A Deep Learning-Based Automatic Recognition Model for Polycystic Ovary Ultrasound Images

Area of Science:

• Artificial Intelligence in Medical Imaging
• Reproductive Endocrinology
• Diagnostic Accuracy Studies

Background:

• Polycystic ovary syndrome (PCOS) significantly impacts women's health, affecting metabolism, reproduction, and mental well-being.
• Current ultrasound methods for diagnosing polycystic ovarian morphology have variable accuracy.

Purpose of the Study:

• To develop a deep learning model for rapid and accurate PCOS identification from ovarian ultrasound images.
• To evaluate the diagnostic performance of the YOLOv11 framework in detecting PCOS.

Main Methods:

• A prospective diagnostic accuracy study involving 1,751 women with suspected PCOS.
• Data from two centers were used, with patients randomly assigned to training, internal validation, and external validation sets.
• A YOLOv11 deep learning model was constructed for automated recognition of PCOS in ovarian ultrasound images.

Main Results:

• The YOLOv11 model achieved high mean average precision (95.7%-97.8%) and F1 scores (95.0%-96.9%) across training and validation sets.
• Area under the curve values ranged from 0.953 to 0.973, indicating strong diagnostic capability.
• The model's evaluation time (0.1 seconds) was significantly faster than clinicians (5.0 seconds).

Conclusions:

• The YOLOv11-based model demonstrates robust performance in detecting PCOS from ovarian ultrasound images.
• This automated approach can enhance the efficiency and generalizability of ultrasound-based PCOS assessment.
• The model streamlines follicle counting and improves diagnostic efficiency.