FIELD OF THE INVENTION

The invention relates to a method and apparatus for classifying abnormal growths in a human body, and a computer program product for carrying out the method when the computer program product is loaded and run on a computer. The invention further relates to a method and apparatus for retrieving a stored record of abnormal growths in a human body, and a computer program product for carrying out the method when the computer program product is loaded and run on a computer. The invention also relates to a medical imaging system comprising an apparatus for classifying abnormal growths in a human body or comprising an apparatus for retrieving a stored record of abnormal growths in a human body.

BACKGROUND OF THE INVENTION

In the course of diagnosis, staging, and therapy planning of a cancer patient, a number of clinical pathway decisions have to be made regarding further diagnostic work-up and therapy of the patient. The cancer staging and therapy decisions are influenced by the occurrence, position and size of abnormal growths, such as lymph nodes, primary tumors and metastatic tumors (metastases). The characteristics of such abnormal growths may be determined by any suitable diagnostic technique, including physical examination and diagnostic imaging. Additionally, the characteristics of such abnormal growths may be determined during therapeutic imaging. For example, imaging may be used during radiation therapy to guide a radiation beam, so that the beam is focused on an abnormal growth, such as cancerous tissue, to limit damage to the surrounding healthy tissue.

Because of the wide range of possible diseases and therapies, it is helpful to consider earlier cases which resemble the new case and to review their respective diagnosis, therapy, and outcome. To help facilitate this, various classification systems are used comprising multiple geometric descriptors, the classification system varying depending upon the nature of the cancer. For example, lung cancer is staged using the "International System for Staging Lung Cancer" and the TNM system (primary Tumor, lymph Nodes, distant Metastases). In the semantic description of this staging system, multiple geometric descriptors are used:

• "> 2 cm distal to carina" • "regional lymph node"

• "invasion more proximal than the lobar bronchus"

• "ipsilateral peribronchial"

• "contralateral nodes"

• "extends to hilar region but does not involve entire lung" • "nodes adjacent to distal lobar bronchi"

• "nodes lying adjacent to wall of esophagus"

• "lower paratracheal nodes defined by a horizontal line extending across the trachea and drawn tangentially to the cephalic border of the azygos vein"

By virtue of the overall staging system, these geometric descriptors thus have a direct impact on the diagnosis and therapy decision. However, using such semantic descriptions of the geometric descriptors, it is necessary that all abnormal growths are classified according to certain organ instances. This can be difficult and prone to errors because abnormal growths may be visually indistinguishable in the diagnostic data, they may be located using different imaging modalities, and they may only be differentiable by their location or by being embedded within a particular organ.

Additionally, because of the diversity of cancerous diseases and the seldom occurrence of rare types, it may be difficult and unreliable to recall the most resembling case from memory. Using a complex classification with a multitude of possibilities makes it difficult to locate similar cases in the past. If the classification system is prone to errors, this reduces the chances of a match even more. Moreover, the most similar case might have occurred at another institution where the classification may have been applied differently. Additionally, psychological 'satisfaction of search' may lead to the neglect of other similar cases (with possibly different outcomes) once one similar case has been recalled - this may however not be the "most similar" case on record.

SUMMARY OF THE INVENTION

It is an object of the invention to provide an apparatus for classifying abnormal growths in a human body. The invention is defined by the independent claims. Advantageous embodiments are defined in the dependent claims.

According to a first aspect of the invention, the object is achieved by means of an apparatus comprising an analyzer configured for determining the positions in the body of a first and second growth and for determining the position of a reference in the body; a calculator configured for calculating a first vector using the positions of the first and second growths and for calculating a second vector using the position of the reference and the position of the first growth; a classifier for defining a classification comprising an identification of the reference, the first vector and the second vector. The invention provides a classification method of normalized geometric features describing the tumor / node constellation.

Using the invention, one or more parts of the classification method may be performed by a computer processor comprised in the apparatus, thereby improving the reliability of the system and reducing the likelihood of error. With such a computerized system, it is possible to avoid a distinct classification of each tumor / lymph node to the anatomical region it belongs to, and instead just use a system of normalized relative coordinates, mutual distances, and likelihood atlases. Additionally, the classification translates to a graphical representation which is more familiar to users who analyze diagnostic images to detect such growths. According to an aspect of the invention, the analyzer is further configured for determining the position of a further reference in the body; the calculator is further configured for calculating a third vector using the position of the further reference and the position of the first growth, and the classifier is further configured for defining the classification which further comprises an identification of the further reference and the third vector.

