Cervical lesion assessment using real-time microendoscopy image analysis in Brazil: The CLARA study

Area of Science:

• Gynecology
• Medical Imaging
• Oncology

Background:

• Cervical intraepithelial neoplasia (CIN) screening relies on methods like colposcopy.
• There is a need for accessible, cost-effective diagnostic tools for CIN.
• High-resolution microendoscopy (HRME) offers potential for improved cervical cancer screening.

Purpose of the Study:

• To evaluate the diagnostic performance of HRME for detecting CIN.
• To compare HRME with automated image analysis against colposcopy.
• To assess the impact of a neural network algorithm on HRME's diagnostic accuracy.

Main Methods:

• Prospective study comparing HRME and colposcopy with biopsy.
• Automated morphologic image analysis for HRME.
• Retrospective analysis using a multi-task convolutional neural network for HRME image classification.
• Histopathology served as the gold standard for CIN diagnosis.

Main Results:

• HRME with morphologic analysis showed similar sensitivity and specificity to colposcopy for CIN3+ detection.
• HRME with morphologic analysis was slightly less sensitive and specific for CIN2+ detection compared to colposcopy.
• HRME with neural network analysis demonstrated comparable sensitivity and specificity to colposcopy for CIN2+ and CIN3+ detection.

Conclusions:

• HRME, particularly with neural network analysis, is a viable alternative to colposcopy for CIN detection.
• HRME offers comparable diagnostic accuracy to colposcopy for cervical intraepithelial neoplasia.
• HRME presents a potential low-cost, point-of-care solution for cervical cancer prevention.