Background: Transvaginal ultrasound (TVUS) is the primary imaging modality for initial evaluation of endometrial pathology. Deep learning models applied to TVUS imaging could improve detection rates of endometrial carcinoma while reducing inter-observer variability. Methods: We prospectively enrolled 512 patients referred for endometrial evaluation between 2020 and 2023. TVUS images were collected and annotated by two expert gynecologists. A ResNet-50 convolutional neural network was fine-tuned on the dataset using transfer learning from ImageNet weights. Results: The model achieved AUC 0.94 (95% CI: 0.91–0.97), sensitivity 0.91, specificity 0.87 on the external validation cohort (n=128). Performance was superior to the junior radiologist baseline (AUC 0.79, p<0.001) and comparable to senior expert evaluation (AUC 0.96). Conclusions: Our CNN-based model demonstrates high diagnostic accuracy for endometrial cancer detection from TVUS images and represents a viable decision support tool for clinical practice.