The first growth is the reference position for the second growth. Errors in the position of the first growth may result in errors for the second growth. However, if a second reference is used to locate the first growth, then the likelihood of error in the position of the first growth is significantly reduced. According to an aspect of the invention, the calculator is further configured for calculating a third vector, using the position of the reference and the position of the second growth, and the classifier is further configured for defining the classification which further comprises the third vector. Although this requires an extra calculation, it may be advantageous to significantly reduce the likelihood of error in the position of the second growth because vectors pointing to it are calculated, each starting from a different point.

According to another aspect of the invention, the classifier is further configured for defining the classification which further comprises at least one textual representation. Textual representations give the user more contextual information than numerical representations, which reduces the chance of errors and allows the resulting classification to be manually checked.

According to an aspect of the invention, an apparatus is provided for retrieving a stored record of abnormal growths in a human body, comprising an apparatus for classifying abnormal growths in a human body according to the invention, wherein the classifier is further configured for defining a first classification of the stored record and for defining a second classification of a further record; the apparatus further comprising a comparator for comparing the first and second classification to determine a degree of similarity, and a retriever for retrieving the stored record if the degree of similarity is greater than a predetermined threshold.

The invention provides the user with a more intuitive classification system which makes it simpler to verify that records retrieved as a match to a current case are truly similar, because the classification system is less abstract. The users may embed their personal experience into a broader, more comprehensive clinical background ensuring that a high proportion of related cases have been taken into account during a record search.

According to an aspect of the invention, a method is provided for classifying abnormal growths in a human body, the method comprising determining the positions in the body of a first and second growth; calculating a first vector using the positions of the first and second growths; determining the position of a reference in the body; calculating a second vector, using the position of the reference and the position of the first growth; defining a classification comprising an identification of the reference, the first vector, and the second vector.

According to a further aspect of the invention, a method is provided for retrieving a stored record of abnormal growths in a human body, the method comprising defining a first classification of the stored record according to the method for classifying abnormal growths in a human body according to the invention, defining a second classification of a further record according to the method for classifying abnormal growths in a human body according to the invention; comparing the first and second classification to determine a degree of similarity, and retrieving the stored record if the degree of similarity is greater than a predetermined threshold.

According to another aspect of the invention, a medical imaging system is provided comprising an apparatus for classifying abnormal growths in a human body according to the invention or an apparatus for retrieving a stored record of abnormal growths in a human body according to the invention.

According to another aspect of the invention, a computer program product is provided for carrying out the method of classifying abnormal growths in a human body according to the invention or for carrying out the method of retrieving a stored record of abnormal growths in a human body, when said computer program product is loaded and run on a computer.

It will be appreciated by those skilled in the art that two or more of the above- mentioned embodiments, implementations, and/or aspects of the invention may be combined in any way deemed useful. Modifications and variations of the method, of the workstation, and/or of the computer program product, which correspond to the described modifications and variations of the image acquisition apparatus, can be carried out by a person skilled in the art on the basis of the present description.

A person skilled in the art will appreciate that the method may be applied to multidimensional image data, e.g., to 2-dimensional (2-D), 3-dimensional (3-D) or 4- dimensional (4-D) images, acquired by various acquisition modalities such as, but not limited to, standard X-ray Imaging, Computed Tomography (CT), Magnetic Resonance Imaging (MRI), Ultrasound (US), Positron Emission Tomography (PET), Single Photon Emission Computed Tomography (SPECT), and Nuclear Medicine (NM).

BRIEF DESCRIPTION OF THE DRAWINGS

These and other aspects of the invention are apparent from and will be elucidated with reference to the embodiments described hereinafter.

In the drawings: Figure 1 shows an example of a diagnostic scan in which positions of abnormal growths have been determined according to the invention,

Figure 2 depicts an example of a diagnostic scan in which organs have been detected according to the invention, Figure 3 shows an example of a diagnostic scan in which the positions of organs have been determined according to the invention,

Figure 4 depicts an example of a diagnostic scan in which the outline of the torso has been determined according to the invention, Figure 5 shows a first embodiment of a vector calculation in which the vectors between a reference organ and abnormal growths have been calculated according to the invention,

Figure 6 shows a second embodiment of a vector calculation in which the vectors between a reference organ and abnormal growths have been calculated according to the invention,

Figure 7 shows a third embodiment of a vector calculation in which the vectors between a reference organ and abnormal growths have been calculated according to the invention,

Figure 8 shows a fourth embodiment of a vector calculation in which the vectors between a reference organ and abnormal growths have been calculated according to the invention,

Figure 9 shows a fifth embodiment of a vector calculation in which the vectors between a reference organ and abnormal growths have been calculated according to the invention, Figure 10 shows a sixth embodiment of a vector calculation in which the vectors between a reference organ and abnormal growths have been calculated according to the invention,

Figure 11 shows a seventh embodiment of a vector calculation in which the vectors between a reference organ and abnormal growths have been calculated according to the invention,

Figure 12 depicts a method of classifying abnormal growths according to the invention, and

Figure 13 depicts a method of retrieving a stored record of abnormal growths in a human body according to the invention. The Figures are purely diagrammatic and not drawn to scale. Particularly for clarity, some dimensions are exaggerated strongly. Similar components in the Figures are denoted by the same reference numerals as much as possible.

