人工智能预测模型在肠梗阻管理中的应用综述及对小儿肠梗阻的启示

A Review of Artificial Intelligence Prediction Models in the Management of Intestinal Obstruction and Their Implications for Pediatric Intestinal Obstruction

摘要: 目的：系统评估人工智能(AI)与深度学习(DL)在肠梗阻诊疗中的应用现状，重点探讨其向小儿肠梗阻领域转化的潜力、挑战与路径。方法：通过系统文献回顾，梳理AI在腹部医学影像及临床决策支持中的应用进展，总结小儿肠梗阻的临床研究现状与决策困境，并分析成人肠梗阻AI模型的技术经验与局限。结果：DL在成人肠梗阻的影像自动检测与风险预测方面已建立较完整的方法学体系，部分模型达到临床级性能(AUC > 0.95)。然而，针对小儿肠梗阻的AI研究仍属空白。儿科独特生理特征、高质量标注数据稀缺、模型可解释性需求及临床转化障碍构成主要挑战。结论：开发基于多模态数据的儿科专用可解释预测模型具有明确临床必要性与技术可行性。未来需通过联邦学习构建儿科数据集，利用迁移学习与域自适应技术突破数据瓶颈，并通过前瞻性临床试验验证模型效用。跨学科协作是推动该领域从经验驱动向数据驱动转型的关键。

Abstract: Objective: To systematically evaluate the current applications of artificial intelligence (AI) and deep learning (DL) in the diagnosis and treatment of intestinal obstruction, with a focus on their potential for translation into pediatric intestinal obstruction, along with associated challenges and pathways. Methods: Through a systematic literature review, key advances of DL in medical fields such as pathological diagnosis, image analysis, and clinical decision support were synthesized. The current state of clinical research and decision-making dilemmas in pediatric intestinal obstruction were summarized, and the technical experiences and limitations of AI models in adult intestinal obstruction were analyzed. Results: DL has established a relatively complete methodological framework for the automatic detection of imaging features and risk prediction in adult intestinal obstruction, with some models achieving clinical-grade performance (AUC > 0.95). However, AI research specifically targeting pediatric intestinal obstruction remains lacking. Major challenges include the unique physiological characteristics of the pediatric population, scarcity of high-quality annotated data, demand for model interpretability, and barriers to clinical translation. Conclusion: The development of a pediatric-specific, interpretable prediction model based on multimodal data demonstrates clear clinical necessity and technical feasibility. Future efforts should focus on constructing pediatric datasets through federated learning, overcoming data bottlenecks using transfer learning and domain adaptation techniques, and validating model utility via prospective clinical trials. Interdisciplinary collaboration is essential to drive the transformation of this field from experience-driven to data-driven paradigms.