Computer-aided detection (CAD) of a disease

FIELD OF THE INVENTION

The present invention relates to a method for analyzing a medical image data set obtained from a medical imaging modality, such as computer tomography (CT), magnetic resonance imaging (MRI), positron electron tomography (PET), single photon emission computed tomography (SPECT), ultrasound scanning, and rotational angiography, and other medical imaging modalities. The invention also relates to a corresponding computer system and a corresponding computer program product.

BACKGROUND OF THE INVENTION In recent years advances within medical imaging, including modalities like computer tomography (CT) or magnetic resonance imaging (MRI) etc., have provided for improved detection of various types of diseases, in particular cancer tumours. This progress within medical imaging has resulted in a vast amount of medical image data that has to be carefully analysed and evaluated in order to obtain sound diagnostic results there from. This analysis phase of medical images can be quite time consuming even for trained personnel e.g. radiologists. Experts trained in analysis of medical imaging are also a precious resource making medical imaging analysis a possible bottleneck of the diagnostic process.

Accordingly, there is an increasing need for computer-aided detection (CAD) helping the radiologist to find and identify possible diseases in the medical image data available. It should be emphasised that the CAD systems are not, at least currently, intended to replace the radiologist but merely to support or guide him during the image analysis. Also the benefit of using CAD is also questioned for some diseases, thus great care should be taking in implementing a CAD system for a disease. In particular, the receiver operating characteristics (ROC) should be well understood, and preferably controlled, before application of a CAD system.

US 2005/0200763 discloses such a computer assisted method of detecting and classifying lung nodules within a set of CT images. Initially, lung and esophagus segmentation is performed to identify the regions of the CT images in which to search for potential lung nodules. The lungs are processed to identify the left and right sides of the lungs

and each side of the lung is divided into subregions including upper, middle and lower subregions and central, intermediate and peripheral subregions. The computer analyzes each of the lung regions to detect and identify a three-dimensional vessel tree representing the blood vessels at or near the mediastinum. The computer then detects objects that are attached to the lung wall or to the vessel tree to assure that these objects are not eliminated from consideration as potential nodules. Thereafter, the computer performs a pixel similarity analysis on the appropriate regions within the CT images to detect potential nodules and performs one or more expert analysis techniques using the features of the potential nodules to determine whether each of the potential nodules is or is not a lung nodule. Thereafter, the computer uses further features, such as speculation features, growth features, etc. in one or more expert analysis techniques to classify each detected nodule as being either benign or malignant, possibly the nodule is assigned with a certain probability of benign or malignant. The computer then displays the detection and classification results to the radiologist to assist the radiologist in interpreting the CT exam for the patient. However, in order to perform this method a relatively large computational power has to be provided, which may limit the spreading of the disclosed CAD method. Because the computing power is limited, there is a need for a way of simplifying the necessary calculation in order to have a practically working CAD system.

Another disadvantage of the method disclosed in US 2005/0200763 is the fixed point of operation on the receiver operation characteristics (ROC) known from detection theory. A compromise between sensitivity and specificity also has to be made but this may lead to undesirable high number of false positives (FP) or, even worse, false negatives.

Hence, an improved a method for analyzing a medical image data set would be advantageous, and in particular a more efficient and/or reliable method would be advantageous.

SUMMARY OF THE INVENTION

Accordingly, the invention preferably seeks to mitigate, alleviate or eliminate one or more of the above mentioned disadvantages singly or in any combination. In particular, it may be seen as an object of the present invention to provide a method for easier and/or faster analysis of a medical image data set for likelihood for a disease.

Thus, the above described object and several other objects are intended to be obtained in a first aspect of the invention by providing a method for analyzing a medical image data set, the method comprising the steps of: a) segmenting the medical image data set using an anatomical model, b) analyzing the segmented data for characteristics of a disease resulting in a set of analysis data, and c) evaluating the set of analysis data with respect to the disease, wherein at least one of the above steps a), b) and c) comprises as an input a position dependent probability (P r) for the disease. The invention is particularly, but not exclusively, advantageous for obtaining an improved computer-aided detection (CAD) method facilitating that more efficient computations can be performed because the degree of analysis in a certain region of the part of the patient, e.g. the lung, can be adjusted or tailored to the level of probability of the disease in the region. Thus by application of the present invention it is possible to increase computational speed and thereby diseases like cancer, in particular cancer nodules in the lungs, can be more effectively found from medical image analysis.

