Detecting and localizing cervical lesions in colposcopic images with deep semantic feature mining

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Cervical cancer screening relies on colposcopy, which can be subjective.
• Accurate detection and localization of cervical lesions are crucial for effective treatment.

Purpose of the Study:

• To assess the feasibility of AI models for detecting and localizing cervical lesions.
• To leverage deep semantic features from colposcopic images for improved diagnostic accuracy.

Main Methods:

• A segmentation-based deep learning architecture with a deep decoding network was employed.
• A two-stage decision model combined image segmentation and classification for pathological changes.
• Transfer learning and an attention mechanism were utilized for feature extraction and precise lesion segmentation.

Main Results:

• The AI approach outperformed traditional segmentation or classification methods.
• A fully automated pixel-based cervical lesion segmentation model was developed.
• The model achieved high sensitivity (96.38%), specificity (95.84%), precision (97.56%), and F1 score (96.96%).

Conclusions:

• The AI model shows promise for identifying normal and cancerous cervical lesions, especially in area segmentation.
• The AI's accuracy in guiding biopsy site selection and treatment localization offers significant support to clinicians.
• This technology can enhance disease assessment and management in cervical cancer screening.