摘要
目的提出一种基于多源特征融合技术结合限制内存-拟牛顿法-反向传播(Limited-Memory Broyden-Fletcher-Goldfarb-Shanno-Back Propagation,L-BFGS-BP)神经网络模型,以期为乳腺癌筛选和诊断提供参考依据。方法选取2016年9月1日至2022年8月31日于南京大学医学院附属金陵医院收治的388例乳腺癌确诊患者及同时期收治的288例非乳腺癌患者为研究对象,从生物遗传学、临床特征、血清标志物、影像学等方向收集并整理多源特征集,建立L-BFGS优化算法及L-BFGS-BP模型。结果相对于随机森林、BP神经网络模型、支持向量机、朴素贝叶斯模型,L-BFGS-BP模型测试准确度分别提高了8.07%、13.55%、3.55%和8.39%,且差异具有统计学意义(P<0.05);精准度分别提高了9.12%、16.42%、7.50%和7.19%,且差异具有统计学意义(P<0.05);L-BFGS-BP模型在召回率、F1分数方面亦有相同的结果。结论L-BFGS-BP模型具有更好的鲁棒性、更快的收敛速度和更好的优化能力,预测能力较强,具有广泛的应用前景和研究价值。
Objective To propose a multi-source features fusion technology combined with limited-memory broyden-fletcher-goldfarbshanno-back propagation(L-BFGS-BP)neural network model to provide reference for screening and diagnosis of breast cancer.Methods A total of 388 breast cancer patients and 288 non breast cancer patients who were diagnosed in Jinling Hospital,Affiliated Hospital of Medical School,Nanjing University,from September 1,2016 to August 31,2022 were collected as research objects.Multi source feature sets were collected and sorted out from the aspects of biogenetics,clinical characteristics,serum markers,imaging,etc.The L-BFGS optimization algorithm and L-BFGS-BP model were established.Results Compared with random forest,BP neural network model,support vector machine,naive Bayes model,the accuracy of L-BFGS-BP model test increased by 8.07%,13.55%,3.55%and 8.39%,with statistically significant differences(P<0.05);the accuracy had been improved by 9.12%,16.42%,7.50%,and 7.19%respectively,with statistically significant differences(P<0.05).The L-BFGS-BP model also showed the same results in recall rate and F1 score.Conclusion L-BFGS-BP model has a better robustness,faster rate of convergence,better optimization ability,strong prediction ability,which has broad application prospects and research value.
作者
王龙琦
周学超
WANG Longqi;ZHOU Xuechao(Department of Vascular Surgery,Jinling Hospital,Affiliated Hospital of Medical School,Nanjing University,Nanjing Jiangsu 210000,China;Second Department of Interventional Oncology and Vascular Disease,Nanjing Hospital Affiliated to Nanjing University of Chinese Medicine,Nanjing Jiangsu 210000,China)
出处
《中国医疗设备》
2024年第8期55-61,共7页
China Medical Devices