DETAILED DESCRIPTION OF PREFERRED EMBODIMENTS: Figure 12 depicts a method 500 of classifying abnormal growths according to the invention. The method comprises: determining (510) the positions of at least two growths; calculating (520) a vector, using the positions of these two growths, the vector indicating the position of one growth relative to the other growth; determining (530) the position of a reference in the body; calculating (540) a vector, using the position of the reference and the position of one of the growths, the vector indicating the position of one of the growths relative to the reference; - defining (550) a classification comprising an identification of the reference and the two vectors.

The classification is sufficient to identify the position of the two growths - one growth is found at the end of the vector from the reference in the body, and the second growth is found by following the vector from the first growth. Determining (530) the position of a reference in the body may be executed before, in parallel with, or subsequent to determining (510) the positions in the body of the two growths. Determination of the three positions may be performed using different techniques and at different times - it is only important that this is performed in a reproducible and consistent way. In the method, the positions of abnormal growths are determined 510. These may be any type of abnormal growths encountered in human bodies, such as tumors, primary tumor, metastatic tumor, cysts, pseudocysts, neoplasms, lymph nodes, lymphomas, fibroids, and nevi. There is no limitation to the kind of growth which may be classified according to the invention - it must merely be possible to detect the growth, determine its position, and determine its position relative to one or more other growths and relative to a reference point.

It may be advantageous to extend the classification to include a category for each abnormal growth in order to cover the wide varieties of medical conditions and the different ways that such conditions may progress throughout the human body. For example, metastatic tumors are very common in the late stages of cancer. The spread of metastases may occur via the blood system, the lymphatic system or even through both routes.

Cancer cells may spread to lymph nodes (regional lymph nodes) near the primary tumor. This is called nodal involvement, positive nodes, or regional disease. If it is desirable to register the type of growth for each member in the classification, the skilled person may be faced with a choice about whether a growth should be labeled as a lymph node or a metastasis. Although localized spreading to regional lymph nodes near a primary tumor is not normally counted as metastasis, the skilled person may choose to label all tumors except the primary tumor as metastasis to simplify and speed up the determination of the classification based upon a particular diagnostic imaging operation. In addition to the above routes, metastasis may occur by direct seeding, e.g., in the peritoneal cavity or pleural cavity, so for such a case it may not be necessary to include lymph nodes in the classification.

For the classification according to the invention, two or more abnormal growths are preferably used. This may represent all the abnormal growths, or a selection. This may be determined by the skilled person, based upon a compromise between the speed, ease and accuracy of classification and the chance of incorrect matches when a search is made for records with a similar classification.

The abnormal growths may be detected using one or more modalities, such as CT, PET, MRI or even physical examination. Automated techniques for scanning images, such as automatic computer aided detection tools (CADe), to detect abnormalities and determine their positions are particularly suitable. However, the method of the invention may also be performed manually.

Figure 1 shows a diagnostic image 100 made using PET modality in which six abnormal growths 110 have been detected in a human torso: three growths 110 in the region of the right lung of the patient (visible on the left-hand side of the image 100) two growths 110 on the patient's left-hand side in the middle of the torso, and one growth 110 lower down on the patient's left-hand side, at the waist.

The abnormal growths are also indicated by arrows in the diagnostic image 100. Once growths 110 have been detected, the position of each growth may be determined through a suitable characteristic of the growth, using techniques known in the art, such as the geometric centre, the centroid position, the mean coordinate, the weighted mean coordinate, the center of mass, the center of density, and a point on the surface. These may be calculated by computer algorithms, or by manual measurements and manual calculations using known techniques.

In the method of Figure 12, a reference position is determined, 530, in the body. The reference may be any detectable and reproducible reference such as an organ, a part of an organ, a skeletal bone, a part of a skeletal bone, a muscle, a part of a muscle, a lymph node, part of a lymph node, a vessel, or part of a vessel. There is no limitation on the type of structure used as a reference point according to the invention - it must merely be possible to detect the structure and determine a reference position based on some characteristic of the structure.

A reproducible reference position based on a structure may be determined using techniques known in the art, such as the geometric centre, the centroid position, the mean coordinate, the weighted mean coordinate, the center of mass, the center of density, and a point on the surface.

