SFPL: Sample-specific fine-grained prototype learning for imbalanced medical image classification.

Area of Science:

• Medical Image Analysis
• Machine Learning
• Computer Vision

Background:

• Imbalanced classification poses significant challenges in medical image analysis.
• Existing methods often overlook intra-class heterogeneity and individual sample variations.
• Addressing these limitations is crucial for improving diagnostic accuracy.

Purpose of the Study:

• To propose a novel sample-specific fine-grained prototype learning (SFPL) method.
• To enhance classification models by capturing individual sample characteristics.
• To improve performance on imbalanced medical image datasets.

Main Methods:

• SFPL learns fine-grained representations of the majority class using multiple prototypes.
• Prototypes are updated via a mixture weighting strategy and a uniform loss function.
• A selective attention aggregation module links prototypes to sample-specific cosine classifiers.

Main Results:

• SFPL demonstrated superior performance compared to state-of-the-art methods across three tasks.
• Performance gains increased with higher imbalance ratios and reduced training data.
• The method effectively handles intra-class heterogeneity and sample individuality.

Conclusions:

• SFPL offers a robust solution for imbalanced classification in medical imaging.
• The approach improves model adaptability to individual sample characteristics.
• SFPL shows significant potential for advancing medical image analysis applications.