Network Visualization and Pyramidal Feature Comparison for Ablative Treatability Classification Using Digitized...

Area of Science:

• Gynecologic Oncology
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Uterine cervical cancer is a significant global health concern for women.
• Cervical tissue ablation effectively treats precancerous lesions.
• Automated Visual Examination (AVE) shows promise in identifying cervical precancer from images.

Purpose of the Study:

• To develop a machine learning algorithm for determining cervical ablation eligibility based on visual characteristics.
• To enhance diagnostic accuracy and treatment decision-making for cervical precancer.

Main Methods:

• A novel deep learning object detection architecture, based on RetinaNet, was developed.
• The algorithm utilizes upsampling and concatenation of pretrained RetinaNet layers.
• Classification results were visualized using Class-selective Relevance Maps (CRM) and Class Activation Maps (CAM).

Main Results:

• The customized deep learning architecture outperformed the baseline RetinaNet in treatability classification.
• The algorithm provided insights into significant features and regions for treatment recommendations.
• Image quality degradation was shown to negatively impact classification accuracy.

Conclusions:

• The developed machine learning algorithm accurately predicts cervical ablation eligibility from images.
• The AI model offers explainable insights into treatment recommendations, aiding clinical decisions.
• High-quality cervical images are crucial for reliable AI-driven diagnostic assessments.