AMD Classification Model Training with Stage Penalty Weights
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing AI models for classifying age-related macular degeneration severity based on fundus images often suffer from overfitting due to varying annotations by medical personnel, leading to reduced classification accuracy.
Innovation Solution
An electronic device and method that calculates a loss function vector using a machine learning algorithm, incorporating penalty weights based on stage differences to update the loss function values, thereby preventing overfitting and improving model accuracy by adjusting the focus on easily confused classifications.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If training data annotated by different medical personnel is used to train the AI model, then the model can learn from diverse clinical perspectives, but the classification accuracy deteriorates due to overfitting from varying macular area identifications
Solution Approach 1:
The patent changes the parameter of loss function weighting by introducing penalty weights that are proportional to the stage differences between AMD classifications. This modifies the training objective to account for the varying difficulty of distinguishing between different stages, thereby resolving the overfitting issue while maintaining adaptability to diverse clinical annotations
Solution Approach 2:
The patent performs preliminary action by pre-calculating the penalty weights based on stage differences before the actual model training. This allows the training process to incorporate the difficulty information in advance, preventing overfitting from the outset while still utilizing the diverse clinical perspectives in the training data
2Measurement precision
If the AI model focuses on distinguishing between easily confused stages (e.g., first and second stage), then the classification precision for those stages improves, but the model becomes overfitted to specific patterns, reducing overall generalization accuracy
Solution Approach 1:
The patent applies parameter changes by modifying the loss function to include penalty weights that are proportional to the absolute difference in stages. This creates a balanced training objective that prevents the model from over-focusing on easily confused stages while maintaining adequate attention to all stage distinctions, thereby improving generalization
Solution Approach 2:
The patent implements feedback by using the stage difference information to dynamically adjust the penalty weights during training. The model receives feedback about which stage distinctions are more difficult, and the loss function is adjusted accordingly to prevent overfitting to specific easy-to-distinguish patterns
Data Source
AI summary
An electronic device and a method of training a classification model for age-related macular degeneration (AMD) are provided. The method includes the following steps. Training data is obtained. A loss function vector corresponding to the training data is calculated based on a machine learning algorithm, in which the loss function vector includes a first loss function value corresponding to a first stage of AMD and a second loss function value corresponding to a second stage of AMD. A first penalty weight is generated according to a stage difference between the first stage and the second stage. The first loss function value is updated according to the second loss function value and the first penalty weight, so as to generate an updated loss function vector. The classification model is trained according to the updated loss function vector.