构建基于机器学习与计算机视觉技术的非接触式精准监测系统，以实现对皮瓣血运状态的连续、客观评估及血管危象的早期预警。

纳入2024年3月至2025年4月首都医科大学附属北京积水潭医院手外科接受游离皮瓣移植手术患者84例，录制术后72 h皮瓣移植区域视频，构建包含580个视频的数据集，以8∶2的比例划分为训练集和测试集。采用Mask R-CNN模型实现皮瓣区域自动分割与追踪，提取颜色、纹理及远程光电容积脉搏波信号等8项特征，并基于集成学习算法构建风险预警模型，同时与传统人工监测进行对照验证。

该系统在测试集中表现出较高的分割精度，交并比（IoU）为0.75时，平均精度达1.000；分类模型准确率达0.851±0.085。对于测试集中2例发生血管危象患者，模型预警时间均早于传统人工监测。

本研究构建的人工智能监测系统可初步实现皮瓣血运的无创、连续、精准监测，提升预警时效性与临床实用性，具有降低皮瓣坏死风险、改善术后管理质量的潜力。

To develop a non-contact precision monitoring system based on machine learning and computer vision technology，enabling continuous and objective assessment of flap perfusion status and early warning of vascular crisis.

From March 2024 to April 2025，84 patients who underwent free flap transplantation in the Department of Hand Surgery at Beijing Jishuitan Hospital，Capital Medical University were enrolled. Videos of the flap transplanted areas were recorded during the first 72 hours postoperatively. A dataset comprising 580 videos was constructed and randomly divided into training set and test set at a ratio of 8∶2. The Mask R-CNN model was employed for automatic flap segmentation and tracking. Eight features，including color，texture，and photoplethysmographic signals，were extracted，and a risk warning model was established based on ensemble learning algorithms. The system's performance was compared with conventional manual monitoring for validation.

The system demonstrated high segmentation accuracy on the test set, the average precision reached 1.000 at an intersection over union of 0.75，and the classification model achieved an accuracy of 0.851±0.085. For the two patients who developed vascular crisis in the test set，the model provided earlier warnings than conventional manual monitoring in all cases.

The artificial intelligence–based monitoring system constructed in this study enables non-invasive，continuous，and precise assessment of flap perfusion，enhances warning timeliness and clinical practicality，and holds potential for reducing the risk of flap necrosis and improving the quality of postoperative management.