Introduction:
Endometriosis, a chronic inflammatory gynecological condition characterized by ectopic endometrial-like tissue, is frequently subject to substantial diagnostic delays. Although transvaginal ultrasound (TVS) and magnetic resonance imaging (MRI) are highly effective in specialized settings, their interpretation remains heavily dependent on operator expertise.

Aim:
To synthesize current evidence on the diagnostic performance of artificial intelligence (AI) methods in endometriosis imaging via TVS and MRI, and to critically compare the clinical applicability, operator dependency, and regulatory challenges governing AI across both modalities.

Material and methods:
A review of studies published between 2020 and 2026 was conducted across PubMed/MEDLINE, Embase, Scopus, and Google Scholar, evaluating AI models in endometriosis imaging and reporting diagnostic performance metrics.

Results and Discussion:
Deep learning models (e.g., ConvNeXt, IC3D, ResNet) and radiomic nomograms achieved high performance in diagnosing ovarian endometriomas and pouch of Douglas obliteration on TVS (sensitivity: 59.5%-90.0%, specificity: 71.7%-100.0%, AUC: 0.760-0.987, accuracy: 73.0%-95.1%). On pelvic MRI, deep learning models demonstrated high performance in deep endometriosis (DE) (diagnostic accuracy: 92.4%, AUC: 0.880 in reader-assist validation).

Conclusions:
Artificial intelligence can aid in diagnosis of endometriosis by reducing subjectivity and automating image analysis. However, the current evidence is preliminary and multicenter prospective studies are needed. Clinical integration requires adherence to the 2022 ESHRE guidelines and emerging EU AI Act regulatory frameworks.