in Table 5 (or table of the bonds of the lateral zone of the control volume).

Figure 8 shows a graphical representation of the conformations of the fracture in the lateral zone.

Table 5

In addition, a second value is associated with each lateral configuration ΔC L . This second value depends on the additional level of comminution that is automatically found in the case examined.

In particular, reference is made to the difference (ΔC L ) between the detected comminution C LR and the reference medial comminution C LRif . In particular, the value associated with the additional comminution of the lateral zone is that indicated in Table 6 (or table of the additional comminution of the lateral zone of the control volume).

Table 6

In addition, a third value B LL is added to the evaluation of the severity index of the lateral zone IS L . This third value B LL depends on the bone loss of the lateral zone (lateral bone loss).

In particular, the value associated with the bone loss of the lateral zone is that indicated in Table 7 (or the table of the bone loss of the lateral zone of the control volume).

Table 7

Therefore, the sum of the three values described above allows the identification of the severity index of the lateral zone (IS L ) , according to the following formula:

Lid severity index ( IS C )

The lid is the joint part where the movement takes place and is therefore the area that most easily undergoes necrosis. The lid bone is a bone that is likely to easily de-vascularize. The blood vessels that pass through the calcar and lead to the lid are unique and very important.

For this reason, the lid has greater importance than all the other parts. In particular, the value associated with each position of the lid is that indicated in Table 8 (or table of the lid positions).

Figure 9 shows a graphical representation of the conformations of the positions of a lid.

In particular, the front luxation is the anterior dislocation is less serious because it is more easily identified without any damage when the reduction is carried out.

The posterior dislocation, on the other hand, is more difficult to manage due to its position, which is why it has a greater weight. Table 8

In addition, a second value is associated with each position of the lid.

This second value influences the severity of the fracture of the lid and depends on the size of the largest fragment of the lid, which is automatically found in the case examined.

In general, but this should not be considered as limiting, the fragment size is reported as a percentage of the total surface of the lid.

In particular, the value associated with the size of the largest fragment (HF - head fragment) of the lid is that indicated in Table 9 (or table of the severity of the head split).

Table 9 Also, a third value B LC is added to the valuation.

This third value B LC depends on the number of fragments in which the head is fractured and on the bone loss of the lid.

In particular, the value associated with the number of fragments and the bone loss of the lid is that indicated in Table 10 (or table of the number of fragments and bone loss of the lateral zone of the control volume).

Table 10

Therefore, the sum of the values described above allows the identification of the severity index of the lid (IS C ) :

In light of the above, the severity index will be given by the following formula: The calculation of the severity index has required the consideration of various factors or parameters characterizing the fracture.

In light of the foregoing, the present invention makes it possible to construct a real model of the humerus fracture and to compare this real model with a reference model of the intact bone structure of the humerus, in order to provide an optimal method of reconstruction of the humerus fracture by means of the calculation of indices of severity of such fracture.

With particular reference to figure 10, the system S, which implements the method for calculating a severity index in traumatic fractures, object of the present invention, essentially comprises a processing unit SI, display means S2, storage means S3, interface means S4 and power supply means S5.

The processing unit SI is the functional and main element of the system S, and, for this reason, it is connected and in communication with the other elements of the system S itself.

The processing unit SI is equipped with calculation and processing means, configured to run a software for calculating a severity index in traumatic fractures, as well as to interface with the other elements of the system S.

Said processing unit SI is also configured to control and coordinate the operation of the elements of the system S, with which it is connected and in communication .

Said processing unit SI can be constituted by a computer or a plurality of computers, possibly connected in the cloud.

In particular, in the embodiment described, said processing unit SI comprises a motherboard, a processor with a minimum frequency of 1,5 GHz, and a RAM memory having a minimum size of 4 Mb, said display means S2 comprises a monitor with a minimum resolution of 800x600, said storage means S3 comprises a hard-disk having a minimum size of 500Mb and the interface means S4 comprises a mouse and a keyboard. Finally, the supply means S5 allows supplying said system S.

However, in other embodiments of the present invention, said system S can comprise further elements other than those mentioned above.

Advantages

An advantage of the method according to the present invention is that of automatically analyzing the severity of traumatic fractures of the proximal third of the humerus, providing all the information necessary to carry out an appropriate therapeutic choice.

A further advantage of the method according to the present invention is that of constructing a real model of the fracture of the humerus and comparing this real model with a reference model of the intact bone structure of the humerus, to provide an optimal way of reconstructing the fracture by means of a calculation, fracture severity indexes.

The present invention has been described for illustrative but not limitative purposes, according to its preferred embodiments, but it is to be understood that modifications and/or changes can be introduced by those skilled in the art without departing from the relevant scope as defined in the enclosed claims.