YOLOv5-FPN: A Robust Framework for Multi-Sized Cell Counting in Fluorescence Images

Area of Science:

• Biomedical research
• Cellular dynamics
• Disease progression analysis

Background:

• Manual cell counting and threshold-based segmentation are time-consuming and error-prone.
• Deep learning offers automation but existing segmentation-based methods demand extensive data and computational power.

Purpose of the Study:

• To develop an efficient and accurate method for detecting and counting multiple-size cells in fluorescence microscopy images.
• To overcome limitations of existing segmentation-based deep learning techniques.

Main Methods:

• Utilized You Only Look Once version 5 (YOLOv5) integrated with a feature pyramid network (FPN).
• Employed an object detection approach, eliminating the need for pixel-level segmentation.
• Tested on publicly available fluorescence microscopy datasets.

Main Results:

• Achieved an average precision of 0.8 for cell detection and counting.
• Demonstrated a processing time of 43.9 ms per image, indicating high computational efficiency.
• Outperformed state-of-the-art segmentation-based methods in accuracy and efficiency.

Conclusions:

• The proposed YOLOv5-based method provides a more efficient and accurate solution for cell counting in fluorescence microscopy.
• This approach requires less labeled data and computational resources compared to traditional and segmentation-based deep learning methods.
• Offers a valuable tool for biomedical research analyzing cellular dynamics and disease progression.