Cardiac Image Classification Using Probabilistic Feedback and User Confirmation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current medical imaging technologies face challenges in accurately evaluating cardiac images due to factors like continuous motion of the heart, insufficient resolution, and low contrast in tissue structure, leading to uncertain diagnoses and disagreements among physicians.
Innovation Solution
A method using a trained machine learning data processing model to classify image features in cardiac images by evaluating probabilities for predetermined classes, providing a certainty level, and requesting user input when uncertainty exists, thereby improving decision-making and reducing incorrect diagnoses.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If image recognition based on artificial intelligence is used, then evaluation accuracy is improved, but certainty and agreement among different evaluations is not fully resolved
Solution Approach 1:
The system implements feedback by evaluating probability data for multiple classes, identifying candidate classes with significant probabilities, and determining certainty levels. When certainty is insufficient, the system requests user input to confirm or disaffirm classifications, creating a feedback loop that improves both accuracy and reliability.
Solution Approach 2:
The system changes the parameter of evaluation from binary classification to probabilistic multi-class classification. By providing probability data for each class and determining certainty levels, the system transforms the evaluation process to simultaneously improve accuracy and provide measurable reliability metrics.
2Adaptability or versatility
If multiple candidate classes are identified with significant probability, then classification completeness is improved, but decision complexity increases
Solution Approach 1:
The system applies partial action by identifying only candidate classes with significant probabilities rather than evaluating all possible classes. This selective approach maintains classification completeness for relevant cases while reducing decision complexity by filtering out unlikely candidates.
Solution Approach 2:
The system substitutes manual evaluation of multiple possibilities with an automated machine learning model that processes probability data and identifies candidate classes. This replacement reduces the complexity burden on users while maintaining comprehensive classification.
3Measurement precision
If user input is requested for confirmation, then diagnostic accuracy is improved, but evaluation time increases
Solution Approach 1:
The system applies partial action by requesting user input only when certainty levels are insufficient, not for all cases. When the machine learning model is highly confident, the classification is accepted automatically, reducing evaluation time. User input is requested only partially, when needed for low-certainty cases.
Solution Approach 2:
The system performs self-service by automatically evaluating probability data and determining certainty levels. The machine learning model independently processes classifications and only involves users when necessary, allowing the system to handle most evaluations autonomously and efficiently.
Data Source
Figure 1~2
Figure 3~4
Figure 5
AI summary
The invention is directed at a classifying image features to determine a cardiac characteristic, using a data processing model trained to evaluate a probability that the image feature matches a certain class. According to the method, the data processing model receives cardiac image data, and obtains the region of interest. Probability data per class is collected from the data processing model and evaluated to identify candidate classes likely of matching the image feature. Dependent on a certainty level a next step is performed. If there's one candidate class and the certainty exceeds a threshold, the candidate class is associated with the image feature, or if the certainty is below threshold user input is requested. If there is more candidate classes, user input is also requested.