Medical researchers developed a Mamba-based deep learning model to predict early liver metastases in patients with pancreatic ductal adenocarcinoma, according to a peer-reviewed study published in Nature Communications. The system integrates imaging features captured from both the primary pancreatic tumor and the liver to evaluate metastatic risk.
Testing across a multi-institutional cohort of 1,063 patients demonstrated prediction scores between 0.806 and 0.890 for the area under the receiver operating characteristic curve. Patients designated as high risk by the model faced significantly worse outcomes, recording a hazard ratio of 1.93 for progression-free survival and 1.89 for overall survival, with both metrics reaching statistical significance at p < 0.001.
Treatment outcomes varied sharply between patient risk groups receiving neoadjuvant therapy. In the high-risk category, neoadjuvant therapy extended overall survival from 17.4 months to 34.1 months, a difference that held statistical significance after propensity score matching at p = 0.004. In contrast, low-risk patients derived no survival benefit from the therapy. Accompanying radiotranscriptomic analyses indicated increased biological aggressiveness within the high-risk segment.
Research teams from Southeast University, Zhejiang University School of Medicine, the University of Science and Technology of China, Northern Jiangsu Province Hospital, Wenzhou Medical University, and Siemens Healthineers contributed to the work. Co-first authors Ben Zhao, Zhifang Gong, and Wenbo Xiao contributed equally to the paper, which was accepted on August 25, 2026, following submission in May 2025. Project funding came from the National Natural Science Foundation of China and the Science and Technology Major Project of Nanjing.
