A prediction model based on deep learning and radiomics features of DWI for the assessment of microsatellite instability in endometrial cancer

IF 2.9 2区医学 Q2 ONCOLOGY

Cancer Medicine Pub Date : 2024-08-22 DOI:10.1002/cam4.70046

Jing Wang, Pujiao Song, Meng Zhang, Wei Liu, Xi Zeng, Nanshan Chen, Yuxia Li, Minghua Wang

{"title":"A prediction model based on deep learning and radiomics features of DWI for the assessment of microsatellite instability in endometrial cancer","authors":"Jing Wang, Pujiao Song, Meng Zhang, Wei Liu, Xi Zeng, Nanshan Chen, Yuxia Li, Minghua Wang","doi":"10.1002/cam4.70046","DOIUrl":null,"url":null,"abstract":"<div>\n \n \n <section>\n \n <h3> Background</h3>\n \n <p>To explore the efficacy of a prediction model based on diffusion-weighted imaging (DWI) features extracted from deep learning (DL) and radiomics combined with clinical parameters and apparent diffusion coefficient (ADC) values to identify microsatellite instability (MSI) in endometrial cancer (EC).</p>\n </section>\n \n <section>\n \n <h3> Methods</h3>\n \n <p>This study included a cohort of 116 patients with EC, who were subsequently divided into training (<i>n</i> = 81) and test (<i>n</i> = 35) sets. From DWI, conventional radiomics features and convolutional neural network-based DL features were extracted. Random forest (RF) and logistic regression were adopted as classifiers. DL features, radiomics features, clinical variables, ADC values, and their combinations were applied to establish DL, radiomics, clinical, ADC, and combined models, respectively. The predictive performance was evaluated through the area under the receiver operating characteristic curve (AUC), total integrated discrimination index (IDI), net reclassification index (NRI), calibration curves, and decision curve analysis (DCA).</p>\n </section>\n \n <section>\n \n <h3> Results</h3>\n \n <p>The optimal predictive model, based on an RF classifier, comprised four DL features, three radiomics features, two clinical variables, and an ADC value. In the training and test sets, this model exhibited AUC values of 0.989 (95% CI: 0.935–1.000) and 0.885 (95% CI: 0.731–0.967), respectively, demonstrating different degrees of improvement compared with the clinical, DL, radiomics, and ADC models (AUC-training = 0.671, 0.873, 0.833, and 0.814, AUC-test = 0.685, 0.783, 0.708, and 0.713, respectively). The NRI and IDI analyses revealed that the combined model resulted in improved risk reclassification of the MSI status compared to the clinical, radiomics, DL, and ADC models. The calibration curves and DCA indicated good consistency and clinical utility of this model, respectively.</p>\n </section>\n \n <section>\n \n <h3> Conclusions</h3>\n \n <p>The predictive model based on DWI features extracted from DL and radiomics combined with clinical parameters and ADC values could effectively assess the MSI status in EC.</p>\n </section>\n </div>","PeriodicalId":139,"journal":{"name":"Cancer Medicine","volume":null,"pages":null},"PeriodicalIF":2.9000,"publicationDate":"2024-08-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://onlinelibrary.wiley.com/doi/epdf/10.1002/cam4.70046","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Cancer Medicine","FirstCategoryId":"3","ListUrlMain":"https://onlinelibrary.wiley.com/doi/10.1002/cam4.70046","RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q2","JCRName":"ONCOLOGY","Score":null,"Total":0}

引用次数: 0

Abstract

Background

To explore the efficacy of a prediction model based on diffusion-weighted imaging (DWI) features extracted from deep learning (DL) and radiomics combined with clinical parameters and apparent diffusion coefficient (ADC) values to identify microsatellite instability (MSI) in endometrial cancer (EC).

Methods

This study included a cohort of 116 patients with EC, who were subsequently divided into training (n = 81) and test (n = 35) sets. From DWI, conventional radiomics features and convolutional neural network-based DL features were extracted. Random forest (RF) and logistic regression were adopted as classifiers. DL features, radiomics features, clinical variables, ADC values, and their combinations were applied to establish DL, radiomics, clinical, ADC, and combined models, respectively. The predictive performance was evaluated through the area under the receiver operating characteristic curve (AUC), total integrated discrimination index (IDI), net reclassification index (NRI), calibration curves, and decision curve analysis (DCA).