Figure 2 shows the diagnostic image 100 of Figure 1, wherein three organs have been detected - bladder 210, the patient's left lung 220, which is shown on the right- hand side of the image 100, and the patient's right lung 230. Figure 3 depicts the same diagnostic image 100, wherein the outline of each of said three organs is transformed into respectively bladder reference 310, left lung reference 320 and right lung reference 320. One or more of these can suitably be used as a reference according to the invention. The choice of reference structure may be based upon reproducibility of position, or ease of detection in one or more modalities. For example, a skeletal bone as a reference may generally be detected using X-ray imaging.

The reference structure may also be selected based upon the medical condition to be classified. For example, the most common places for metastases to occur are the lungs, liver, brain, and the bones. There is also a propensity for certain tumors to seed in particular organs - often locations having similar characteristics. For example, prostate cancer usually metastasizes to the bones, colon cancer has a tendency to metastasize to the liver, stomach cancer often metastasizes to the ovary in women, breast tumor cells metastasize to bone tissue, and malignant melanoma spreads to the brain.

Figure 4 depicts the same diagnostic image 100, wherein the outline detection of the human torso is indicated as reference 340. The human torso is of predictable shape, comprising a head, thorax and limbs, which may therefore form the basis for determining a reproducible reference. Additionally, it may be more convenient to classify a skin-related condition, such as skin cancer, with respect to this as a reference.

In the method of Figure 12, a first vector is calculated, 520, using the positions of the first and second growths, and a second vector is calculated using the position of the reference and the position of the first growth.

Figure 5 depicts an embodiment of the invention, using the six abnormal growths 110 detected in the diagnostic image 100 of Figure 1. In Figure 5, the growths 110 are depicted in the same relative positions. The bladder reference 310 depicted in the diagnostic image 100 of Figure 3 is depicted in Figure 5 as the reference position for the growths 110. In Figure 5, the diagnostic image 100 itself is not depicted to improve the visibility of the vectors 361, 362, 363, 364, 365 between the different points of the constellation. The relationship between the different growths may be described as a constellation 410. In Figure 5, one of the growths 110 is selected as a root point 350, and the vectors 361, 362, 363, 364, 365 are calculated between the root point 350 and each growth 110. The selection of the growth 110 to be used as the root point 350 may be based on its proximity to the reference 310, its direction from the reference 310, its proximity to a particular organ, or the spatial relationship to all the other growths 110. Similar to the choice of reference point 310, the choice of root point 350 may depend upon the medical condition involved.

The Figure is limited to 2-dimensional vectors, but 3-dimensional vectors may also be used, and the convention adopted is that the sense of the vector directions 361, 362, 363, 364, 365 is away from the root point 350. Additionally, a vector 370 is calculated between the reference 310 and the root point 350, using the convention that the sense of the vector direction 370 is away from the reference point. Note that any other convention for the directions of the vectors may be used, so long as it is applied consistently. The vector is of the conventional type, having a magnitude and a direction. According to the classification definition 550 part of the method depicted in

Figure 12, the classification of the constellation 410 depicted in Figure 5 would be: bladder 310

Vector 370 [magnitude, direction]

Vector 361 [magnitude, direction] Vector 362 [magnitude, direction]

Vector 363 [magnitude, direction]

Vector 364 [magnitude, direction]

Vector 365 [magnitude, direction]

As will be apparent to the skilled person, the classification is sufficient to reconstruct the positions of the growths 110 in relation to the same reference point 310 in a different diagnostic image.

Additionally, the constellation 410 of Figure 5 may be extended to include a second reference point (not shown) and to calculate a second reference vector (not shown) between the second reference and the root point 350, using the arbitrary convention that the sense of the second reference vector direction is away from the second reference point. As errors in the position of the root point 350 may affect the accuracy with which the positions of the other growths 110 may be determined, use of a second reference point and a second reference vector may be advantageous in reducing such errors. According to the classification definition 550 part of the method depicted in

Figure 12, the classification of the constellation 410 depicted in Figure 5 would then be: bladder 310

Vector 370 [magnitude, direction] second reference second vector to root point 350 [magnitude, direction]

Vector 361 [magnitude, direction]

Vector 362 [magnitude, direction]

Vector 363 [magnitude, direction]

Vector 364 [magnitude, direction] Vector 365 [magnitude, direction]

As will be apparent to the skilled person, the classification is sufficient to reconstruct the positions of the growths 110 in relation to the same reference point 310 in a different diagnostic image.