It should be noted that the present invention applies a positional or regional dependent probability as an input in certain part of the computer-aided detection of a possible diseased structured within the medical images. The position dependent probability is an a priori probability that may be obtained from medical journal, well-established medical statistics, authority approved software and other reliable sources. One example is J. J. Nigro et al, "PREVALENCE AND LOCATION OF NODAL METASTASES IN DISTAL ESOPHAGEAL DENOCARCINOMA CONFINED TO THE WALL: IMPLICATIONS FOR THERAPY", The Journal of Thoracic and Cardiovascular Surgery Volume 117, Number 1. The application of region dependent probabilities is not to be confused with the fact that some CAD methods from an evaluation step of the suspected object in medical images, e.g. nodules in the lung, have as a result or output a certain probability that the suspected object is malignant or benign. Thus for example the CAD method disclosed in US 2005/0200763 may use a classifier routine to assign likelihood with malignancy to a candidate nodule in the lung. However, this is a preliminary result or final result of the method and is not used as an input to simplify subsequent calculations during the CAD process.

In the context of the present invention, it is also to be understood that the term "disease" is to construe broadly, and comprising in particular the situation where an

anomalous structure of examined tissue is not malignant i.e. the anomalous structure may be benign. One example may be the finding of a tumour that is benign, such a tumour also being a "disease" in the context of the present invention.

Optionally, the step a) of segmenting the medical image data set using an anatomical model may comprise as an input a position dependent probability (P r) for the disease. Thus, segmentation resolution may be fine or coarse depending on the level of position dependent probability (P r). In particular, one or more regions of the medical image data set may not be segmented due to the position dependent probability (P d) for the disease in the one or more regions being below a predetermined threshold value thereby excluding regions with quite low probability in order to simplify further processing and computations.

Optionally, the step b) of analyzing the segmented data for characteristics of the disease resulting in a set of analysis data may comprise as an input a position dependent probability (P r) for the disease. In particular, the position dependent probability (P r) may be applied to change the operation point of the receiver operation characteristics (ROC) of the computer-aided detection (CAD) method or algorithm. The level of probability (P r) may determine the direction of the change. Thus, in one embodiment the specificity of the performed analysis may be increased in dependency of the position dependent probability (P r) for the disease for example if specificity should be increased in a relative high P r region. Alternatively, the sensitivity of the performed analysis may be increased in dependency of the position dependent probability (P r) for the disease. Thus, for relatively low P_r regions sensitivity may be increased.

Optionally, the step c) of evaluating the set of analysis data with respect to the disease may comprise as an input a position dependent probability (P r) for the disease. Thus, statistical classification of e.g. nodules, using for instance Bayesian network can have as input P r for that specific region.

In one embodiment, the disease may be a tumour and the corresponding position dependent probability (P r) may then be the probability for having a tumour at that position (P r tum). In that case, the position dependent probability for having a tumour (P r tum) may be combined with a probability that the tumour is malignant (P r tum M) and/or benign (P r tum B). Thus, there is introduced two levels of probabilities with respect to tumours in lungs for example. Similarly, the multiple levels of probabilities may be introduced within the teaching of the present invention.

Preferably, the position dependent probability (P r) for the disease may be obtained from a reference database of position dependent probabilities. The database can in

turn be based on literature sources, standard works etc. The database may be remotely based. Possibly, the reference database of position dependent probabilities can be internal firmware.

In one embodiment, the position dependent probability (P r) for the disease may be further dependent on characteristics of the patient being examined. Thus, P r may be dependent on the patient features like age, sex, habits (especially smoking habits or other risk increasing behaviour), and other characteristics known to increase or decrease the probability of incurring a certain disease.

In another embodiment, the position dependent probability (P r) for the disease may be combined with a probability (P CAD) for finding the disease dependent on the CAD method being applied. Such knowledge about the CAD method may require extensive testing and/or experience with the CAD method in question. Once such knowledge is available it may however relatively easy be applied in the context of the present invention.

In yet another embodiment, where the position dependent probability (P r) for the disease may be combined with a probability (P IM) for finding the disease dependent on the imaging modality used to obtain the medical image data set. This may be e.g. the probability for beam starvation or metal artefacts in CT, or motion error bands in MRI etc.

In a preferred embodiment, one or more position dependent probabilities (P r) for the disease may be displayed to a user, e.g. a radiologist, together with a medical image. An indication to a user of the value of P r may be made on a screen, but it could also be a value combined therewith or derived there from.

In a second aspect, the present invention relates to a computer system arranged for analyzing a medical image data set, the system comprising: segmentation means for segmenting the medical image data set using an anatomical model, - analysis means for analyzing the segmented data for characteristics of a disease resulting in a set of analysis data, and evaluation means for evaluating the set of analysis data with respect to the disease, wherein at least one of the segmentation means, the analysis means and the evaluation means, is arranged to receive as an input a position dependent probability (P r) for the disease.

