CADx Gatekeeper Circuit for Out-of-Distribution Image Screening
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
State-of-the-art CADx devices struggle with out-of-distribution (OOD) data, leading to unreliable predictions and user trust issues due to the inability to effectively detect and manage OOD inputs, especially in clinical settings.
Innovation Solution
A CADx system equipped with a gatekeeper circuit that includes an input similarity estimator and rejection explanation generator to identify and alert users when OOD data is encountered, preventing model evaluation and providing explanations for rejections.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a machine learning model is trained on a specific dataset to predict disease risk, then it can provide disease risk scores for in-distribution data, but it produces unreliable predictions and misclassifications when encountering out-of-distribution data that was not represented in the training dataset
Solution Approach 1:
The gatekeeper is trained in advance on a diverse set of out-of-distribution examples to learn what characteristics indicate unreliable predictions. Before the main model makes predictions, the gatekeeper pre-assesses the input data to determine if it falls within the reliable operating range, preventing misclassifications before they occur.
Solution Approach 2:
The gatekeeper acts as an intermediary layer between the input data and the main prediction model. It filters and gates the input data flow, allowing only in-distribution data to proceed to the model while blocking out-of-distribution data, thus protecting the system from unreliable predictions.
2Productivity
If the CADx system processes all incoming medical images without verification, then productivity is maintained, but misclassifications occur on out-of-distribution data leading to loss of user trust
Solution Approach 1:
The gatekeeper performs a preliminary assessment of each input image before it reaches the main processing pipeline. This early filtering action identifies out-of-distribution images that would likely cause misclassifications, allowing the system to handle them appropriately while maintaining efficient processing of valid images.
Solution Approach 2:
The gatekeeper extracts and identifies the subset of out-of-distribution images from the overall input stream. By separating these problematic cases from the normal workflow, the system can maintain high productivity for in-distribution images while preventing misclassifications on extracted out-of-distribution cases.
3Ease of operation
If the system provides disease risk scores for all inputs including OOD data, then ease of operation is maintained, but the scores become unreliable leading to potential harmful misclassifications
Solution Approach 1:
The gatekeeper performs a preliminary check to prevent the harmful effect of misclassification before it can occur. By identifying out-of-distribution inputs in advance, the system can take preventive action such as rejecting the input or flagging it for manual review, thereby avoiding the generation of unreliable risk scores that could harm patient care.
Solution Approach 2:
The gatekeeper converts the potential harm of processing out-of-distribution data into a benefit by using the gatekeeping mechanism itself as a useful feature. Instead of silently producing unreliable results, the system actively identifies and flags problematic inputs, turning a vulnerability into a quality assurance mechanism that enhances overall system reliability.
Data Source
Figure 1
Figure 2A
Figure 2B
AI summary
A CADx system for analysing medical images and determining if the images are acceptable for analysis, by determining if the images contain out-of-distribution input data is described. The CADx system comprises: an input circuit for receiving at least one medical image; a gatekeeper circuit for determining if the at least one received medical image contains out-of-distribution input data and so does not meet the requirements of the CADx system for acceptable images; and an output circuit to produce an output that is either a determination that the at least one received medical image contains out-of-distribution data and is not suitable for analysis by the CADx system, or a determination that the medical image is acceptable.