Case-Based Decision Support System Personalized Similarity Weights
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing case-based decision support systems, such as CADx systems, assume similarity is uniform across radiologists, but similarity is subjective and varies between individuals, leading to inconsistent retrieval of similar cases.
Innovation Solution
A method that trains on inputs from multiple radiologists to establish a baseline system, allowing users to refine features and weights for personalized similarity distance computation, using genetic algorithms to adapt to individual opinions and experiences.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If a default set of features and weights is used for similarity computation, then the system can operate with simple and fast retrieval, but the retrieval results do not reflect individual radiologists' subjective similarity assessments
Solution Approach 1:
The system transitions from static default weights to dynamic personalized weights that adapt to each radiologist's preferences. The weights are updated iteratively based on user feedback, allowing the system to evolve from a generic state to a personalized state that reflects individual similarity assessments while maintaining operational efficiency.
Solution Approach 2:
The system modifies the weight parameters of features based on user feedback and preferences. By adjusting these parameters iteratively through machine learning algorithms, the system optimizes the balance between retrieval speed and similarity assessment accuracy for each individual radiologist.
2Measurement precision
If the system is personalized for each user, then similarity retrieval accuracy improves, but system complexity increases
Solution Approach 1:
The system performs self-training by automatically learning from user feedback and interactions. Through iterative refinement processes, the system adapts to individual radiologists' preferences without requiring manual configuration, thereby reducing operational complexity while maintaining high personalization levels.
Solution Approach 2:
The system implements a baseline model before personalization, providing immediate functional capability. This preliminary structure allows the system to start with default functionality and progressively refine itself through user feedback, avoiding the need to build complete personalization mechanisms from scratch.
3Measurement precision
If more features are used for similarity computation, then retrieval accuracy improves, but computation time increases
Solution Approach 1:
The system uses a subset of the most relevant features for initial retrieval, then refines results using additional features. This approach achieves high accuracy without requiring all features to be processed simultaneously, thereby reducing computation time while maintaining retrieval quality.
Solution Approach 2:
The system dynamically adjusts feature weights to emphasize the most discriminative features for each query. By optimizing which features contribute most to similarity computation, the system achieves high retrieval accuracy with reduced computational burden compared to uniformly processing all features.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
This invention relates to a method and device for case-based decision support. It proposes that a case-based decision support system is trained on inputs from several radiologists in order to have a "baseline" system, and then the system provides an option to a radiologist to refine the baseline system based on his/her inputs which either refine weights of features for similarity distance computation directly or provide new similarity ground truth clusters. By enabling modifying the similarity distance computation based on user inputs, this invention adapts similarity ground truth to different users with different experience and/or different opinions.