Abstract
BACKGROUND: Endometriosis is a prevalent gynecological disorder marked by the growth of endometrial-like tissue outside the uterus, often causing pelvic pain, irregular menstruation, and infertility. Despite ongoing research, timely diagnosis remains challenging due to the complex etiology, non-specific symptoms, and the lack of reliable non-invasive diagnostic tools. Current diagnostic approaches, particularly for early-stage endometriosis, are limited, highlighting a critical knowledge gap in accurate and timely detection. Artificial intelligence (AI), when applied to ultrasound imaging, shows promise in addressing this gap by potentially enabling earlier and more accurate diagnosis. This systematic review aims to evaluate the role of AI in improving the diagnosis and prediction of endometriosis using ultrasound images, addressing the unmet need for more effective diagnostic strategies. METHODS: This systematic review was conducted in 2025 following the PRISMA guidelines. A comprehensive search was performed in reputable databases, including PubMed, Web of Science, and Scopus, included as a supplementary source to capture additional relevant studies. The search used the keywords “artificial intelligence,” “diagnosis,” “endometriosis,” and “ultrasound images” without time restrictions. Only English-language studies examining the role of AI in diagnosing endometriosis were included. Two independent reviewers screened titles and abstracts, followed by a full-text review of eligible articles. Data extraction was conducted using two standardized forms: one recording study title, country, number of participants, objectives, and main findings; and the other documenting the type of AI model used, error rate, accuracy, and diagnostic performance. FINDINGS: Five studies were included, applying machine learning and deep learning algorithms to diagnose or predict endometriosis using ultrasound. Deep learning models achieved the highest accuracies (0.89–0.93) and AUC values around 0.90. Machine learning models showed slightly lower performance (accuracy 0.80–0.85, AUC 0.75–0.80) but offered better interpretability. Sensitivity ranged from 0.78 to 0.92 and specificity from 0.74 to 0.89, indicating quantitative improvements in diagnosis using AI compared to traditional methods. CONCLUSION: This review underscores the promising role of artificial intelligence algorithms in improving the accuracy of endometriosis diagnosis through ultrasound imaging, which could facilitate earlier and more effective treatment. The findings suggest that integrating AI into clinical practice has the potential to enhance diagnostic efficiency and patient outcomes. Future research should focus on validating these approaches in real-world settings and promoting awareness among clinicians and patients about the practical benefits and limitations of AI-assisted endometriosis care.