Multimodal AI system for the rapid diagnosis and surgical prediction of necrotizing enterocolitis

Abstract

The rapid diagnosis and surgical prediction of necrotizing enterocolitis (NEC) remain a challenge because its complex pathogenesis has not been completely elucidated, and no single medical examination is specific for diagnosing NEC. Artificial intelligence (AI) has proven the robustness of multivariate analysis and been widely used in the diagnosis of complex diseases in the past decade. In this paper, a new multimodal AI system including feature engineering, machine learning (ML), and deep learning (DL) was constructed based on abdominal radiographs (ARs) and clinical data. A total of 4,535 ARs from 1,823 suspected NEC patients were analyzed by transfer learning, and then medical images and clinical parameters from 827 suspected NEC patients were used to train, validate, and test the AI system. Our results demonstrated that the system was effective in diagnosing NEC. In addition, the clinical datasets obtained one week before surgery from 379 NEC patients were studied by the multimodal AI system, and the results showed that it was capable of predicting which NEC patients required surgery. We compared the results in external test sets with those made by clinicians and found that the diagnostic and surgical predictive ability of the AI system was equivalent to that of experienced clinicians. This multimodal AI system can help clinicians improve diagnostic efficiency, reduce the number of missed diagnoses, and facilitate early diagnosis and treatment to prevent disease progression or even death.