AMD Classification Model Training With Stage-Weighted Loss

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing AI models for classifying age-related macular degeneration stages suffer from overfitting due to varying medical personnel annotations, leading to inaccurate stage classification, particularly misjudging severe stages as less severe, which can delay timely medical referrals.

Innovation Solution

A method and electronic device that updates loss function values using penalty weights based on stage differences and group penalties to train a classification model, reducing overfitting and improving accuracy in distinguishing between stages requiring urgent referrals.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Ease of manufacture

If training data annotated by different medical personnel is used to train the AI model, then the model can be trained with available data, but the classification accuracy deteriorates due to overfitting caused by varying annotations

Engineering Contradiction:
Improveease of training data preparationVSAvoidclassification accuracy
Core Design Contradiction:
Ease of manufactureVSMeasurement precision

Solution Approach 1:

The patent applies parameter changes by modifying the loss function parameters to incorporate uncertainty weights. Different loss function values are assigned to different classification targets based on the severity of potential misclassification, thereby changing the optimization parameters of the training process to account for annotation variability and prevent overfitting

Inventive Principle:
Principle #35Parameter changes

Solution Approach 2:

The patent implements local quality by applying different loss function values to different classification targets. Severe stages (third and fourth stages) are assigned higher loss function values compared to less severe stages, creating a non-uniform quality distribution across classification targets that reflects the clinical importance of each stage

Inventive Principle:
Principle #3Local quality

2Measurement precision

If the classification model misclassifies severe stages as less severe stages, then the model may show higher overall accuracy, but the reliability of medical referral decisions deteriorates

Engineering Contradiction:
Improveoverall classification accuracyVSAvoidreliability of referral decisions
Core Design Contradiction:
Measurement precisionVSReliability

Solution Approach 1:

The patent changes the parameter weighting in the loss function to reflect the clinical consequences of misclassification. By assigning higher loss values to misclassification of severe stages, the optimization process is shifted to prioritize correct identification of severe cases over overall accuracy metrics

Inventive Principle:
Principle #35Parameter changes

Solution Approach 2:

The patent converts the harmful effect of annotation variability and potential misclassification into a beneficial training signal. The uncertainty and varying annotations from different medical personnel are transformed into weighted loss functions that guide the model to be more conservative in classifying severe stages, thereby improving referral decision reliability

Inventive Principle:
Principle #22Blessing in disguise (Convert harm into benefit)

Data Source

PatentEP4099279B1Electronic device and method of training classification model for age-related macular degeneration
Publication Date: 2025.11.26 ACER BEING HEALTH INC
  • EP4099279B1 patent drawingFigure 1~2
  • EP4099279B1 patent drawingFigure 3
  • EP4099279B1 patent drawing

AI summary

An electronic device (100) 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 (23) corresponding to the training data is calculated based on a machine learning algorithm, in which the loss function vector (23) includes a first loss function value corresponding to a first classification of AMD and a second loss function value corresponding to a second classification of AMD, the first classification corresponds to a first group, and the second classification corresponds to one of the first group and a second group. The first loss function value is updated according to the second loss function value and a group penalty weight in response to the second classification corresponding to the second group to generate an updated loss function vector (25). The classification model is trained according to the updated loss function vector (25).