Results

The optimal predictive model, based on an RF classifier, comprised four DL features, three radiomics features, two clinical variables, and an ADC value. In the training and test sets, this model exhibited AUC values of 0.989 (95% CI: 0.935–1.000) and 0.885 (95% CI: 0.731–0.967), respectively, demonstrating different degrees of improvement compared with the clinical, DL, radiomics, and ADC models (AUC-training = 0.671, 0.873, 0.833, and 0.814, AUC-test = 0.685, 0.783, 0.708, and 0.713, respectively). The NRI and IDI analyses revealed that the combined model resulted in improved risk reclassification of the MSI status compared to the clinical, radiomics, DL, and ADC models. The calibration curves and DCA indicated good consistency and clinical utility of this model, respectively.

Conclusions

The predictive model based on DWI features extracted from DL and radiomics combined with clinical parameters and ADC values could effectively assess the MSI status in EC.

Abstract Image

查看原文本刊更多论文

基于深度学习和 DWI 放射组学特征的预测模型，用于评估子宫内膜癌的微卫星不稳定性。

研究背景目的：探讨基于从深度学习（DL）和放射组学中提取的弥散加权成像（DWI）特征，结合临床参数和表观弥散系数（ADC）值的预测模型在识别子宫内膜癌（EC）微卫星不稳定性（MSI）方面的功效：这项研究包括116名子宫内膜癌患者，随后将他们分为训练组（81人）和测试组（35人）。从 DWI 中提取传统放射组学特征和基于卷积神经网络的 DL 特征。采用随机森林（RF）和逻辑回归作为分类器。应用 DL 特征、放射组学特征、临床变量、ADC 值及其组合分别建立了 DL、放射组学、临床、ADC 和组合模型。通过接收者操作特征曲线下面积（AUC）、总综合判别指数（IDI）、净再分类指数（NRI）、校准曲线和决策曲线分析（DCA）对预测性能进行评估：基于 RF 分类器的最佳预测模型由四个 DL 特征、三个放射组学特征、两个临床变量和一个 ADC 值组成。在训练集和测试集中，该模型的AUC值分别为0.989（95% CI：0.935-1.000）和0.885（95% CI：0.731-0.967），与临床、DL、放射组学和ADC模型相比有不同程度的提高（AUC-训练=0.671、0.873、0.833和0.814，AUC-测试=0.685、0.783、0.708和0.713）。NRI和IDI分析表明，与临床、放射组学、DL和ADC模型相比，联合模型提高了MSI状态的风险再分类能力。校准曲线和 DCA 分别表明该模型具有良好的一致性和临床实用性：基于从DL和放射组学提取的DWI特征的预测模型与临床参数和ADC值相结合，可有效评估EC的MSI状态。

本文章由计算机程序翻译，如有差异，请以英文原文为准。

求助全文

约1分钟内获得全文求助全文

来源期刊

Cancer Medicine ONCOLOGY-

CiteScore

5.50

自引率

2.50%

发文量

907

审稿时长

19 weeks

期刊介绍： Cancer Medicine is a peer-reviewed, open access, interdisciplinary journal providing rapid publication of research from global biomedical researchers across the cancer sciences. The journal will consider submissions from all oncologic specialties, including, but not limited to, the following areas: Clinical Cancer Research Translational research ∙ clinical trials ∙ chemotherapy ∙ radiation therapy ∙ surgical therapy ∙ clinical observations ∙ clinical guidelines ∙ genetic consultation ∙ ethical considerations Cancer Biology: Molecular biology ∙ cellular biology ∙ molecular genetics ∙ genomics ∙ immunology ∙ epigenetics ∙ metabolic studies ∙ proteomics ∙ cytopathology ∙ carcinogenesis ∙ drug discovery and delivery. Cancer Prevention: Behavioral science ∙ psychosocial studies ∙ screening ∙ nutrition ∙ epidemiology and prevention ∙ community outreach. Bioinformatics: Gene expressions profiles ∙ gene regulation networks ∙ genome bioinformatics ∙ pathwayanalysis ∙ prognostic biomarkers. Cancer Medicine publishes original research articles, systematic reviews, meta-analyses, and research methods papers, along with invited editorials and commentaries. Original research papers must report well-conducted research with conclusions supported by the data presented in the paper.