Apparatus and method for automated analyses of ultrasound images
A non-transitory computer readable storage medium has instructions executed by a processor to execute a feature extractor to form extracted features from images formed by a first ultrasound scanner and a second ultrasound scanner. A decision maker is operated to form predictions of medical conditions based upon patterns identified in the extracted features. An evaluator is utilized to compare the predictions to labels in images to form a feature extractor performance measure and a decision maker performance measure. A dissimilarity estimator is operated to compute a difference measure between a probability distribution of features extracted from images formed by the first ultrasound scanner and the second ultrasound scanner.
1. A non-transitory computer readable storage medium with instructions for execution by a processor, including instructions for:
training a machine learning model to analyze ultrasound data acquired using different ultrasound scanners by:
executing a feature extractor to form extracted features from images formed by a first ultrasound scanner and a second ultrasound scanner;
operating a decision maker to form predictions of medical conditions based upon patterns identified in the extracted features;
utilizing an evaluator to compare the predictions of medical conditions to labels in at least a subset of the images to form a feature extractor performance measure;
operating a dissimilarity estimator to compute a difference measure between the extracted features formed by the first ultrasound scanner and the extracted features formed by the second ultrasound scanner; and
updating one or more parameters of the feature extractor based on the feature extractor performance measure and the computed difference measure, so as to minimize a dissimilarity between the extracted features from the first ultrasound scanner and the second ultrasound scanner.
2. The non-transitory computer readable storage medium of claim 1 wherein the instructions for executing the feature extractor include instructions for:
executing a first feature extractor to extract features from images formed by the first ultrasound scanner; and
executing a second feature extractor to extract features from images formed by the second ultrasound scanner.
3. The non-transitory computer readable storage medium of claim 1 , wherein the labels are solely associated with images formed by the first ultrasound scanner.
4. The non-transitory computer readable storage medium of claim 1 , wherein the feature extractor comprises a plurality of convolution neural network processing layers.
5. The non-transitory computer readable storage medium of claim 1 , wherein the decision maker is a fully connected neural network.
6. The non-transitory computer readable storage medium of claim 1 , wherein the dissimilarity estimator is a fully connected neural network.
7. The non-transitory computer readable storage medium of claim 1 , further including instructions for:
utilizing the evaluator to compare the predictions of medical conditions to the labels in at least a subset of the images to form a decision maker performance measure, wherein the decision maker performance measure is used to refine one or more parameters of the decision maker.
8. A non-transitory computer readable storage medium with instructions for execution by a processor, including instructions for:
updating a trained model for analyzing ultrasound data acquired using different ultrasound scanners by:
executing a feature extractor that has been trained on images from a first ultrasound scanner, to extract features from images formed by a second ultrasound scanner, distinct from the first ultrasound scanner;
operating a decision maker that has been trained on the images from the first ultrasound scanner to form predictions of medical conditions based upon patterns identified in the extracted features;
utilizing an evaluator to compare the predictions of medical conditions to labels corresponding to the images from the second ultrasound scanner to form a feature extractor performance measure and a decision maker performance measure; and
updating one or more parameters of the feature extractor and the decision maker based on the feature extractor performance measure and the decision maker performance measure.
9. The non-transitory computer readable storage medium of claim 8 , wherein the feature extractor comprises a plurality of convolution neural network processing layers.
10. The non-transitory computer readable storage medium of claim 8 , wherein the decision maker is a fully connected neural network.
11. The non-transitory computer readable storage medium of claim 8 , wherein the feature extractor performance measure is used to refine the one or more parameters of the feature extractor.
12. The non-transitory computer readable storage medium of claim 8 , wherein the decision maker performance measure is used to refine the one or more parameters of the decision maker.
13. A computer system, comprising:
one or more processors; and
memory storing one or more programs, the one or more programs including instructions for:
training a machine learning model to analyze ultrasound data acquired using different ultrasound scanners by:
executing a feature extractor to form extracted features from images formed by a first ultrasound scanner and a second ultrasound scanner;
operating a decision maker to form predictions of medical conditions based upon patterns identified in the extracted features;
utilizing an evaluator to compare the predictions of medical conditions to labels in at least a subset of the images to form a feature extractor performance measure;
operating a dissimilarity estimator to compute a difference measure between the extracted features formed by the first ultrasound scanner and the extracted features formed by the second ultrasound scanner; and
updating one or more parameters of the feature extractor based on the feature extractor performance measure and the computed difference measure, so as to minimize a dissimilarity between the extracted features from the first ultrasound scanner and the second ultrasound scanner.
14. The computer system of claim 13 wherein the instructions for executing the feature extractor include instructions for:
executing a first feature extractor to extract features from images formed by the first ultrasound scanner; and
executing a second feature extractor to extract features from images formed by the second ultrasound scanner.
15. The computer system of claim 13 , wherein the labels are solely associated with images formed by the first ultrasound scanner.
16. The computer system of claim 13 , wherein the feature extractor comprises a plurality of convolution neural network processing layers.
17. The computer system of claim 13 , wherein the decision maker is a fully connected neural network.
18. The computer system of claim 13 , wherein the dissimilarity estimator is a fully connected neural network.
19. The computer system of claim 13 , wherein the memory further stores instructions for:
utilizing the evaluator to compare the predictions of medical conditions to the labels in at least a subset of the images to form a decision maker performance measure, wherein the decision maker performance measure is used to refine one or more parameters of the decision maker.