The FetalMind intelligent ultrasound interpretation system.

A team led by Professor Du Bo from Wuhan University's School of Computer Science received the Best Paper Award at the 32nd ACM SIGKDD Conference on Knowledge Discovery and Data Mining.

Their paper, Epistemic-Aware Vision-Language Foundation Model for Fetal Ultrasound Interpretation, offers a novel approach to automatically integrating multi-view fetal ultrasound information, anomaly detection, and report generation, setting a new standard at the intersection of artificial intelligence and clinical medicine.

By integrating prenatal screening big data with AI technologies, the team aims to improve the intelligent identification of birth defect risks, assistive diagnostics, and clinical interventions.

In collaboration with 12 medical centers, the team developed the FetalSigma-1M dataset, comprising 20,566 examinations and 1.19 million ultrasound images, covering 54 sectional views and over 300 diseases across early, mid, and late pregnancy stages.

By using pregnancy-specific spatial alignment, an expert "disease-section" knowledge graph, and salient cognitive disentanglement, the model improves localization of critical abnormal evidence and cross-sectional reasoning.

Compared to models like GPT-5 and Gemini 2.5 Pro, the FetalMind model demonstrated an average performance improvement of 14 percent, achieving an overall diagnostic accuracy of 81.3 percent.