Deep learning classification of cervical dysplasia using depth-resolved angular light scattering profiles.

Area of Science:

• Biomedical optics
• Artificial intelligence in healthcare
• Gynecologic oncology

Background:

• Cervical dysplasia detection relies on visual inspection and biopsy, which can be subjective and time-consuming.
• Accurate and rapid staging of cervical dysplasia is crucial for effective treatment and patient outcomes.
• Advancements in light scattering techniques offer potential for objective tissue analysis.

Purpose of the Study:

• To develop and validate a machine learning model for detecting and staging cervical dysplastic tissue.
• To assess the performance of a convolutional neural network (CNN) using depth-resolved angular scattering data.
• To compare the efficiency of the deep learning approach with traditional methods for cervical dysplasia assessment.

Main Methods:

• Utilized depth-resolved angular scattering measurements from two clinical trials.
• Developed a convolutional neural network (CNN) architecture for data analysis.
• Trained and validated the model using independent datasets to ensure robustness.

Main Results:

• Achieved high classification performance with 90.3% sensitivity, 85.7% specificity, and 87.5% accuracy.
• Demonstrated consistent classification across different instruments for automated/light-based cytology imaging (a/LCI) scans.
• Reduced processing time by a hundredfold compared to the traditional Mie theory inverse light scattering analysis (ILSA) method.

Conclusions:

• The CNN-based machine learning method provides a sensitive, specific, and accurate approach for cervical dysplasia detection and staging.
• This AI-driven technique ensures uniformity in a/LCI scan classification across various devices.
• The significant speed improvement offers a promising, efficient tool for clinical application in cervical dysplasia assessment.