STUDY PROTOCOL: Diagnostic Accuracy of Artificial Intelligence and Deep Learning Models in the Imaging Diagnosis of Adenomyosis

This registration protocol details a systematic review and meta-analysis focused on the diagnostic performance of Artificial Intelligence (AI) and Deep Learning models for detecting uterine adenomyosis. Adenomyosis is a prevalent gynecological condition often underdiagnosed due to significant interobserver variability in Transvaginal Ultrasound (TVUS) and Magnetic Resonance Imaging (MRI). The primary objective is to provide a high-level quantitative synthesis of AI diagnostic accuracy, specifically focusing on Convolutional Neural Networks (CNNs) and Radiomics. Unlike previous qualitative reviews, this study will employ advanced Bivariate Mixed-Effects Models to calculate pooled sensitivity and specificity, overcoming the threshold effect inherent in computer vision algorithms. Methodological quality will be assessed using QUADAS-2 and the PROBAST+AI tool to evaluate risk of bias and data leakage. The findings aim to establish clinical benchmarks for the integration of AI in precision gynecology, aligning with the "Office of the Future" framework for the SOGESP 2026 congress.