Attention-Based Dual-Path Deep Learning for Blood Cell Image Classification Using ConvNeXt and Swin Transformer.

Area of Science:

• Medical Image Analysis
• Hematology
• Deep Learning

Background:

• Accurate blood cell classification is vital for diagnosing hematological disorders.
• Traditional methods are labor-intensive and can be inconsistent.
• Automated analysis is needed to improve diagnostic efficiency.

Purpose of the Study:

• To develop a highly accurate automated blood cell classification system.
• To leverage deep learning for enhanced feature extraction and context integration.
• To improve the quality of medical images for better cellular differentiation.

Main Methods:

• A dual-path deep learning architecture combining ConvNeXt and Swin Transformer networks.
• Implementation of a Multi-scale Preprocessing Module (MPM) for image enhancement.
• Training and validation on a dataset of 17,092 blood cell images.

Main Results:

• Achieved an unprecedented classification accuracy of 99.98%.
• Demonstrated superior performance compared to existing blood cell analysis methods.
• The MPM significantly improved the visibility and differentiation of cellular features.

Conclusions:

• The proposed dual-path deep learning model offers a highly accurate solution for blood cell classification.
• This technology has the potential to enhance clinical diagnostic workflows and patient outcomes.
• Automated, precise blood cell analysis can reduce medical professional workload and improve diagnostic speed.