MDE-YOLO: Edge-Aware Multi-Scale Fusion Lightweight High-Precision Model for Cervical Cell Detection

Area of Science:

• Medical Imaging
• Computer Vision
• Oncology

Background:

• Cervical cancer is a leading cause of cancer death in women, making early diagnosis critical.
• Traditional object detection methods struggle with edge information and detail performance in medical images.
• Accurate detection of cervical cells is essential for timely diagnosis and treatment planning.

Purpose of the Study:

• To propose an improved object detection method for precise cervical cell identification.
• To enhance the detection performance of cervical cell targets in cervical images using deep learning.
• To reduce false detection and missed detection rates in cervical cell analysis.

Main Methods:

• Developed MDE-YOLO, an enhanced object detection model based on YOLOv11n.
• Introduced a multiscale feature fusion pyramid network (MSFFPN) for improved scale detection.
• Incorporated a shared detail-enhanced detection head (SDEDH) for superior detail processing.
• Implemented an edge-enhanced fusion stem (EEFStem) module to boost cell contour recognition.

Main Results:

• MDE-YOLO achieved an mAP50 of 80.1% on the Cervix Cell Detection dataset, a 2.6% improvement over YOLOv11n.
• MDE-YOLO demonstrated superior performance on a challenging dataset with higher noise and detection difficulty.
• The model has 24.5% fewer parameters than YOLOv11n, indicating improved efficiency.
• MDE-YOLO showed significant improvements in mAP50 (3.0%) and mAP50-95 (1.6%) on the Comparison Detector Dataset.

Conclusions:

• MDE-YOLO offers enhanced accuracy and robustness for cervical cell detection compared to standard YOLO models.
• The proposed model effectively reduces false and missed detection rates in cervical cell images.
• MDE-YOLO shows promising application prospects for improving early diagnosis of cervical cancer.