IEGS-BoT: An Integrated Detection-Tracking Framework for Cellular Dynamics Analysis in Medical Imaging.

Area of Science:

• Biomedical image analysis
• Microscopic video analysis
• Cellular imaging

Background:

• Cell detection and tracking are crucial for biomedical image analysis, aiding clinical diagnosis and treatment.
• Challenges in microscopic videos include ambiguous boundaries and complex backgrounds, causing missed or false detections and tracking loss.
• Existing methods struggle with object loss and identity confusion in dynamic cellular environments.

Purpose of the Study:

• To develop an enhanced multiple object tracking algorithm, IEGS-BoT, for accurate cell detection and tracking in microscopic sequences.
• To address limitations of current methods in handling complex backgrounds and maintaining object identity.
• To improve the reliability of cell tracking for clinical applications.

Main Methods:

• Developed the IEGS-YOLO detector incorporating the iEMA module for enhanced feature fusion and GSConv in the neck for reduced complexity.
• Integrated the BoT-SORT tracker with camera motion compensation and Kalman filter for precise bounding box localization.
• Evaluated the IEGS-BoT algorithm on the CTMC dataset for comprehensive performance analysis.

Main Results:

• IEGS-YOLO achieved superior detection performance with map50 of 73.2% and map50-95 of 32.6%, outperforming YOLO11n.
• IEGS-BoT demonstrated significant improvements in tracking metrics: MOTA (53.97%), HOTA (51.30%), and IDF1 (67.52%).
• Reduced ID switches from 1170 to 894 compared to the base BoT-SORT, indicating enhanced identity preservation.

Conclusions:

• The proposed IEGS-BoT algorithm effectively overcomes challenges in cell detection and tracking, including object loss and identity switching.
• This method provides a more robust and accurate solution for analyzing microscopic video sequences in biomedical research.
• IEGS-BoT offers a promising tool for advancing medical image analysis and clinical decision-making.