深度学习模型构建及其在骨肉瘤生存预测中的应用  

Construction of deep learning models and their application in the prediction of survival of osteosarcoma

作　　者：王燕 杨恒[2] 邵彩霞[3] 李奇俊 孟祥光[5] WANG Yan;YANG Heng;SHAO Caixia;LI Qijun;MENG Xiangguang(Department of Emergency,Zhengzhou Orthopaedic Hospital,Zhengzhou 450052,China;Department of Physical Examination,the Seventh People’s Hospital of Zhengzhou,Zhengzhou 450016,China;Operating Room,the Seventh People’s Hospital of Zhengzhou,Zhengzhou 450016,China;Department of Dermatology,Puyang Oilfield General Hospital,Puyang 457001,China;Department of Pharmacy,the Seventh People’s Hospital of Zhengzhou,Zhengzhou 450016,China)

机构地区：[1]郑州市骨科医院急诊科,河南郑州450052 [2]郑州市第七人民医院体检科,河南郑州450016 [3]郑州市第七人民医院手术室,河南郑州450016 [4]濮阳油田总医院皮肤科,河南濮阳457001 [5]郑州市第七人民医院药学部,河南郑州450016

出　　处：《肿瘤基础与临床》2025年第1期14-20,共7页journal of basic and clinical oncology

基　　金：国家自然科学基金面上项目(81974511)。

摘　　要：目的基于深度学习框架构建1、3和5年生存模型以预测骨肉瘤患者的生存情况。方法从SEER数据库获取骨肉瘤患者临床资料,利用pycox生存分析包构建骨肉瘤生存预测分类器,并进行模型预测和评估。结果数据清洗后最终3190例原发性骨肉瘤患者被入组。在3个数据集中,Log-rank检验结果显示,年龄、种族、肿瘤原发位、肿瘤转移程度、婚否、居住地和收入与生存有关(P<0.05)。3个预测模型的最小损失值分别为0.89、1.39和1.46,C指数分别为0.76、0.70和0.70,Brier分数分别为0.09、0.15和0.18。结论年龄、肿瘤原发位、肿瘤转移程度、婚否和居住地对骨肉瘤生存分类器具有较大的影响。Objective To construct the one-,three-,and five-year survival model based on a deep learning framework to predict the survival of patients with osteosarcoma.Methods The clinical data of patients with osteosarcoma were obtained from the SEER database.The pycox survival analysis package was used to construct a survival prediction classifier for osteosarcoma,and model prediction and evaluation were conducted.Results After data cleaning,a total of 3190 patients with primary osteosarcoma were enrolled.In the three datasets,the Log-rank test results showed the feature variables of age,race,tumor primary site,tumor metastasis degree,marital status,residence and income were related survival.The minimum loss values of the three prediction models were 0.89,1.39,and 1.46,respectively,with evaluation indexes C-index of 0.76,0.70,and 0.70,and Brier Score of 0.09,0.15,and 0.18,respectively.Conclusion Age,tumor primary site,tumor metastasis degree,marital status,and residence have large effects on the survival classifier for osteosarcoma.

关 键 词：骨肉瘤 生存率 深度学习 机器学习 人工神经网络 

分 类 号：R738.1[医药卫生—肿瘤]