论文部分内容阅读
目的 探讨人工神经网络对颅内胶质瘤术后早期反应性强化与肿瘤残存强化的MRI表现的鉴别诊断价值。资料与方法 构建一个 3层误差反向传播 (backpropagation ,BP)网络来鉴别诊断术后肿瘤残存与正常脑组织的反应性强化。研究对象为 1 0 0例颅内胶质瘤术后患者 ,术后 1个月内至少行 1次MRI检查 ,80例作为训练数据 ,2 0例作为检验数据。对MR图像的 5个指标进行描述 ,作为输入层的输入数据输入网络。用检验数据检验网络性能 ,并用受试者操作特征 (receiveroperatingcharacteristiccurves,ROC)方法评价网络性能。结果 人工神经网络诊断出 1 3例残存肿瘤中的 1 0例 ,敏感性为 76 .9% ,诊断出 7例反应性强化中的 6例 ,特异性为 85 .7% ;放射科医师诊断出 8例残存肿瘤 ,敏感性为 61 .5 % ,诊断出 5例反应性强化 ,特异性为 71 .4%。神经网络ROC曲线下面积Az=0 .81 3± 0 .1 0 5 ,显著高于放射科医师ROC曲线下的面积Az =0 .665± 0 .1 30。结论 人工神经网络能成功鉴别颅内胶质瘤术后的反应性强化与肿瘤残存 ,可以辅助放射科医师鉴别肿瘤残存与正常脑组织的反应性强化
Objective To explore the differential diagnosis value of artificial neural network (MRI) in the MRI manifestations of postoperative early reactivity enhancement and residual tumor enhancement after intracranial glioma. Materials and Methods A 3-layer backpropagation (BP) network was constructed to differentiate the remnants of tumors from those of normal brain tissue after diagnosis. The subjects were 100 cases of intracranial glioma postoperative patients, at least one line after 1 month MRI examination, 80 cases as training data, 20 cases as test data. The five indicators of the MR image are described and input to the network as input data for the input layer. The network performance was tested with test data and the network performance was evaluated using the receiver operating characteristiccharacteristics (ROC) method. Results The artificial neural network detected 10 of 13 residual tumors with a sensitivity of 76.9%. Six of 7 patients with reactive enhancement were diagnosed with a specificity of 85.7%. Radiographers diagnosed Eight cases of residual tumor, the sensitivity was 61.5%, 5 cases were diagnosed with enhanced reactivity, specificity was 71.4%. The area under the ROC curve of neural network, Az = 0.81 3 ± 0.l0 5, was significantly higher than the area under radiosurgical ROC curve, Az = 0.666 ± 0.30. Conclusion Artificial neural network can successfully differentiate intracranial glioma postoperative reactivity enhancement and tumor remnant, which can assist radiologists in differentiating tumor remnant from normal brain tissue.