Automated screening of histopathology tissue samples via analysis of a normal model
Systems and methods are provided for screening histopathology tissue samples. An anomaly detection system is trained on a plurality of training images. Each of the plurality of training images represents a tissue sample that is substantially free of abnormalities. A test image, representing a tissue sample, is provided to the anomaly detection system. A deviation from normal score is generated for at least a portion of the test image. The deviation from normal score represents a degree of abnormality in the tissue sample represented by the test image.
1. A system for screening histopathology tissue samples, comprising:
a processor; and
a non-transitory computer readable medium storing executable instructions comprising:
an anomaly detection system trained on a plurality of training images, each of the plurality of training images representing a tissue sample that is substantially free of abnormalities, the anomaly detection system extracting a plurality of features from each of the plurality of training images that includes a set of features derived from a set of multi-scale histograms of color features;
an image interface that receives a test image and provides the test image to the anomaly detection system, the anomaly detection system generating a deviation from normal score for at least a portion of the test image, the deviation from normal score representing a degree of abnormality in the tissue sample represented by the test image; and
a user interface that provides the deviation score to a user at an associated output device.
2. The system of claim 1 , wherein the anomaly detection system extracts a plurality of features from each of the plurality of training images, the plurality of features including a set of features derived from a latent space of a variational autoencoder.
3. The system of claim 1 , wherein the anomaly detection system comprises an isolation forest algorithm that determines the deviation from normal score as an as average path length along a decision tree to isolate an observation represented by the test image.
4. The system of claim 1 , wherein the anomaly detection system comprises a single class support vector machine that generates the deviation from normal score.
5. The system of claim 1 , wherein the anomaly detection system comprises a plurality of expert systems that provide respective outputs an arbitrator to provide a deviation from normal score.
6. The system of claim 1 , wherein the anomaly detection system extracts a plurality of features from each of the plurality of training images, the plurality of features including a set of features derived from latent vectors generated by a convolutional neural network.
7. The system of claim 1 , wherein the anomaly detection system extracts a plurality of features from each of the plurality of training images, the plurality of features including a set of features formed in at least one hidden layer of a generative adversarial network.
8. A method for evaluating the effect of a therapeutic on an organ of a subject from histopathology tissue samples, the method comprising:
administering the therapeutic to a subject;
extracting a tissue sample to be tested from the organ of the subject;
training an anomaly detection system on a plurality of training images, each of the plurality of training images representing a tissue sample that is substantially free of abnormalities;
providing a test image, representing the tissue sample to be tested, to the anomaly detection system; and
generating a deviation from normal score for at least a portion of the test image at the anomaly detection system, the deviation from normal score representing a degree of abnormality in the tissue sample represented by the test image.
9. The method of claim 8 , further comprising extracting the tissue sample to be tested via a biopsy of a human patient.
10. The method of claim 8 , wherein the deviation from normal score is generated for each of a plurality of locations on the test image.
11. The method of claim 10 , further comprising comparing the deviation from normal score at each of the plurality of locations to a threshold value to identify a location in the tissue sample represented by the test image containing an abnormality.
12. The method of claim 10 , wherein the plurality of locations on the test images are a plurality of pixels comprising the test image.
13. The method of claim 8 , wherein generating the deviation from normal score for at least a portion of the test image at the anomaly detection system comprises extracting a plurality of features from the test image, the plurality of features including a set of features derived from one of a latent space of a variational autoencoder, a dense Speeded-Up Robust Features feature detection process, a set of multi-scale histograms of color and texture features, a set of latent vectors generated by a convolutional neural network, and a hidden layer of a generative adversarial network.
14. The method of claim 8 , wherein generating the deviation from normal score for at least a portion of the test image at the anomaly detection system comprises providing the extracted plurality of features to one of an isolation forest algorithm and a single class support vector machine.
15. A method for screening histopathology tissue samples, comprising:
training an anomaly detection system on a plurality of training images, each of the plurality of training images representing a tissue sample that is substantially free of abnormalities;
providing a first test image, representing a first tissue sample to be tested, to the anomaly detection system;
generating a first deviation from normal score for at least a portion of the first test image at the anomaly detection system, the first deviation from normal score representing a degree of abnormality in the tissue sample represented by the first test image, administering a therapeutic to a subject associated with the first tissue sample;
extracting a second tissue sample from the subject;
providing a second test image, representing the second tissue sample to be tested, to the anomaly detection system;
generating a second deviation from normal score for at least a portion of the second test image; and
comparing the second deviation from normal score to the second deviation from normal score to the first deviation from normal score to determine an efficacy of the therapeutic.
16. The method of claim 15 , wherein the first deviation from normal score is generated for each of a plurality of locations on the test image.
17. The method of claim 16 , further comprising comparing the first deviation from normal score at each of the plurality of locations to a threshold value to identify a location in the first tissue sample containing an abnormality.
18. The method of claim 16 , wherein the plurality of locations on the first test image are a plurality of pixels comprising the test image.
19. The method of claim 15 , wherein generating the first deviation from normal score for at least a portion of the first test image at the anomaly detection system comprises extracting a plurality of features from the first test image, the plurality of features including a set of features derived from one of a latent space of a variational autoencoder, a dense Speeded-Up Robust Features feature detection process, a set of multi-scale histograms of color and texture features, a set of latent vectors generated by a convolutional neural network, and a hidden layer of a generative adversarial network.
20. The method of claim 15 , wherein generating the first deviation from normal score for at least a portion of the first test image at the anomaly detection system comprises providing the extracted plurality of features to one of an isolation forest algorithm and a single class support vector machine.