OvAi Focus: AI segments ovarian lesions with 0.87 DICE score

Ovarian cancer remains the deadliest gynecological malignancy, yet ultrasound interpretation suffers from high operator variability and morphological complexity. To address this, SynDiag s.r.l. (Italy) developed OvAi Focus, a CE-marked AI medical device that performs automated multi-class semantic segmentation directly on ultrasound images. The software distinguishes functional ovaries from adnexal masses and further separates cystic versus solid components—key for cancer risk stratification. The model was trained and independently validated on a multicenter dataset of 1,081 adult women from six clinical centers in Italy and Israel, covering diverse acquisition settings.

Performance metrics are strong: DICE scores reached 0.87 for complete lesion segmentation, 0.85 for cystic regions, 0.68 for solid components, and 0.62 for functional ovaries. These results are in line with or superior to state-of-the-art approaches, despite the challenging heterogeneity of real-world ultrasound data. OvAi Focus will be presented at Ital-IA 2026 (Rome) and published in CEUR-WS proceedings. By providing objective, reproducible segmentation, the software could aid radiologists and gynecologists in earlier and more accurate assessment of ovarian pathology.