In a third aspect, the invention relates to a computer program product being adapted to enable a computer system comprising at least one computer having data storage means in connection therewith.

This aspect of the invention is particularly, but not exclusively, advantageous in that the present invention may be accomplished by a computer program product enabling a computer system to carry out the operations of the invention when down- or uploaded into the computer system. Such a computer program product may be provided on any kind of computer readable medium, or through a network.

The individual aspects of the present invention may each be combined with any of the other aspects. These and other aspects of the invention will be apparent from the following description with reference to the described embodiments.

BRIEF DESCRIPTION OF THE FIGURES

The various aspects of the invention will now be described in more detail with regard to the accompanying figures. The figures show one way of implementing the present invention and is not to be construed as being limiting to other possible embodiments falling within the scope of the attached claim set. Figure 1 schematic drawing of a combined imaging modality and computer system for one embodiment of the present invention,

Figure 2 is a schematic flow chart of the method steps according to an embodiment of the present invention,

Figure 3 is a receiver operating characteristics (ROC) graph, Figure 4 is a cross-sectional CT image of a patient illustrating a false positive

(FP) nodule candidate,

Figure 5 is a cross-sectional CT image of a patient illustrating a true nodule (TP),

Figure 6 is a schematic flow chart of a method according to the invention.

DETAILED DESCRIPTION OF AN EMBODIMENT

Figure 1 schematic drawing of a combined imaging modality and computer system 1 for one embodiment of the present invention. The computer system 1 being arranged for performing CAD from a medical image data set 20 obtained from a medical imaging modality IM, such as computer tomography (CT), magnetic resonance imaging

(MRI), positron electron tomography (PET), single photon emission computed tomography (SPECT), ultrasound scanning, and rotational angiography, or any other medical imaging modalities. The transmission from the modality IM to the unit 12 can be via a dedicated

connection means 11 (short range or long range, possibly via internet) or by wireless transmission.

The unit 12 of the computer system 1 is arranged for performing computer- aided detection (CAD) of a disease on a medical image data set 20. Segmentation means 13 are provided for segmenting the medical image data set 20 using an anatomical model, preferably an augmented model. For general reference to segmentation in medical imaging, the reader is referred to C. Xu, D. Pham, and J.L. Prince, "Medical Image Segmentation Using Deformable Models," Handbook of Medical Imaging, Volume 2: Medical Image Processing and Analysis, pp. 129-174, edited by J.M. Fitzpatrick and M. Sonka, SPIE Press, May 2000, which is hereby incorporated by reference in its entirety.

Additionally, analysis means 14 for analyzing the segmented data for characteristics of the disease is provided in unit 12, the analysis resulting in a set of analysis data 25, i.e. candidates for the disease. Also, evaluation means 15 for evaluating the set of analysis data 25 with respect to the disease. Evaluation of the analysis data 25 may be performed by any kind of expert algorithm available to the skilled person, such Bayesian network, artificial neural network (ANN), fuzzy logical based network, etc.

Applying the principle of the invention, at least one of the segmentation means 13, the analysis means 14 and the evaluation means 15, is arranged to receive as an input to their processing a position dependent probability P r for the disease as indicated within the unit 12. Final and/or preliminary results are subsequently transferred via dedicated connection means 16 (short range or long range, possibly via internet) or by wireless transmission to a general user interface (UI) 17 where a radiologist or similarly trained personnel can benefit from the improved CAD process provided by the present invention. For a range of possible applications an individualized general patient model, or aspects of it, can be shown to the medically trained user, either for verification or for facilitation of reporting (e.g. automated description or picture generation for lesion location definition). Potentially, the attached meta- information, such as locally variable lesion probabilities, can also be shown to the clinician for diagnosis support.

Figure 2 is a schematic flow chart of the method steps according to an embodiment of the present invention for performing computer-aided detection (CAD) of a disease (including also an anomalous structure that is not malignant) on a medical image data set 20, the data set 20 being obtained from an imaging modality IM as described above. After segmenting the medical image data set 20 using an anatomical model relevant for the patient

under examination, the medical image data set 20 can be divided into regions where each region have an associated position dependent probability P r for the disease.

