A validation system has been developed for an artificial intelligence–based prototype designed to automatically analyze digital colposcopic images for the detection of precancerous lesions and cervical cancer.

An image standardization algorithm was developed, and a dataset comprising 2,525 colposcopic images was collected and annotated. A MobileNetV3 Large convolutional neural network was subsequently trained using the prepared dataset.

The prototype achieved a classification accuracy of 95%, sensitivity of 94%, and specificity of 96% in distinguishing among three categories of colposcopic findings: Normal, LSIL (low-grade squamous intraepithelial lesion), and HSIL+ (high-grade squamous intraepithelial lesion or more severe pathology).

The developed system demonstrates the potential of deep learning technologies in gynecologic oncology and may serve as an auxiliary clinical decision-support tool during screening examinations.

TBA