基于人工智能技术的危及器官自动勾画在胸部肿瘤中的应用

目的:评估基于人工智能技术的自动勾画软件勾画胸部危及器官轮廓的几何学精度，为临床应用提供依据。方法:选择30例胸部肿瘤患者的CT图像，分别使用基于人工智能技术的自动勾画软件勾画和医师手动勾画胸部危及器官。采用Hausdorff距离、形状相似性指数及Jaccard系数这3个指标评价自动勾画与手动勾画危及器官的几何学一致性。结果:在肺、心脏和脊髓的Hausdorff距离中，最大为右肺的（22.31±4.50） mm，最小为脊髓的（3.17±0.80） mm。危及器官的形状相似性指数值均≥0.91。Jaccard系数中左肺和右肺的均值≥0.95，脊髓的为0.84±0.02，心脏的略低为0.83±0.04。结论:基于人工智能技术的危及器官自动勾画软件对于胸部危及器官勾画能够达到较高的准确性和精度，可以满足临床工作。【关键词】胸部肿瘤；人工智能；危及器官；自动勾画；放射治疗

Application of automatic organs-at-risk segmentation based on artificial intelligence technology in thoracic tumors

Objective To evaluate the geometric accuracy of automatic segmentation software based on artificial intelligence technology for segmenting the organs-at-risk (OAR) in patients with thoracic tumors, so as to provide a basis for its clinical application. Methods A total of 30 patients with thoracic tumors were enrolled in the study, and the thoracic OAR was automatically delineated by segmentation software and manually segmented by physicians. Three evaluation indexes, namely Hausdorff distance, Dice similarity coefficient and Jaccard coefficient, were used to evaluate the geometric consistency between automatic segmentation and manual segmentation. Results Among the Hausdorff distances of lung-L, lung-R, heart and spinal cord, the maximum Hausdorff distance was (22.31±4.50) mm in lung-R, and the minimum was (3.17±0.80) mm in spinal cord. The Dice similarity coefficient of all OAR (lung-L, lung-R, heart, spinal cord) was not less than 0.91. The mean value of Jaccard coefficient in lung-L and lung-R were greater than or equal to 0.95, while that in spinal cord and heart was 0.84±0.02 and 0.83±0.04, respectively. Conclusion The automatic segmentation software based on artificial intelligence technology can achieve a high accuracy and precision in thoracic OAR segmentation, which can meet the needs of clinical practices.