Figure 6 depicts a further embodiment of the invention, which is an alternative classification of the constellation depicted in Figure 5, using the same six abnormal growths 110 in the same relative positions, and using the bladder reference 310 as the same reference position. For clarity, the labels of Figure 5 for the growths 110 have been omitted in Figure 6. The relationship between the different growths may also be described as constellation 420. One of the growths is selected as a root point 350, and the vectors 361, 362, 363, 364, 365 are calculated around the periphery of the constellation 420. The convention adopted here is that the sense of the direction of the vector is clockwise around the periphery.

Additionally, a vector 370 is calculated between the reference 310 and the root point 350, using the convention that the sense of the vector direction 370 is away from the reference point. Note that again any other convention for the directions of the vectors may be used, so long as it is applied consistently.

According to the classification definition 550 part of the method depicted in Figure 12, the classification of the constellation 420 depicted in Figure 6 would be: bladder 310

Vector 370 [magnitude, direction] Vector 361 [magnitude, direction]

Vector 362 [magnitude, direction]

Vector 363 [magnitude, direction]

Vector 364 [magnitude, direction] Vector 365 [magnitude, direction]

As will be apparent to the skilled person, the classification is sufficient to reconstruct the positions of the growths 110 in relation to the same reference point 310 in a different diagnostic image.

Figure 7 depicts a further embodiment of the invention, which is an alternative classification of the constellation depicted in Figure 5, using the same six abnormal growths 110 in the same relative positions, and using the bladder reference 310 as the same reference position. For clarity, the labels of Figure 5 for the growths 110 have been omitted in Figure 7. The relationship between the different growths may also be described as constellation 430. In this case no root point is selected - the vectors 361, 362, 363, 364, 365 are calculated around the periphery of the constellation 430 from the reference 310. In other words, the reference point 310 is included in the constellation 430, whereas in Figure 6, the constellation 420 does not include the reference point. The arbitrary convention is adopted that the sense of the direction of the vector is clockwise around the periphery.

In this embodiment, the vector 366 between the last growth 110 in the vector chain and the reference point is also calculated. To determine the positions of all the growths 110, this is not essential, but it may be advantageous to reduce an accumulation of errors in the vector calculations when the constellation 430 is determined or reconstructed.

According to the classification definition 550 part of the method depicted in Figure 12, the classification of the constellation 430 depicted in Figure 7 would be: bladder 310

Vector 370 [magnitude, direction]

Vector 361 [magnitude, direction]

Vector 362 [magnitude, direction]

Vector 363 [magnitude, direction] Vector 364 [magnitude, direction]

Vector 365 [magnitude, direction]

Vector 366 [magnitude, direction] As will be apparent to the skilled person, the classification is sufficient to reconstruct the positions of the growths 110 in relation to the same reference point 310 in a different diagnostic image.

Figure 8 depicts a further embodiment of the invention, which is an alternative classification of the constellation depicted in Figure 5, using the same six abnormal growths 110 in the same relative positions, and using the bladder reference 310 as the same reference position. For clarity, the labels of Figure 5 for the growths 110 have been omitted in Figure 8. The relationship between the different growths may be described as constellation 440. In this case, two root points 350, 351 are selected, creating constellation 440 with two clusters of three growths each. This may be preferable when the abnormal growths 110 show two clusters. This is possibly the case if the medical condition involved affects one of the bilaterally paired organs such as the kidneys or the lungs, or an organ with more than one lobe, such as the brain or liver. For the first cluster, the vectors 361, 362 are calculated between the first root point 350 and each growth 110 in the first cluster, using the arbitrary convention that the sense of the vector direction is away from the first root point 350.

Additionally, a vector 370 is calculated between the reference 310 and the first root point

350, using the arbitrary convention that the sense of the vector direction 370 is away from the reference point. For the second cluster, the vectors 363, 364 are calculated between the second root point 351 and each growth 110 in the second cluster, using the arbitrary convention that the sense of the vector direction is away from the second root point 351.

Additionally, a vector 371 is calculated between the reference 310 and the second root point

351, using the arbitrary convention that the sense of the vector direction 371 is away from the reference point.

According to the classification definition 550 part of the method depicted in Figure 12, the classification of the constellation 440 depicted in Figure 8 would be: bladder 310

Vector 370 [magnitude, direction]

Vector 361 [magnitude, direction]

Vector 362 [magnitude, direction] Vector 370 [magnitude, direction]

Vector 363 [magnitude, direction]

Vector 364 [magnitude, direction] As will be apparent to the skilled person, the classification is sufficient to reconstruct the positions of the growths 110 in relation to the same reference point 310 in a different diagnostic image.

