基于权重分配的直肠癌病理完全反应预测算法
Prediction Algorithm of Pathological Complete Response of Rectal Cancer Based on Weight Distribution
李兰兰 1徐斌 1李娟 2王大彪3
作者信息
- 1. 福州大学物理与信息工程学院福建省媒体信息智能处理与无线传输重点实验室,福建 福州 350108
- 2. 中山大学附属第六医院广东省结直肠盆底疾病研究重点实验室,广东 广州 510655
- 3. 福州大学石油化工学院,福建 福州 350108
- 折叠
摘要
研究的目的是建立一个深度学习模型,用于进行直肠癌患者新辅助放化疗后的病理完全反应的预测.回顾性分析了99 例直肠癌患者的MR影像资料,并按照训练组(71 例)和测试组(28 例)进行划分构成数据集.通过U-Net定位分割出肿瘤大致区域,在预测阶段通过改变神经网络卷积层数和切片大小得到了 9 个基础预测模型,并且利用权重分配法对预测得分进行修正.在验证组9 个模型中,切片大小为256*256 时,包含4 个卷积层的模型整体性能最好,3 折交叉验证中平均准确率、特异性和敏感性分别达到了 0.714、0.717 和 0.708.研究构建的模型可以作为辅助工具对结直肠癌晚期患者对新辅助治疗的病理反应进行预测,预测精度较好,可为临床治疗提供参考.
Abstract
The purpose of this study is to establish a deep learning model for predicting the pathological complete response of rectal cancer patients after neoadjuvant chemoradiotherapy.The MR imaging data of 99 patients with rectal cancer were retrospectively analyzed,and the data set was divided according to the training group(71 cases)and the test group(28 cases).The approximate tumor area was segmented by U-Net positioning.In the prediction stage,nine basic prediction models were obtained by changing the convolution layers and slice size of the neural net-work,and the prediction score was modified by using the weight distribution method.Among the 9 models in the vali-dation group,when the slice size is 256*256,the model with 4 convolution layers has the best overall performance.The average accuracy,specificity and sensitivity in the 3-fold cross-validation are 0.714,0.717 and 0.708 respec-tively.The model constructed in this study can be used as an auxiliary tool to predict the pathological response of pa-tients with advanced colorectal cancer to neoadjuvant therapy.The prediction accuracy is good and can provide a ref-erence for clinical treatment.
关键词
直肠癌/神经网络/新辅助放化疗/磁共振图像/病理完全反应预测Key words
Rectal cancer/Neural network/Neoadjuvant chemoradiotherapy/Magnetic resonance imaging/Path-ologic complete response引用本文复制引用
基金项目
福建省自然科学基金(2020J01453)
出版年
2024
NETL
NSTL
万方数据