Deep self-supervised transformation learning for leukocyte classification.

Area of Science:

• Medical Image Analysis
• Deep Learning
• Computer Vision

Background:

• Annotation scarcity is a major challenge for deep learning in medical image analysis.
• Self-supervised learning (SSL) offers a solution by leveraging unlabeled data for feature extraction.

Purpose of the Study:

• To propose a simple and effective SSL method for leukocyte classification.
• To address the challenge of limited annotated data in medical imaging.

Main Methods:

• Utilized a convolutional neural network backbone for feature extraction from transformed leukocyte images.
• Implemented a pretext task of self-supervised transformation recognition.
• Avoided large batches of negative sampling and specialized architectures.

Main Results:

• The proposed method demonstrated superior performance compared to five typical self-supervised image classification baselines.
• Achieved better results in linear evaluation, domain transfer, and finetuning protocols.
• Effectively learned generalizable representations for different leukocyte types and datasets.

Conclusions:

• The developed self-supervised learning method is effective for leukocyte classification.
• The approach successfully extracts useful features from unlabeled medical images.
• Validated through systematic study of transformation compositions and comparative experiments.