Endoscopic Lesion Inference Using Accuracy-Driven Model Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing technologies for endoscopic examination systems fail to effectively address the variability in the user's requirements for specific issues within a certain technical area that existing technologies have not addressed or effectively solved. These are the challenges or needs the patent application aims to tackle.
Innovation Solution
An image processing device that includes a first acquisition means to acquire a set value of a first index indicating accuracy relating to lesion analysis, a second acquisition means to acquire a predicted value of a second index for each of plural models, and an inference means to make inference regarding the lesion based on the predicted value and plural models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a single lesion detection model is used, then the device complexity is reduced, but the adaptability to different user requirements for accuracy indices deteriorates
Solution Approach 1:
The system dynamically selects among multiple pre-trained lesion detection models based on user-defined accuracy requirements. Instead of using a fixed single model, the system adapts its model selection in real-time according to the specific accuracy indices (e.g., sensitivity, specificity) that the user prioritizes for their examination needs.
Solution Approach 2:
The system changes operational parameters by selecting different models with varying performance characteristics (different accuracy profiles) based on user specifications. Each model has been trained to optimize for different accuracy indices, and the system adjusts which model is deployed by changing the selection criterion according to user input.
2Adaptability or versatility
If multiple lesion detection models are used to satisfy different user requirements, then the adaptability improves, but the device complexity increases
Solution Approach 1:
Multiple lesion detection models are pre-trained in advance, each optimized for different accuracy indices (e.g., one model for high sensitivity, another for high specificity). This preliminary preparation allows the system to quickly select the appropriate pre-trained model based on user requirements without needing to train models on-demand, thereby managing complexity.
Solution Approach 2:
The system introduces an intermediary model selection mechanism that acts as a mediator between user requirements and the actual lesion detection process. This intermediary layer evaluates user-defined accuracy indices and selects the most appropriate pre-trained model, simplifying the interface between diverse user needs and multiple specialized models.
3Measurement precision
If the lesion detection model is optimized for high sensitivity, then the detection rate improves, but the false positive rate increases
Solution Approach 1:
Different models are specialized for different performance characteristics: some models are locally optimized for high sensitivity (detecting all possible lesions), while others are optimized for high specificity (minimizing false positives). The system selects the model with the appropriate local quality (performance profile) based on the user's priorities for the current examination.
Solution Approach 2:
The system dynamically adjusts the trade-off between sensitivity and false positive rate by selecting different models based on user-defined accuracy indices. If the user prioritizes minimizing false positives, the system selects a model optimized for specificity rather than sensitivity, thereby dynamically adjusting the operational characteristics to match user needs.
Data Source
AI summary
The image processing device 1X includes a first acquisition means 30X, a second acquisition means 31X, and an inference means 33X. The first acquisition means 30X acquires a set value of a first index indicating an accuracy relating to a lesion analysis. The second acquisition means 31X acquires, for each of plural models which make inference regarding a lesion, a predicted value of a second index, which is an index of the accuracy other than the first index, on an assumption that the set value of the first index is satisfied. The inference means 33X makes inference regarding the lesion included in an endoscopic image of an examination target, based on the predicted value of the second index and the plural models.