人工神经元网络鉴别星形胶质细胞瘤良恶性的初步研究

Predicting the Malignant Degree of Astrocytoma with Use of Artificial Neural Networks:Pilot Study

目的:基于磁共振影像特点 ,应用人工神经元网络建立计算机辅助诊断系统 ,研究其判断星形胶质细胞肿瘤良、恶性的可行性及其诊断效果。材料和方法:搜集280例星形胶质细胞肿瘤病例的MRI影像资料 ,其中良性169例 ,恶性111例。由放射科医生对MRI图像进行12方面的特征提取并记录。然后将其输入人工神经元网络 ,对网络训练 ,建立计算机辅助诊断系统 ,以数据库病例初步评价其诊断效果并与放射科专家比较其诊断准确性。结果:数据库病例测试表明人工神经元网络的诊断结果为 ,对于良性和恶性星形胶质细胞瘤的诊断准确率分别为92.1 %和94.3 %,特异性分别为93.6 %和89.9%诊断准确性接近放射科专家。结论:神经元网络可以用来进行星形胶质细胞瘤良、恶性的鉴别诊断。本研究建立的计算机辅助诊断系统对于提高良、恶性星形胶质细胞瘤鉴别诊断的准确性和医学影像学教学方面具有一定的实用价值。随着人工智能的快速发展 ,建立计算机辅助诊断系统帮助放射科医生提高诊断的准确性逐渐成为可能。

Purpose:To develop a computer-aided diagnosis(CAD)scheme by using artificial neural networks(ANN)based on MRI features and to study its feasibility of help radiologist in the differential diagnosis of high-grade astrocytomas from low-grade ones.Materials and Methods:Totally280cases of astrocyˉtomas were investigated,among them169are of low-grade astrocytomas and111are of high-grade ones.Totally12MRI features were extracted and recorded by radiologists.ANN was used as classifier to distinguish malignant astrocytoma from benign one based on MRI features extracted by radiologists.The performance of ANN was evaluated and compared with neuroradiologists.Results:The performance of ANN was satisfying.ANN analysis gave a correct diagnosis for high-grade and low-grade astrocytomas with average accuracy of94.3%and92.1%,specificity93.6%and89.9%respectively.Conclusion:ANN can be used to differentiate benign and malignant astrocytoma.The computerized scheme has the potenˉtial to increase diagnostic accuracy of radiologists and can be used routinely in the teaching of radioloˉgy.With the development of artificial intelligence it is possible to design computer-aided diagnosis sysˉtems to help radiologists increase their diagnostic accuracy.