Researchers have developed a novel dual-branch framework for segmenting ovaries in medical images, specifically addressing the challenges of endometriosis analysis using both transvaginal ultrasound (TVUS) and magnetic resonance imaging (MRI). This approach adapts the MedSAM3 model by incorporating a prototype bank derived from TVUS data to create anatomically consistent feature representations across both modalities. Experiments on endometriosis datasets demonstrated quantitative and qualitative improvements, with the proposed method outperforming state-of-the-art techniques by over 5 percentage points. An ablation study further highlighted the effectiveness of the prototype bank and the benefits of warm-up pretraining in the source TVUS domain. AI
IMPACT This research could lead to more accurate and efficient diagnostic tools for endometriosis, improving patient outcomes through better image analysis.
RANK_REASON Academic paper detailing a new AI model and methodology. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
