A Few-shot learning approach for Monkeypox recognition from a cross-domain perspective.

Area of Science:

• Medical Imaging
• Computer Vision
• Machine Learning

Background:

• Monkeypox is an emerging zoonotic infectious disease with increasing global concern.
• Accurate and rapid diagnosis is crucial for disease control.
• Traditional deep learning models require large annotated datasets, which are scarce for rare diseases like monkeypox.

Purpose of the Study:

• To develop a few-shot learning approach for effective human monkeypox recognition in images.
• To address the challenge of limited labeled data in medical image analysis.
• To improve the efficiency and accuracy of computer-aided diagnosis for infectious skin diseases.

Main Methods:

• A novel few-shot learning framework utilizing a normal backbone and auxiliary backbones.
• Co-training with Self-supervised Learning and Cross-domain Adaptation techniques.
• A power transform layer for unifying features across different domains.

Main Results:

• The proposed method achieved superior performance compared to mainstream few-shot learning algorithms.
• Demonstrated effectiveness in a three-way few-shot classification task involving chickenpox, measles, and human monkeypox.
• Successfully recognized human monkeypox with a minimal number of training samples.

Conclusions:

• The developed few-shot learning framework offers a promising solution for diagnosing diseases with limited data.
• This approach can significantly aid in the early detection and management of emerging infectious diseases like monkeypox.
• Highlights the potential of self-supervised and cross-domain learning in medical image recognition.