Figure 9 depicts a further embodiment of the invention, which is an alternative classification of the constellation depicted in Figure 5, using the same six abnormal growths 110 in the same relative positions, and using the bladder reference 310 as the same reference position. For clarity, the labels of Figure 5 for the growths 110 have been omitted in Figure 9. The relationship between the different growths may be described as constellation 450. In this case, compared to constellation 410 in Figure 5, a different root point 350 is selected. The vectors 361, 362, 363, 364, 365 are calculated between the root point 350 and each growth 110. The arbitrary convention adopted is that the sense of the vector direction 361 to 365 is away from the root point 350. Additionally, a vector 370 is calculated between the reference 310 and the root point 350, using the arbitrary convention that the sense of the vector direction 370 is away from the reference point 310. According to the classification definition 550 part of the method depicted in

Figure 12, the classification of the constellation 450 depicted in Figure 9 would be: bladder 310

Vector 370 [magnitude, direction]

Vector 361 [magnitude, direction] Vector 362 [magnitude, direction]

Vector 363 [magnitude, direction]

Vector 364 [magnitude, direction]

Vector 365 [magnitude, direction]

Vector 366 [magnitude, direction] As will be apparent to the skilled person, the classification is sufficient to reconstruct the positions of the growths 110 in relation to the same reference point 310 in a different diagnostic image.

Figure 10 depicts a further embodiment of the invention, which is an alternative classification of the constellation depicted in Figure 8, using the same six abnormal growths 110 in the same relative positions, and using the bladder reference 310 as the same reference position. Additionally, left lung reference 320 as depicted in Figure 3 is used as a second reference position. The relationship between the different growths may be described as constellation 460. In this case, two root points 350, 351 are selected, creating a constellation with two clusters of three growths each. For the first cluster, the vectors 361, 362 are calculated between the first root point 350 and each growth 110 in the first cluster, using the arbitrary convention that the sense of the vector direction is away from the first root point 350. Additionally, a vector 370 is calculated between the first reference 310 and the first root point 350, using the arbitrary convention that the sense of the vector direction 370 is away from the reference point 310. For the second cluster, the vectors 363, 364 are calculated between the second root point 351 and each growth 110 in the second cluster, using the arbitrary convention that the sense of the vector direction is away from the second root point 351. Additionally, a vector 371 is calculated between the second reference 320 and the second root point 351, using the arbitrary convention that the sense of the vector direction 371 is away from the reference point.

According to the classification definition 550 part of the method depicted in Figure 12, the classification of the constellation 460 depicted in Figure 10 would be: bladder 310

Vector 370 [magnitude, direction] Vector 361 [magnitude, direction]

Vector 362 [magnitude, direction]

Right lung 320

Vector 371 [magnitude, direction]

Vector 363 [magnitude, direction] Vector 364 [magnitude, direction]

As will be apparent to the skilled person, the classification is sufficient to reconstruct the positions of the growths 110 in relation to the same reference points 310,320 in a different diagnostic image.

It will be apparent to the skilled person that the different types of vector calculations described in relation to Figures 5 to 10 and constellations 410 to 460, may be combined in a customized classification. Figure 11 depicts a further embodiment of the invention, which is an alternative classification of the constellation depicted in Figure 5, using the same six abnormal growths 110 in the same relative positions, and using the bladder reference 310 as the same reference position. For clarity, the labels of Figure 5 for the growths 110 have been omitted in Figure 11. The relationship between the different growths may be described as constellation 470. Similar to the constellation 410 in Figure 5, a root point 350 is selected and a vector 370 is calculated between the reference 310 and the root point 350, using the arbitrary convention that the sense of the vector direction 370 is away from the reference point 310. For each growth, the vectors 360-1 to 360-6 are calculated to each other member of the growth constellation 470. The skilled person will recognize this constellation 470 as a customized combination of different conventions from the constellations 410 to 460.

According to the classification definition 550 part of the method depicted in Figure 12, the classification of the constellation 470 depicted in Figure 11 would be: bladder 310

Vector 370 [magnitude, direction]

Vectors 360-1 [5 x magnitude, direction]

Vectors 360-2 [5 x magnitude, direction] Vectors 360-3 [5 x magnitude, direction]

Vectors 360-4 [5 x magnitude, direction]

Vectors 360-5 [5 x magnitude, direction]

Vectors 360-6 [5 x magnitude, direction]

As will be apparent to the skilled person, the classification is sufficient to reconstruct the positions of the growths 110 in relation to the same reference point 310 in a different diagnostic image.

