Disease detection algorithms trainable with small number of positive samples
Disease detection from medical images is provided. In various embodiments, a medical image of a patient is read. The medical image is provided to a trained anatomy segmentation network. A feature map is received from the trained anatomy segmentation network. The feature map indicates the location of at least one feature within the medical image. The feature map is provided to a trained classification network. The trained classification network was pre-trained on a plurality of feature map outputs of the segmentation network. A disease detection is received from the trained classification network. The disease detection indicating the presence or absence of a predetermined disease.
1. A method comprising:
reading a medical image of a patient;
providing the medical image to a trained anatomy segmentation network, wherein the trained anatomy segmentation network comprises a contracting path, the contracting path comprising a repeated application of one or more convolutions and a pooling layer, and an expanding path following the contracting path, the contracting path decreasing an image size while doubling a number of channels, and the expanding path upsampling the image size by repetition;
receiving from the trained anatomy segmentation network a feature map, the feature map indicating a location of at least one feature within the medical image;
training a classification network using a set of feature maps, the set of feature maps including both normal and abnormal samples, and each of the set of feature maps being an output of the trained anatomy segmentation network;
concatenating the medical image and its feature map, thereby forming a combined input;
providing the combined input to the trained classification network; and
receiving from the trained classification network a disease detection, the disease detection indicating a presence or absence of a predetermined disease.
2. The method of claim 1 , further comprising:
providing at least one additional feature map to the trained classification network with the medical image.
3. The method of claim 1 , wherein the trained anatomy segmentation network comprises a convolutional neural network.
4. The method of claim 1 , wherein the trained classification network comprises a convolutional neural network.
5. The method of claim 1 , wherein the medical image is a computed tomography image.
6. The method of claim 1 , wherein the medical image is a cardiac computed tomography image.
7. The method of claim 1 , further comprising:
training the anatomy segmentation network on a set of medical images depicting normal anatomy.
8. The method of claim 1 , wherein a majority of the set of feature maps correspond to normal anatomy.
9. A system comprising:
a data store comprising a plurality of medical images;
a trained anatomy segmentation network, wherein the trained anatomy segmentation network comprises a contracting path and an expanding path following the contracting path, the contracting path decreasing the image size while increasing a number of channels, and the expanding path upsampling the image size;
a classification network;
a computing node comprising a computer readable storage medium having program instructions embodied therewith, the program instructions executable by a processor of the computing node to cause the processor to perform a method comprising:
reading a medical image of a patient;
providing the medical image to a trained anatomy segmentation network, wherein the trained anatomy segmentation network comprises a contracting path, the contracting path comprising a repeated application of one or more convolutions and a pooling layer, and an expanding path following the contracting path, the contracting path decreasing an image size while doubling a number of channels, and the expanding path upsampling the image size by repetition;
receiving from the trained anatomy segmentation network a feature map, the feature map indicating a location of at least one feature within the medical image;
training a classification network using a set of feature maps, the set of feature maps including both normal and abnormal samples, and each of the set of feature maps being an output of the trained anatomy segmentation network;
concatenating the medical image and its feature map, thereby forming a combined input;
providing the combined input to the trained classification network; and
receiving from the trained classification network a disease detection, the disease detection indicating a presence or absence of a predetermined disease.
10. The system of claim 9 , the method further comprising:
providing at least one additional feature map to the trained classification network with the medical image.
11. The system of claim 9 , wherein the trained anatomy segmentation network comprises a convolutional neural network.
12. The system of claim 9 , wherein the classification network comprises a convolutional neural network.
13. The system of claim 9 , wherein the medical image is a computed tomography image.
14. The system of claim 9 , wherein the medical image is a cardiac computed tomography image.
15. The system of claim 9 , the method further comprising:
training the anatomy segmentation network on a set of medical images depicting normal anatomy.
16. The system of claim 9 , wherein a majority of the set of feature maps correspond to normal anatomy.
17. A computer program product for disease detection from medical images, the computer program product comprising a computer readable storage medium having program instructions embodied therewith, the program instructions executable by a processor to cause the processor to perform a method comprising:
reading a medical image of a patient;
providing the medical image to a trained anatomy segmentation network, wherein the trained anatomy segmentation network comprises a contracting path, the contracting path comprising a repeated application of one or more convolutions and a pooling layer, and an expanding path following the contracting path, the contracting path decreasing an image size while doubling a number of channels, and the expanding path upsampling the image size by repetition;
receiving from the trained anatomy segmentation network a feature map, the feature map indicating a location of at least one feature within the medical image;
training a classification network using a set of feature maps, the set of feature maps including both normal and abnormal samples, and each of the set of feature maps being an output of the trained anatomy segmentation network;
concatenating the medical image and its feature map, thereby forming a combined input;
providing the combined input to the trained classification network; and
receiving from the trained classification network a disease detection, the disease detection indicating a presence or absence of a predetermined disease.
18. The computer program product of claim 17 , the method further comprising:
providing at least one additional feature map to the trained classification network with the medical image.
19. The computer program product of claim 17 , wherein the trained anatomy segmentation network comprises a convolutional neural network.
20. The computer program product of claim 17 , wherein the trained classification network comprises a convolutional neural network.