For merely illustrative reasons, the number of regions in Figure 2 is four; 20a, 20b, 20c, and 2Od, with associated probabilities P a, P b, P c, and P d, respectively. Of course the number of regions can be adapted to the number of available probabilities; the higher the number of probabilities, the more finely divided the data set 20. Assuming that the probability P d in region 2Od is below a predetermined limit, e.g. 1%, it is possible to discard this region 2Od for segmentation and/or any further analysis and evaluation. This is shown symbolically in Figure 2, part b, where region 2Od is crossed over. During analysis of the segmented data for characteristics of the disease, the region 2Od is not studied resulting in a simplification of CAD process. The analysis may also be adapted to the region dependent probabilities P r. Thus, the internal parameter of the computer-aided detection can in general be made dependent, either directly or indirectly, on these probabilities. Also, the mesh or grid for analysis can be coarser and finer depending on the probabilities. The analysis results in a set of analysis data 25, i.e. candidates from the disease. In Figure 2 this is symbolically indicated as round circle, e.g. a tumour, but the exterior shape of the analysis data 25 could of course be of any kind.

In the last step, shown in part c of Figure 2, there is performed an evaluation of the set of analysis data 25 with respect to the disease. In generally, the evaluation can be either positive POS or negative NEG with respect to the disease for the given analysis data 25. According to the present invention, the probability P b for the disease in the region 20b where the candidate 25 was found is actively used as an input in the evaluation process. Thus, the probability P b can for example be used as input in one or more nodes in a Bayesian network used for evaluation of the disease candidate 25 with respect to the disease. Figure 3 is a receiver operating characteristics (ROC) graph for a binary classifier system with the sensitivity SENS on the vertical scale and one minus the specificity 1-SPEC on the horizontal scale, the present invention being particularly suited for modifying receiver operating characteristics 300 (ROC) of a computer aided detection (CAD) method. Thus, the position dependent probability P r may be applied to change a current operation point 303 a of the receiver operation characteristics 300 (ROC) of the computer-aided detection (CAD) algorithm. Accordingly, the specificity SPEC of the performed analysis can be increased in dependency of the position dependent probability P r by moving along arrow 302 resulting in a new operation point 303c. The specificity can be increased when for example the probability P r for finding the disease in a certain region is high.

Alternatively, the sensitivity SENS of the performed analysis can be increased in dependency of the position dependent probability by moving along arrow 301 resulting in yet another operation point 303b. Thus, for low P r regions the sensitivity can be increased. Notice that this may be practically implemented by changing internal parameters of CAD algorithm; it is generally not possible to shift the entire curve 300 for a certain CAD algorithm. Thus, the present invention can provide a direction for a change in the operation point 303a on the ROC curve 300, the invention does not change the ROC curve 300 itself.

Figure 4 is a cross-sectional CT image of a patient illustrating a false positive (FP) nodule candidate (as indicated by the solid arrow). Purely geometrically, this false positive candidate (a hilus vessel branching out of the mediastinum) looks like a mass, also the size is realistic. This is probably a category of FPs which could only be excluded by anatomical knowledge, e.g. in the form of region dependent probability P r, not by pure shape features. Using anatomical knowledge of typical vessel positions and sizes this FP could have been detected as such and be suppressed using the principles of the present invention.

Figure 5 is a cross-sectional CT image of a patient illustrating a true nodule (true positive, TP) as indicated by the solid arrow i.e. a counter-example of a true nodule/ lesion at the mediastinum of similar geometric appearance as compared to Fig. 4. The decision of a true nodule will be more probable using a variant of the present invention with probabilities of non-diseaseus structures because the anatomical model does not predict a likelihood of a vessel being in that region.

Figure 6 is a schematic flow chart of a method according for performing computer-aided detection (CAD) of a disease on a medical image data set 20 obtained from an imaging modality IM, cf. Figure 1 , the method comprising the steps of: a) segmenting the medical image data set 20 using an anatomical model, b) analyzing the segmented data for characteristics of the disease resulting in a set of analysis data 25, and c) evaluating the set of analysis data 25 with respect to the disease, wherein at least one of the above steps a), b) and c) comprises as an input a position dependent probability P r for the disease.

The invention can be implemented by means of hardware, software, firmware or any combination of these. The invention or some of the features thereof can also be implemented as software running on one or more data processors and/or digital signal processors.

The individual elements of an embodiment of the invention may be physically, functionally and logically implemented in any suitable way such as in a single unit, in a plurality of units or as part of separate functional units. The invention may be implemented in a single unit, or be both physically and functionally distributed between different units and processors.

Although the present invention has been described in connection with the specified embodiments, it should not be construed as being in any way limited to the presented examples. The scope of the present invention is to be interpreted in the light of the accompanying claim set. In the context of the claims, the terms "comprising" or "comprises" do not exclude other possible elements or steps. Also, the mentioning of references such as "a" or "an" etc. should not be construed as excluding a plurality. The use of reference signs in the claims with respect to elements indicated in the figures shall also not be construed as limiting the scope of the invention. Furthermore, individual features mentioned in different claims, may possibly be advantageously combined, and the mentioning of these features in different claims does not exclude that a combination of features is not possible and advantageous.