Alternatively, the classification of the constellation 470 depicted in Figure 11 may be further customized such that no root point 350 is chosen, and that the vector 370-1 to 370-6 is calculated between each growth 110 and the reference point 310, and the vector 360- 1 to 360-6 is calculated between each growth 110 and each other member of the constellation. Although there is an increase in the number of vectors calculated, this may be advantageous because the position of each growth 110 is only partially dependent on the rest of the members in the constellation. This may be useful because the classification is less sensitive to errors in the positions of other growths 110, and less sensitive to the absence of other growths. According to the classification definition 550 part of the method depicted in Figure 12, the classification would then be: bladder 310

Vectors 360-1 [5 x magnitude, direction]

Vector 370-1 [magnitude, direction] Vectors 360-2 [5 x magnitude, direction]

Vector 370-2 [magnitude, direction]

Vectors 360-3 [5 x magnitude, direction]

Vector 370-3 [magnitude, direction]

Vectors 360-4 [5 x magnitude, direction] Vector 370-4 [magnitude, direction]

Vectors 360-5 [5 x magnitude, direction]

Vector 370-5 [magnitude, direction]

Vectors 360-6 [5 x magnitude, direction] Vector 370-6 [magnitude, direction]

As will be apparent to the skilled person, the classification is sufficient to reconstruct the positions of the growths 110 in relation to the same reference point 310 in a different diagnostic image.

If conventions are consistently applied, then the skilled person may easily envision simple algorithms to allow conversions between the different types of classification. As the relative positions of organs within the human body are generally well-defined, even completely different classifications may be converted. Such conversions may be made more reliable by labeling each growth with a suitable type indication, such as tumor, metastasis, or lymph node. To improve the accuracy of the classification, and therefore also the reliability of retrieval of an appropriate record, it may be advantageous to determine further generic parameters and include these in the classification. These may be determined from user- entered selections as well as from automatic image-based organ segmentations. For example: the number of growths, the mean diameter of the growths, the mean volume of the growths, the mean density of layers surrounding a growth, abundance of growths per organ patient relevant data, such as patient disposition and history data (gender, age, habits), and the modality used in determining the positions of growths.

It may also be advantageous to determine further parameters for one or more of the abnormal growths 110 and include these in the classification. These may be determined from user-entered selections as well as from automatic image-based organ segmentations. For example: the mass, the mean density, the mean radius, an ellipsoid measurement of the growth, the triangle mesh representation of a part of the surface, a categorization of the growth, and a characterization of the growth, such as "smooth", "irregular" contrast uptake in CT scans SUV-max values in PET scans location in or proximate to an organ, such as "upper left lung lobe".

To improve the levels of similarity, the tumor positions may be converted into a body normalized coordinate system (in order to be able to match patients of different body sizes), and within the organs into an organ normalized coordinate system ('percentile- coordinates').

For some parameters, textual representations may be used instead of numbers. For example, for the direction of a vector "ipso-lateral" and "contra-lateral" may be used. Textual representation has the advantage that it improves the readability of the classification for a medical professional. Figure 13 depicts a method 600 of retrieving a stored record of abnormal growths in a human body. This method comprises: defining (610) a classification of a stored record; defining (620) a classification of a record for which matches are sought in other records; - comparing (630) the two classifications to determine a degree of similarity, and retrieving (640) the stored record if the degree of similarity is greater than a predetermined threshold.

The stored record is classified 610 according to the classification method of the invention. This may be done immediately prior to retrieval, or it may have been performed at some earlier point in time, and the classification is stored as part of the record.

A second classification of a second record is defined 620 according to the invention. This second record is the starting point for the user to search for similar records. Of course, this may also be done immediately prior to retrieval, or it may have been performed at some earlier point in time, and the classification is stored as part of the record.

The first and second records may be based upon diagnostic data from the same patient at different times, or from different patients.

The first and second classifications are then compared 630 to determine a degree of similarity, and if a stored record shows a high degree of similarity, then the record is retrieved 640. Typically, the degree of similarity that determines whether a match has been found is variable so that the user can perform broad to narrow or narrow to broad searches.

Depending on the medical condition involved, the source of the diagnostic information, the quality of the medical information, the user may also wish to adapt the threshold for which matches are retrieved 640.

Using conventional techniques, the skilled person may also envision a self-learning system, for example using statistics, a neural network or fuzzy logic, where the system is given similar constellations to evaluate. The typical geometric relations between the members of the constellation and the references, as well as typical variations, can be learned by multidimensional classifiers (given a sufficiently high number of training cases) from the relevant parameters. In order to minimize inter-patient variability, the images can first be registered onto a mean patient model, possibly also a collection of typical patients, or eigenmodes of the mean model.

