Researchers have developed AbdomenNet, an AI foundation model that detects 11 acute abdominal conditions and handles three risk-stratification subtasks from non-contrast computed tomography scans, according to a study published in Nature Communications.
Medical teams frequently use non-contrast computed tomography (NCCT) as an initial imaging tool for abdominal emergencies due to contraindications to contrast agents, atypical presentations, or resource constraints. To analyze these images, the researchers pre-trained the self-supervised model on 103,989 NCCT examinations before fine-tuning it on 5,816 annotated cases.
External validation across three independent cohorts comprising 2,528 patients yielded a macro-average area under the receiver operating characteristic curve (AUROC) of 0.919 for five emergent abdominal conditions.
In a multi-reader multi-case crossover study, access to AbdomenNet increased radiologists' mean AUROC from 0.812 to 0.924. The tool also reduced clinicians' median reading time by 52.5 seconds per case. Retrospective workflow reconstruction indicated that prioritizing cases with the model could shorten median report turnaround time by 37 minutes.
Co-first authors Chao Zhu and Ruipeng Zhang developed the system alongside researchers from Shanghai Jiao Tong University School of Medicine, Southeast University Affiliated Nantong First People’s Hospital, Wuxi No.2 People’s Hospital, Shanghai East Hospital, and Renmin Hospital of Wuhan University. Grants from the Ministry of Science and Technology and the National Natural Science Foundation of China supported the project, with the authors declaring no competing interests.
