Few-Shot Learning Geometric Ensemble for Multi-label Classification of Chest X-Rays

Area of Science:

• Medical Imaging
• Machine Learning
• Computer-Aided Diagnosis

Background:

• Classifying rare cardiothoracic diseases from chest X-rays is challenging due to limited labeled data.
• Existing machine learning models struggle with multi-label classification of infrequent conditions.

Purpose of the Study:

• To develop a machine learning model capable of identifying uncommon cardiothoracic diseases using chest X-ray images.
• To address the data scarcity issue in multi-label classification of rare diseases through few-shot learning (FSL).

Main Methods:

• Proposed a multi-label few-shot learning (FSL) framework incorporating neighborhood component analysis loss.
• Implemented distribution calibration for generating additional training samples.
• Developed a geometric DeepVoro Multi-label ensemble by combining Voronoi diagrams from multi-label schemes.

Main Results:

• Demonstrated improved performance in multi-label few-shot classification tasks.
• The DeepVoro Multi-label ensemble effectively leveraged information from common diseases to aid in identifying rare conditions.
• Achieved enhanced accuracy in classifying cardiothoracic diseases with limited training data.

Conclusions:

• The proposed multi-label FSL approach with the DeepVoro ensemble shows significant promise for diagnosing rare cardiothoracic diseases.
• This method offers a viable solution for improving diagnostic accuracy in scenarios with insufficient labeled data for rare conditions.
• The publicly available code facilitates further research and application in medical image analysis.