Systems and methods for processing electronic images using uncertainty estimation
A method for processing electronic images using uncertainty estimation may be used to determine whether to use an artificial intelligence (AI) assisted prediction. The method may include receiving one or more electronic images associated with a pathology specimen and providing the one or more electronic images to a machine learning model. The machine learning model may perform operations including determining a certainty level corresponding to a certainty that a predetermined AI system will provide an accurate prediction, determining whether the certainty level equals or exceeds a predetermined confidence threshold, and, upon determining that the certainty level does not equal or exceed a predetermined confidence threshold, determining to not use the predetermined AI system.
1 . A computer-implemented method of determining whether to use an artificial intelligence (AI) assisted prediction using an AI system, the method comprising:
receiving, by one or more processors of the AI system, one or more electronic slide images associated with a pathology specimen; and
providing, by the one or more processors, the one or more electronic slide images to a machine learning model, the machine learning model trained to:
classify the one or more electronic slide images to determine a certainty level corresponding to a certainty that a predetermined AI system will provide an accurate prediction for the pathology specimen;
determine whether the certainty level equals or exceeds a predetermined confidence threshold; and
upon determining that the certainty level does not equal or exceed a predetermined confidence threshold, output a determination to not use the predetermined AI system for the AI assisted prediction for the pathology specimen.
2 . The method of claim 1 , wherein running the machine learning model to determine the certainty level includes identifying a subpopulation of the pathology specimen.
3 . The method of claim 1 , wherein running the machine learning model to determine whether the certainty level equals or exceeds a predetermined confidence threshold includes running a trained classifier to determine a classified label, the classified label indicating at least one of a subpopulation or whether the pathology specimen is outside of one or more predetermined subpopulations.
4 . The method of claim 3 , wherein running the classifier to determine the classified label includes:
partitioning the received electronic slide images into a plurality of foreground tiles;
extracting a vector of features from each foreground tile; and
running each vector through the trained classifier to determine the classified label.
5 . The method of claim 1 , wherein running the machine learning model to determine whether the certainty level equals or exceeds a predetermined confidence threshold includes detecting one or more features in the received electronic slide images and determining a consistency of the detected one or more features.
6 . The method of claim 5 , wherein detecting the one or more features in the received electronic slide images and determining the consistency includes:
partitioning the received electronic slide images into a plurality of foreground tiles;
extracting a vector of the detected one or more features from each foreground tile;
determining local feature information for each foreground tile; and
analyzing, using statistical analysis, the determined local feature information for all foreground tiles to determine the consistency.
7 . The method of claim 5 , further comprising outputting a visualization of the detected one or more features and/or the determined consistency.
8 . The method of claim 1 , wherein running the machine learning model to determine the certainty level includes:
partitioning the received electronic slide images into a plurality of foreground tiles;
extracting a vector of features from each foreground tile; and
running each extracted vector through the machine learning model to determine at least one of: (i) a subpopulation for the pathology specimen, or (ii) an unknown label indicating that the pathology specimen is outside one or more predetermined subpopulations.
9 . The method of claim 1 , wherein determining to not use the predetermined AI system includes:
detecting that the pathology specimen is outside one or more predetermined subpopulations.
10 . The method of claim 9 , further comprising determining that an alternative method should be used, wherein the predetermined AI system was trained using results from the alternative method as ground truth.
11 . The method of claim 1 , further comprising determining the predetermined confidence threshold based on a user input.
12 . The method of claim 1 , wherein the certainty level is based on a comparison of a performance by the predetermined AI system with a comparison of an alternative method.
13 . The method of claim 1 , wherein the predetermined confidence threshold is based on a certainty level associated with an alternative method.
14 . The method of claim 1 , wherein the predetermined AI system is configured to detect a presence and/or an intensity of HER2 expression.
15 . The method of claim 1 , wherein the predetermined AI system is configured to predict a response to a polymerase chain reaction (PCR) test.
16 . The method of claim 1 , further comprising outputting a recommendation for an alternative method, wherein the alternative method is configured to make a same type of prediction as the predetermined AI system.
17 . The method of claim 1 , further comprising storing or removing the received electronic slide images, based on whether the pathology specimen belongs to one or more predetermined subpopulations, to create a data set based on one or more electronic images for which the certainty level equals or exceeds the predetermined confidence threshold.
18 . A system for determining whether to use an artificial intelligence (AI) assisted prediction, the system comprising:
at least one memory storing instructions; and
at least one processor configured to execute the instructions to perform operations comprising:
receiving, by the at least one processor, one or more electronic slide images associated with a pathology specimen; and
providing, by the at least one processor, the one or more electronic slide images to a machine learning model, the machine learning model trained to:
classify the one or more electronic slide images to determine a certainty level corresponding to a certainty that a predetermined AI system will provide an accurate prediction for the pathology specimen;
determine whether the certainty level equals or exceeds a predetermined confidence threshold; and
upon determining that the certainty level does not equal or exceed a predetermined confidence threshold, output a determination to not use the predetermined AI system for the AI assisted prediction for the pathology specimen.
19 . The system of claim 18 , wherein running the machine learning model to determine the certainty level includes identifying a subpopulation of the pathology specimen.
20 . A non-transitory computer-readable medium storing instructions that, when executed by a processor, perform a method for determining whether to use an artificial intelligence (AI) assisted prediction, the method comprising:
receiving, by the processor, one or more electronic slide images associated with a pathology specimen; and
providing, by the processor, the one or more electronic slide images to a machine learning model, the machine learning model trained to:
classify the one or more electronic slide images to determine a certainty level corresponding to a certainty that a predetermined AI system will provide an accurate prediction for the pathology specimen;
determine whether the certainty level equals or exceeds a predetermined confidence threshold; and
upon determining that the certainty level does not equal or exceed a predetermined confidence threshold, output a determination to not use the predetermined AI system for the AI assisted prediction for the pathology specimen.