论文部分内容阅读
目的建立基于反传(BP)神经网络技术的甲状腺癌诊断模型,并评估该模型的临床应用价值。方法回顾性分析2010年1月至2011年8月期间南京市鼓楼医院收治的甲状腺癌患者103例及甲状腺良性病变患者51例,提取其超声图像的9个特征,循建模规则,建立基于BP神经网络技术的甲状腺癌诊断模型,依此模型对2011年9月至2011年12月期间收治的根据超声图像特征疑为甲状腺癌的42例患者进行术前诊断,其结果与术后病理诊断结果(术后病理诊断为甲状腺癌32例,甲状腺良性病变10例)进行对比研究。结果甲状腺癌诊断模型对建模样本的诊断准确率为95.45%(147/154);术前样本的诊断准确率为90.48%(38/42);所有样本的诊断准确率为94.39%(185/196)。结论从本组有限的病例结果初步得出,基于BP神经网络技术的甲状腺癌诊断模型具有较高的可行性及可靠性,可望成为一种全新的甲状腺癌辅助诊断方法。
Objective To establish a diagnostic model of thyroid cancer based on backpropagation (BP) neural network and evaluate the clinical value of this model. Methods A retrospective analysis of 103 cases of thyroid cancer and 51 cases of thyroid benign lesions admitted to Nanjing Drum Tower Hospital from January 2010 to August 2011 was conducted. Nine features of ultrasound images were extracted and the rules of modeling were followed. Based on BP Neural network technology of thyroid cancer diagnosis model, according to this model from September 2011 to December 2011 were treated according to the characteristics of ultrasound images of 42 cases of suspected thyroid cancer patients preoperative diagnosis, the results and postoperative pathological diagnosis (Pathological diagnosis of thyroid cancer in 32 cases, thyroid benign lesions in 10 cases) were compared. Results The diagnostic accuracy of the thyroid cancer diagnosis model was 95.45% (147/154) for the modeling samples, 90.48% (38/42) for the preoperative samples, and 94.39% (185 / 196). Conclusions From the results of a limited number of cases in this group, we initially conclude that the thyroid cancer diagnosis model based on BP neural network is feasible and reliable, and is expected to become a new diagnostic method for thyroid cancer.