目的:本研究旨在探讨CT图像、血常规及生化数据在预测膀胱癌复发风险中的价值.方法:回顾性纳入2017年3月-2022年7月于徐州中心医院泌尿外科治疗的65例膀胱癌患者.当患者初发膀胱癌时,收集其术前CT图像、血常规及生化数据.将CT图像进行归一化,并随机旋转-40～40°,以增加数据输入.使用CT图像、血常规及生化数据分别构建多种预测模型(三维卷积神经网络、梯度提升机).采用五倍交叉验证实验及曲线下面积(area under the curve,AUC)评价三维卷积神经网络和梯度提升机的预测性能.结果:利用CT图像训练的三维卷积神经网络准确率为89.0％,曲线下面积为0.888.基于血常规和生化数据训练的梯度提升机准确率分别为94.7％、98.8％,曲线下面积分别为0.898和0.996.结论:机器学习方法在对经尿道膀胱肿瘤切除术后患者的复发预测方面显示出巨大的潜力,或可用于膀胱癌的复发风险分层,进一步指导后续的化学治疗.
Clinical study on artificial intelligence-based prediction of bladder cancer recurrence using preoperative CT images,blood and biochemical data
Objective:To investigate the value of machine learning-based utilization of CT scans,routine blood tests and chemistry panels in predicting the risk of bladder cancer recurrence.Methods:Sixty-five patients with bladder cancer treated at the department of urology of Xuzhou Central Hospital in Jiangsu Province were included retrospectively from March 2017 to July 2022.When the patient first developed bladder cancer,the preoperative CT images,blood routine and biochemical data were collected.CT images were normalized,and were also ran-domly rotated-40 to 40 degrees to increase data input.Two prediction models(3D convolutional neural network and gradient boosting machines)were constructed using CT images,blood routine and biochemical data.Five-fold cross-validation experiments and AUC curve were used to evaluate the prediction performance of 3D convolutional neural network and gradient pusher.Results:The accuracy of the 3D convolutional neural network trained by CT images was 89.0％,and the area under the curve was 0.888.The accuracy of the gradient boosting machines trained based on blood routine and biochemical data were 94.7％and 98.8％,and the area under the curve were 0.898 and 0.996,respectively.Conclusion:Machine learning approaches show great potential for predicting recur-rence in patients after transurethral bladder tumor resection,and may provide insights into the design of risk-level-dependent adjuvant therapies.
NETL
NSTL
万方数据
