Cervical cell extraction network based on optimized yolo.

Area of Science:

• Medical Imaging
• Computational Biology
• Cancer Research

Background:

• Early cervical cancer screening is crucial for prevention.
• Segmenting highly overlapping cells in microscopic images presents a significant challenge.
• Identifying individual abnormal cells amidst dense clusters is difficult.

Purpose of the Study:

• To propose an effective object detection algorithm, Cell_yolo, for accurate segmentation of overlapping cervical cells.
• To improve the identification of single cells from overlapping clusters in cervical cell images.
• To enhance the accuracy and efficiency of cervical cancer screening through improved cell segmentation.

Main Methods:

• Developed Cell_yolo, an object detection algorithm featuring a simplified network structure and improved maximum pooling.
• Implemented a novel non-maximum suppression method based on center distance to handle overlapping cells.
• Enhanced the loss function by incorporating the focal loss function to address sample imbalance.

Main Results:

• Cell_yolo demonstrated superior performance in segmenting overlapping cervical cells compared to YOLOv4 and Faster_RCNN.
• The algorithm achieved high detection accuracy with low computational complexity.
• The proposed methods effectively preserved image information and prevented misclassification of overlapping cells.

Conclusions:

• Cell_yolo offers an effective and accurate solution for segmenting overlapping cervical cells.
• The algorithm shows significant potential for improving automated analysis in cervical cancer screening.
• This work contributes to advancing computational methods in medical image analysis for cancer detection.