AI影像辅助诊断系统在急性阑尾炎分型中的准确性与外科决策影响

The Accuracy of AI Imaging-Assisted Diagnostic System in the Classification of Acute Appendicitis and the Impact of Surgical Decision-Making

摘要: 人工智能影像辅助诊断系统在急性阑尾炎分型诊断中展现出显著优势，其通过深度学习算法(如卷积神经网络、随机森林)精准量化阑尾直径、周围渗出及结石特征等影像学参数，有效突破传统诊断瓶颈。研究表明，AI模型对急性阑尾炎分型(单纯性与复杂性)的诊断准确率超过80%，曲线下面积(AUC)峰值达0.985，显著优于传统Alvarado评分系统；其中3D CT分层注意力网络使复杂性阑尾炎分型的AUC提升7.1%，特异性达94%。该系统在多模态影像应用中表现突出：CT导向模型对复杂性阑尾炎(如坏疽或穿孔)的识别敏感度达73%~90.6%；MRI结合AI对孕妇群体的阴性预测值达100%，特异性达97%，有效避免不必要辐射；而超声模型通过迁移学习优化亦提升分型精度。AI分型结果直接影响外科决策：精准区分单纯性(适用抗生素治疗，成功率近70%)与复杂性病变(需紧急手术)后，阴性阑尾切除率降低21%~47%，并优化手术时机选择(如坏疽性阑尾炎的24小时手术窗)。此外，AI系统通过缩短确诊时间(如MRI结合AI使诊断缩短至30分钟内)、减少CT使用率(达27%~40%)及降低穿孔等并发症风险，全面优化患者管理流程。未来需解决多中心数据标准化、算法可解释性及临床整合等挑战，以推进临床转化应用。

Abstract: Artificial Intelligence (AI)-based imaging-assisted diagnostic systems demonstrate significant advantages in the classification diagnosis of acute appendicitis. Utilizing deep learning algorithms (like Convolutional Neural Networks and Random Forests), these systems precisely quantify key imaging parameters, including appendiceal diameter, surrounding exudate, and appendicolith features, effectively overcoming limitations inherent in traditional diagnostic approaches. Research indicates AI models achieve diagnostic accuracy exceeding 80% for differentiating acute appendicitis types (simple vs. complicated), with peak Area Under the Curve (AUC) values reaching 0.985, significantly outperforming the traditional Alvarado scoring system; notably, a 3D CT hierarchical attention network improved the AUC for complicated appendicitis classification by 7.1%, achieving 94% specificity. The system excels in multimodal imaging applications: CT-guided models show 73%~90.6% sensitivity for identifying complicated appendicitis (e.g., gangrenous or perforated); MRI combined with AI achieves a 100% Negative Predictive Value (NPV) and 97% specificity in pregnant populations, effectively avoiding unnecessary radiation; ultrasound models optimized via transfer learning also demonstrate improved classification accuracy. AI classification directly impacts surgical decision-making: by accurately distinguishing simple appendicitis (suitable for antibiotic therapy with nearly 70% success) from complicated disease (requiring urgent surgery), the rate of negative appendectomy rates decreases by 21%~47% and the surgical timing is optimized (e.g., identifying the 24-hour window for gangrenous appendicitis). Furthermore, AI systems optimize patient management processes by shortening diagnostic time (e.g., MRI + AI reducing diagnosis to within 30 minutes), decreasing CT utilization rates (by 27%~40%), and reducing complication risks like perforation. Future advancement requires addressing challenges, including multicenter data standardization, algorithm interpretability, and seamless clinical integration to advance clinical translational implementation.