Using attention to classify medical images with data imbalance

Paper Title:

Class Attention to Regions of Lesion for Imbalanced Medical Image Recognition

Published on:

19 July 2023

Primary Category:

Computer Vision and Pattern Recognition

Paper Authors:

Jia-Xin Zhuang,

Jiabin Cai,

Jianguo Zhang,

Wei-shi Zheng,

Ruixuan Wang

Bullets

Key Details

•

Proposes CARE method to embed attention into CNN training for imbalanced medical image data

•

Uses bounding boxes and Grad-CAM to drive attention to lesion regions

•

Improves classification accuracy on minority classes in skin and chest X-ray datasets

•

Works with different CNN architectures and can combine with existing imbalance approaches

•

Allows automated lesion localization to reduce annotation needs

Explore the topics in this paper

attention mechanism

classification

data imbalance

medical imaging

rare diseases

AI generated summary

Using attention to classify medical images with data imbalance

This paper proposes a method to handle data imbalance in medical image classification, where some disease classes have many more examples than rare classes. It uses an attention mechanism to help models focus on lesion regions when training on rare disease images. This improves classification of minority classes.