Similarity based per item model selection for medical imaging
Embodiments may include techniques to choose a model based on a similarity of computed features of an input to computed features of several models in order to improve feature analysis using Machine Learning models. A method of image analysis may comprise extracting a training feature vector corresponding to each of the plurality of machine learning models from each validation image from a plurality of machine learning models trained using a plurality of validation images, extracting from a new image a new feature vector corresponding to each of the plurality of machine learning models, comparing each new feature vector corresponding to each machine learning model with the training feature vector corresponding to each of the plurality of machine learning models, and selecting and outputting an inference for the new image generated by the machine learning model for which the new feature vector and the training feature vector are most similar.
1. A method of image analysis, implemented in a computer system comprising a processor, memory accessible by the processor, and computer program instructions stored in the memory and executable by the processor, the method comprising:
extracting a plurality of training feature vectors corresponding to each of a plurality of machine learning models from each validation image from a plurality of machine learning models trained using a plurality of validation images;
extracting from a new image a plurality of new feature vectors corresponding to each of the plurality of machine learning models;
comparing each of the plurality of new feature vectors corresponding to each of the plurality of machine learning models with the plurality of training feature vectors corresponding to each of the plurality of machine learning models; and
selecting a machine learning model and outputting an inference for the new image generated by a selected machine learning model for which a selected new feature vector of the plurality of new feature vectors and a selected training feature vector of the plurality of training feature vectors are most similar.
2. The method of claim 1 , wherein each of the plurality of machine learning models is trained using training images corresponding to different human physical characteristics comprising at least one of age, gender, body mass, and ethnicity.
3. The method of claim 2 , wherein each of the plurality of machine learning models is trained using training images corresponding to different types of findings comprising at least one of tumor, calcification, asymmetry, and stenosis.
4. The method of claim 3 , wherein the comparing is performed using at least one of a cosine similarity function, a root mean square distance, L0, L1, L2 norms, and Locality Sensitive Hashing.
5. The method of claim 4 , further comprising computing a plurality of similarity scores based on a result of the comparing.
6. The method of claim 5 , wherein the plurality of similarity scores are computed using at least one of a highest similarity score, a mean of a fixed number of highest similarity scores, and a proportion of training samples whose similarity score is above a threshold.
7. The method of claim 6 , wherein the output inference is an inference generated by the machine learning model selected based on the computed similarity score.
8. A system for image analysis, the system comprising a processor, memory accessible by the processor, and computer program instructions stored in the memory and executable by the processor to perform:
extracting a plurality of training feature vectors corresponding to each of a plurality of machine learning models from each validation image from a plurality of machine learning models trained using a plurality of validation images;
extracting from a new image a plurality of new feature vectors corresponding to each of the plurality of machine learning models;
comparing each of the plurality of new feature vectors corresponding to each of the plurality of machine learning models with the plurality of training feature vectors corresponding to each of the plurality of machine learning models; and
selecting a machine learning model and outputting an inference for the new image generated by a selected machine learning model for which a selected new feature vector of the plurality of new feature vectors and a selected training feature vector of the plurality of training feature vectors are most similar.
9. The system of claim 8 , wherein each of the plurality of machine learning models is trained using training images corresponding to different human physical characteristics comprising at least one of age, gender, body mass, and ethnicity.
10. The system of claim 9 , wherein each of the plurality of machine learning models is trained using training images corresponding to different types of findings comprising at least one of tumor, calcification, asymmetry, and stenosis.
11. The system of claim 10 , wherein the comparing is performed using a cosine similarity function.
12. The system of claim 11 , further comprising computing a plurality of similarity scores based on a result of the comparing.
13. The system of claim 12 , wherein the plurality of similarity scores are computed using at least one of a highest similarity score, a mean of a fixed number of highest similarity scores, and a proportion of training samples whose similarity score is above a threshold.
14. The system of claim 13 , wherein the output inference is an inference generated by the machine learning model selected based on the computed similarity score.
15. A computer program product for image analysis, the computer program product comprising a non-transitory computer readable storage having program instructions embodied therewith, the program instructions executable by a computer, to cause the computer to perform a method comprising:
extracting a plurality of training feature vectors corresponding to each of a plurality of machine learning models from each validation image from a plurality of machine learning models trained using a plurality of validation images;
extracting from a new image a plurality of new feature vectors corresponding to each of the plurality of machine learning models;
comparing each of the plurality of new feature vectors corresponding to each of the plurality of machine learning models with the plurality of training feature vectors corresponding to each of the plurality of machine learning models; and
selecting a machine learning model and outputting an inference for the new image generated by a selected machine learning model for which a selected new feature vector of the plurality of new feature vectors and a selected training feature vector of the plurality of training feature vectors are most similar.
16. The computer program product of claim 15 , wherein each of the plurality of machine learning models is trained using training images corresponding to different human physical characteristics comprising at least one of age, gender, body mass, and ethnicity.
17. The computer program product of claim 16 , wherein each of the plurality of machine learning models is trained using training images corresponding to different types of findings comprising at least one of tumor, calcification, asymmetry, and stenosis.
18. The computer program product of claim 17 , wherein the comparing is performed using at least one of a cosine similarity function, a root mean square distance, L0, L1, L2 norms, and Locality Sensitive Hashing.
19. The computer program product of claim 18 , further comprising computing a plurality of similarity scores based on a result of the comparing.
20. The computer program product of claim 19 , wherein the plurality of similarity scores are computed using at least one of a highest similarity score, a mean of a fixed number of highest similarity scores, and a proportion of training samples whose similarity score is above a threshold, and the output inference is an inference generated by the machine learning model selected based on the computed similarity score.