网站首页            期刊简介             编委会             投稿指南             期刊订阅             下载中心             在线留言            联系我们             English
  2025年1月11日 星期六  
文章快速检索
中国生物医学工程学报  2023, Vol. 42 Issue (5): 583-593    DOI: 10.3969/j.issn.0258-8021.2023.05.008
  论著 本期目录 | 过刊浏览 | 高级检索 |
基于MRI与优化3D-ResNet18的鼻咽癌复发预测模型
李炯逸1, 李彬1*, 邱前辉2*, 刘遗斌2, 田联房1
1(华南理工大学自动化科学与工程学院,广州 510640)
2(南方医科大学附属广东省人民医院(广东省医学科学院)耳鼻咽喉科,广州 510080)
Predictive Model for Nasopharyngeal Carcinoma Recurrence with MRI and
Optimized 3D-ResNet18
Li Jiongyi1, Li Bin1*, Qiu Qianhui2*, Liu Yibin2, Tian Lianfang1
1(School of Automation Science and Engineering, South China University of Technology, Guangzhou 510640, China)
2(Department of Otolaryngology & Head and Neck Surgery, Guangdong Provincial People's Hospital (Guangdong Academy of Medical Sciences), Southern Medical University, Guangzhou 510080, China)
全文: PDF (4862 KB)   HTML (1 KB) 
输出: BibTeX | EndNote (RIS)      
摘要 鼻咽癌的治疗后复发是导致治疗失败的重要原因,同时也对患者的生活质量、治愈率甚至生存率产生严重的不利影响。因此,实现鼻咽癌复发情况的有效预测对患者的预后具有积极作用。磁共振成像(MRI)具有软组织高分辨率功能,是鼻咽癌的首选检查手段。鼻咽癌在MRI中的影像表现存在较大差异,病变组织灰度不均匀且界限模糊,基于MRI的鼻咽癌病灶手工标注的难度大、成本高、准确性存在局限;而自动分割准确率也欠佳,导致依靠鼻咽癌病灶精确分割的浅层影像特征提取和计算精度较低,以致基于影像组学特征工程和传统机器学习方法的鼻咽癌复发预测模型性能不佳。对此,本研究提出一种基于MRI和Nesterov加速梯度优化3D-ResNet18的鼻咽癌复发预测模型。通过距离正则化水平集和均衡化增强的鼻咽癌MRI病灶自动检测,自动获取去冗余的增强影像数据,基于Nesterov加速梯度算法优化的改进3D-ResNet18网络模型,提取鼻咽癌深度特征并实现复发预测,为病人的治疗方案提供指导。研究在140例鼻咽癌患者的MRI影像上展开并完成模型训练与交叉验证分析。改进模型的敏感性、特异性和准确率分别为80.0%、64.6%和72.3%,AUC值为0.75,同条件下分别对比3D-ResNet10模型和Momentum优化方法的配对t检验P值分别为0.040和0.006,所改进模型具有显著优势。基于MRI和优化3D-ResNet18的鼻咽癌复发预测模型可实现鼻咽癌复发的有效预测。
服务
把本文推荐给朋友
加入我的书架
加入引用管理器
E-mail Alert
RSS
作者相关文章
李炯逸
李彬
邱前辉
刘遗斌
田联房
关键词 鼻咽癌复发预测模型残差神经网络网络优化    
Abstract:Recurrence of nasopharyngeal carcinoma (NPC) after treatment is an important factor of treatment failure, which is extremely harmful to the quality of life, cure rate and even survival rate of patients with NPC. Therefore, effective prediction of nasopharyngeal carcinoma recurrence plays crucial roles in the prognosis of NPC. Magnetic resonance imaging (MRI) has high resolution of soft tissue, which is a preferred inspection method of NPC. Each imaging of NPC in MRI is quite different, and the gray level of lesion tissue is uneven and the boundaries are blurred, which cause the fact that manual segmentation is difficult, costly, and has limited accuracy, while automatic segmentation of NPC lesions in MRI has low accuracy rate. As a result, low-level image feature extraction and computation based on NPC lesion segmentation has low accuracy rate as well. Performance of NPC recurrence prediction model with radiomics feature engineering and traditional machine learning methods is poor. To solve this problem, a nasopharyngeal carcinoma recurrence prediction model was proposed with MRI and Nesterov accelerating gradient optimized 3D-ResNet18. Through the automatic detection of NPC lesion by distance regularized level-set evolution and histogram equalization in MRI, the enhanced imaging data without redundancy was automatically obtained. The improved 3D-ResNet18 network model optimized by Nesterov accelerated gradient algorithm was used to extract the deep features of NPC and achieve recurrence prediction, providing guidance on the patients’ treatment plans. The research was conducted on MRI images of 140 patients with NPC to complete the model training and cross-validation analysis. Recurrence of NPC was predicted with sensitivity, specificity, accuracy, and AUC of 80.0%, 64.6%, 72.3% and 0.75 respectively. The p values of paired t-test comparing 3D-ResNet10 model and Momentum optimization method under the same conditions were 0.040 and 0.006 respectively. The results showed that the predictive model of nasopharyngeal carcinoma recurrence with improved 3D-ResNet18 can effectively predict the nasopharyngeal carcinoma recurrence.
Key wordsnasopharyngeal carcinoma    recurrence prediction model    ResNet    network optimization
收稿日期: 2022-07-08     
PACS:  R318  
基金资助:国家自然科学基金(62273155); 2021年广东省科技专项资金(“大专项+任务清单”)(210719145863737)
通讯作者: *E-mail: binlee@scut.edu.cn;qiuqianhui@gdph.org.cn   
引用本文:   
李炯逸, 李彬, 邱前辉, 刘遗斌, 田联房. 基于MRI与优化3D-ResNet18的鼻咽癌复发预测模型[J]. 中国生物医学工程学报, 2023, 42(5): 583-593.
Li Jiongyi, Li Bin, Qiu Qianhui, Liu Yibin, Tian Lianfang. Predictive Model for Nasopharyngeal Carcinoma Recurrence with MRI and
Optimized 3D-ResNet18. Chinese Journal of Biomedical Engineering, 2023, 42(5): 583-593.
链接本文:  
http://cjbme.csbme.org/CN/10.3969/j.issn.0258-8021.2023.05.008     或     http://cjbme.csbme.org/CN/Y2023/V42/I5/583
版权所有 © 2015 《中国生物医学工程学报》编辑部
本系统由北京玛格泰克科技发展有限公司设计开发