宫颈癌CT图像诊断系统对淋巴结勾画的效能评估

Assessment of Efficiency of Cervical Cancer CT Image Diagnostic System for Lymph Node Delineation

目的比较基于卷积神经网络(CNN)的宫颈癌图像诊断系统对CT断层图像阅片与人工阅片判定淋巴结情况,评估图像诊断系统对淋巴结勾画的应用价值。资料与方法利用图像诊断系统对56例宫颈癌患者肾门水平以下腹膜后、髂总血管旁及双侧盆壁淋巴结进行勾画,并由2名影像科副主任医师对所勾画的感兴趣区进行判定,观察有无漏诊的淋巴结。所有患者均选取10枚与感兴趣区淋巴结分布在相同位置的不同大小、密度与淋巴结相近的血管断面作为阴性对照。计算图像诊断系统与人工阅片的一致性、敏感度及特异度。结果共选取679枚淋巴结及阴性对照结节560枚。诊断系统准确勾画643枚,漏诊36枚;将560枚阴性对照结节误勾画为淋巴结78枚。诊断系统与人工阅片结果具有高度一致性(Kappa=0.813,P<0.01)。诊断敏感度为94.70%,特异度为86.07%,受试者工作特征曲线下面积为0.904。结论基于CNN的宫颈癌图像诊断系统对于宫颈癌患者CT图像淋巴结勾画与人工阅片具有高度一致性,可应用于临床辅助诊断。

Purpose To compare the uterus cervical cancer image diagnosis system based on convolution neural networks(CNN)to determine the lymph node condition in CT tomography and manual reading,and to evaluate the application value of image diagnosis system to lymph node delineation.Materials and Methods Using an image diagnosis system,56 patients with cervical cancer were delineated with retroperitoneum,para-iliac vessels,and bilateral pelvic lymph nodes below the renal hilar level.Observed whether there were missed lymph nodes.All patients selected 10 blood vessel sections with different sizes and densities close to the lymph nodes in the same location as the lymph nodes of the region of interest as negative controls.Calculated the consistency,sensitivity and specificity of the image diagnosis system and manual reading.Results A total of 679 lymph nodes and 560 negative control nodules were selected.The diagnostic system accurately delineated 643 and missed 36 diagnosis;mistakenly delineated 560 negative control nodules as 78 lymph nodes.The diagnosis system and the manual reading results were highly consistent(Kappa=0.813,P<0.01).The diagnostic sensitivity,specificity and the area under the receiver operating characteristic curve was 94.70%,86.07%and 0.904,respectively.Conclusion CNN-based cervical cancer image diagnosis system is highly consistent with the CT image lymph node delineation and manual reading of cervical cancer patients,and it can be used in clinical auxiliary diagnosis.