Red Blood Cell Classification Based on Attention Residual Feature Pyramid Network

Area of Science:

• Medical imaging analysis
• Artificial intelligence in healthcare
• Hematology

Background:

• Red blood cell abnormalities are linked to various diseases, including tumors and internal medicine conditions.
• Accurate red blood cell classification is crucial for diagnosing these abnormalities.
• Manual classification by doctors is labor-intensive and prone to subjectivity.

Purpose of the Study:

• To develop an automated system for classifying 14 types of red blood cells.
• To assist medical professionals in diagnosing diseases related to red blood cell abnormalities.
• To improve the efficiency and accuracy of red blood cell classification.

Main Methods:

• Proposed an Attention-based Residual Feature Pyramid Network (ARFPN) for direct red blood cell image classification.
• Integrated spatial and channel attention mechanisms with residual units to enhance feature extraction.
• Utilized the RoI align method to preserve spatial information and improve classification accuracy.

Main Results:

• The Channel Attention Residual Feature Pyramid Network (C-ARFPN) achieved a mean Average Precision (mAP) of 86%.
• The Channel and Spatial Attention Residual Feature Pyramid Network (CS-ARFPN) achieved a higher mAP of 86.9%.
• The proposed models demonstrated effectiveness in classifying multiple red blood cell types.

Conclusions:

• The developed ARFPN models offer a more efficient and accurate alternative to manual red blood cell classification.
• The AI-driven approach can significantly reduce diagnostic time for medical professionals.
• This technology has the potential to enhance diagnostic efficiency in clinical settings.