Automated Pap Smear Cervical Cancer Screening Using Deep Learning.

Area of Science:

• Medical Imaging
• Computational Pathology
• Artificial Intelligence in Healthcare

Background:

• Cervical cancer screening relies on accurate analysis of pap smear histological slides.
• Detecting abnormal nuclear features in cervical cells is crucial for early diagnosis.
• Existing methods may face challenges with artifacts and complex cellular structures.

Purpose of the Study:

• To apply Mask Regional Convolutional Neural Network (Mask R-CNN) for automated detection and analysis of cervical cell nuclei in pap smear images.
• To screen for normal and abnormal nuclear features indicative of cervical cancer.
• To evaluate the performance of Mask R-CNN in a clinical setting with real-world data.

Main Methods:

• Utilized liquid-based histological slides from Thammasat University Hospital, including cervical cells and artifacts.
• Implemented Mask R-CNN for nucleus detection and segmentation on pap smear images.
• Compared the performance of the Mask R-CNN approach with a single-cell classification algorithm.

Main Results:

• The Mask R-CNN algorithm achieved a mean average precision (mAP) of 57.8%, with per-image accuracy, sensitivity, and specificity all at 91.7%.
• When adapted for single-cell classification, Mask R-CNN demonstrated 89.8% accuracy, 72.5% sensitivity, and 94.3% specificity on the test dataset.
• The results indicate Mask R-CNN's potential for robust cervical cancer screening.

Conclusions:

• Mask R-CNN shows significant promise as an automated tool for cervical cancer screening by analyzing nuclear features in pap smear images.
• The algorithm's ability to handle artifacts and achieve high diagnostic metrics suggests its clinical utility.
• Further research and validation are warranted to integrate this AI approach into routine cervical cancer screening protocols.