A lightweight xAI approach to cervical cancer classification

Area of Science:

• Oncology
• Medical Imaging
• Artificial Intelligence

Background:

• Cervical cancer, primarily caused by human papillomavirus (HPV), disproportionately affects regions with limited healthcare access.
• Liquid cytology and AI-powered classifiers offer potential for mass screening and diagnostic assistance.
• Explainable AI is crucial for pathologist verification and building trust in AI diagnostic systems.

Purpose of the Study:

• To develop and optimize a custom deep-learning classifier for cervical cancer detection using liquid cytology images.
• To achieve high accuracy in classifying multiple severity grades of cervical cancer.
• To ensure the AI system provides explainable insights for clinical decision-making.

Main Methods:

• A 4-phase optimization process was employed to create a deep-learning classifier.
• The classifier was trained on liquid cytology images to distinguish between 4 severity classes of cervical cancer.
• Convolutional neural networks and explainable AI techniques were utilized.

Main Results:

• The optimized classifier achieved over 97% accuracy for 4 severity classes and 100% accuracy for 2 classes.
• The system demonstrated rapid execution times, completing classification and report generation in under 1 second.
• The proposed classifier outperformed previous methods in accuracy while maintaining a lower computational cost.

Conclusions:

• The developed deep-learning classifier shows high efficacy and efficiency in diagnosing cervical cancer from liquid cytology images.
• Explainable AI features enhance the system's utility by providing transparency for pathologists.
• This AI approach holds promise for improving cervical cancer screening, particularly in resource-limited settings.