AMD Level Classification with Macula-Aware Fundus Image Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing artificial intelligence models for determining the severity level of age-related macular degeneration (AMD) often provide inaccurate results due to differing determination criteria compared to human doctors, who focus on the macula in fundus images, while AI models consider the entire eyeball area.
Innovation Solution
A classification device and method that utilizes an object detection model to identify the macula in a fundus image, generating a bounding box, and employs two classification models - one for the macula area and another for the entire image - based on an intersection over union threshold to align with human diagnostic criteria.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Area of stationary object
If the AI model uses the entire fundus image for AMD classification, then the coverage area is maximized, but the diagnostic accuracy decreases due to mismatched determination criteria with human doctors
Solution Approach 1:
The patent segments the fundus image processing into two distinct pathways: one for the macula region and one for the entire fundus image. The system dynamically selects which segmentation to use based on the IOU threshold, thereby resolving the contradiction between coverage area and diagnostic accuracy by applying the appropriate segmentation level for each case
Solution Approach 2:
The patent introduces dynamic adaptability by using the IOU threshold as a decision criterion to switch between two classification modes. When the macula position is reasonable (high IOU), the system dynamically selects macula-based classification; when the position is unreasonable (low IOU), it switches to entire-image classification, making the system flexible rather than fixed
2Measurement precision
If the AI model focuses on the macula area to match human diagnostic criteria, then the diagnostic accuracy improves, but the system becomes vulnerable when macula position is unreasonable
Solution Approach 1:
The patent performs a preliminary check by calculating the IOU threshold between the detected macula bounding box and the predetermined area before proceeding to classification. This preliminary action assesses whether the macula position is reasonable, and based on this assessment, the system prepares and executes the appropriate classification pathway, preventing errors from unreasonable macula detection
Solution Approach 2:
The system uses the IOU threshold calculation as a feedback mechanism to evaluate the quality of macula detection. This feedback determines whether to proceed with macula-based classification or switch to entire-image classification, creating a self-correcting system that adapts to detection quality and maintains reliability under varying conditions
Data Source
AI summary
A classification method and a classification device for classifying a level of an age-related macular degeneration are provided. The classification method includes the following. An object detection model and a first classification model are pre-stored. A fundus image is obtained. A bounding box is generated in the fundus image according to a macula in the fundus image detected by the object detection model. An intersection over union between a predetermined area and the bounding box in the fundus image is calculated. A classification of the fundus image is generated according to the first classification model in response to the intersection over union being greater than a threshold.