A numerical representation may be advantageous because of the rich background of mathematical distance metrics. All considered numerical features can be assembled into a numerical vector, and the distance or similarity between vectors can be computed e.g. by matrix multiplications, or by pattern classification schemes, such as linear or polynomial regression, support vector regression, neural networks, etc.

Similarly, the importance of individual parameters within a chosen classification may also be "learned", given a sufficiently large database of constellations. User intervention may also be used to speed-up or streamline this process. Appropriate weights can even be associated with each parameter.

Textual descriptions can be mapped onto discrete numerical (integer) values for processing purposes, particularly if the full range of possible descriptors has been enumerated. In this way, numerical and textual representations could also be used in a mixed fashion.

It may also be advantageous for the user to vary the number of classification parameters used during retrieval. For example, in a broad to narrow search, the user may first perform a comparison based upon a very broad classification, such as only the relative positions of the growths and the respective references, and if the number of successful matches is too high, a second filter may be used, such as the type of each growth.

It is also envisioned that the similarity algorithm may comprise both logical textual comparisons ('nodule in liver', 'smoker') as well as quantitative distance measures (in high dimensional feature spaces, normalized by feature distributions). It may also be advantageous to adapt the similarity algorithm to account for possible mismatches for a certain tumor, because the required imaging modality or an image of the required part of the body is not available.

An apparatus for classifying abnormal growths in a human body according to the embodiments of the invention may comprise an analyzer configured for determining the positions in the body of a first and second growth and for determining the position of a reference in the body; a calculator configured for calculating a first vector, using the positions of the first and second growths, and for calculating a second vector, using the position of the reference and the position of the first growth; a classifier for defining a classification comprising an identification of the reference, the first vector and the second vector. Such an apparatus may be advantageously comprised in a medical imaging system.

Such a first apparatus may also be comprised in a second apparatus for retrieving a stored record of abnormal growth in a human body, comprising said first apparatus wherein the classifier is further configured for defining a first classification of the stored record and for defining a second classification of a further record; the second apparatus further comprising a comparator for comparing the first and second classification to determine a degree of similarity, and a retriever for retrieving the stored record if the degree of similarity is greater than a predetermined threshold. Such a second apparatus may be advantageously comprised in a medical imaging system. The embodiments of these apparatus indicate parts such as an analyzer configured for determining the positions, a calculator configured for calculating vectors, a classifier for defining a classification, a comparator for comparing classifications and a retriever for retrieving records. It will be apparent to the skilled person that this is a separation based upon functionality and does not necessarily indicate separate, discrete pieces of hardware. For example, all functions may be performed by a single processor, each function may be assigned to a separate processor or each function may be divided over a plurality of processors. Additionally, the physical location of a plurality of processors is unimportant - they may be distributed over one or more locations, connected in some way by a communication pathway such as a network, or even over the Internet. The user friendliness of the apparatus may also be improved by providing the user with the most similar case, together with a visual representation of the constellation being matched. The user may also be provided with information about the record retrieved to support clinical decisions, such as: a visual representation of the constellation, the respective diagnosis, the history, the chosen therapy, and the therapy response. In a practical implementation of such an apparatus, the user may also be provided with ancillary information about the further progression of the disease in similar cases. If some of the similar cases also include information about patient treatment and outcome, the apparatus can serve as a gateway to the evaluation and ranking of alternative therapeutic options for cases-of- interest in relation to what has been achieved in similar cases.

In summary, the invention relates to a method of classifying abnormal growths in a human body. For support of diagnosis and clinical decisions in respect of cancer patients, a database of earlier cases is searched, and the cases with the most resembling constellation of positions, sizes etc. of such growths are presented to the clinical user, together with their respective diagnosis, therapy choice and outcome.

The resemblance is not based on the similarity of an individual growth, such as a tumor, but rather on the similarity of the constellation of the entirety of, for example, tumors, metastases and lymph nodes, as identified in diagnostic measurements and on scans of the patient. Techniques such as physical examination, CT scan, PET scan, and MRI scan may be used to collect the data for the constellation records in the database.

It should be noted that the above-mentioned embodiments illustrate rather than limit the invention, and that those skilled in the art will be able to design many alternative embodiments without departing from the scope of the appended claims. In the claims, any reference signs placed between parentheses shall not be construed as limiting the claim. Use of the verb "comprise" and its conjugations does not exclude the presence of elements or steps other than those stated in a claim. The article "a" or "an" preceding an element does not exclude the presence of a plurality of such elements. The invention may be implemented by means of hardware comprising several distinct elements, and by means of a suitably programmed computer. In the device claim enumerating several means, several of these means may be embodied by one and the same item of hardware. The mere fact that certain measures are recited in mutually different dependent claims does not indicate that a combination of these measures cannot be used to advantage.