Machine learning-based radiomics prognostic model for patients with proximal esophageal cancer after definitive chemoradiotherapy

Abstract Objectives To explore the role of radiomics in predicting the prognosis of proximal esophageal cancer and to investigate the biological underpinning of radiomics in identifying different prognoses. Methods A total of 170 patients with pathologically and endoscopically confirmed proximal eso...

Full description

Saved in:
Bibliographic Details
Main Authors: Linrui Li, Zhihui Qin, Juan Bo, Jiaru Hu, Yu Zhang, Liting Qian, Jiangning Dong
Format: Article
Language:English
Published: SpringerOpen 2024-11-01
Series:Insights into Imaging
Subjects:
Online Access:https://doi.org/10.1186/s13244-024-01853-y
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1846147767342202880
author Linrui Li
Zhihui Qin
Juan Bo
Jiaru Hu
Yu Zhang
Liting Qian
Jiangning Dong
author_facet Linrui Li
Zhihui Qin
Juan Bo
Jiaru Hu
Yu Zhang
Liting Qian
Jiangning Dong
author_sort Linrui Li
collection DOAJ
description Abstract Objectives To explore the role of radiomics in predicting the prognosis of proximal esophageal cancer and to investigate the biological underpinning of radiomics in identifying different prognoses. Methods A total of 170 patients with pathologically and endoscopically confirmed proximal esophageal cancer from two centers were enrolled. Radiomics models were established by five machine learning approaches. The optimal radiomics model was selected using receiver operating curve analysis. Bioinformatics methods were applied to explore the potential biological mechanisms. Nomograms based on radiomics and clinical–radiomics features were constructed and assessed by receiver operating characteristics, calibration, and decision curve analyses net reclassification improvement, and integrated discrimination improvement evaluations. Results The peritumoral models performed well with the majority of classifiers in the training and validation sets, with the dual-region radiomics model showing the highest integrated area under the curve values of 0.9763 and 0.9471, respectively, and outperforming the single-region models. The clinical–radiomics nomogram showed better predictive performance than the clinical nomogram, with a net reclassification improvement of 34.4% (p = 0.02) and integrated discrimination improvement of 10% (p = 0.007). Gene ontology enrichment analysis revealed that lipid metabolism-related functions are potentially crucial in the process by which the radiomics score could stratify patients. Conclusions A combination of peritumoral radiomics features could improve the predictive performance of intratumoral radiomics to estimate overall survival after definitive chemoradiotherapy in patients with proximal esophageal cancer. Radiomics features could provide insights into the lipid metabolism associated with radioresistance and hold great potential to guide personalized care. Critical relevance statement This study demonstrates that incorporating peritumoral radiomics features enhances the predictive accuracy of overall survival in proximal esophageal cancer patients after chemoradiotherapy, and suggests a link between radiomics and lipid metabolism in radioresistance, highlighting its potential for personalized treatment strategies. Key Points Peritumoral region radiomics features could predict the prognosis of proximal esophageal cancer. Dual-region radiomics features showed significantly better predictive performance. Radiomics features can provide insights into the lipid metabolism associated with radioresistance. Graphical Abstract
format Article
id doaj-art-ad64937cedd44763a958e99e7fec2cad
institution Kabale University
issn 1869-4101
language English
publishDate 2024-11-01
publisher SpringerOpen
record_format Article
series Insights into Imaging
spelling doaj-art-ad64937cedd44763a958e99e7fec2cad2024-12-01T12:29:25ZengSpringerOpenInsights into Imaging1869-41012024-11-0115111210.1186/s13244-024-01853-yMachine learning-based radiomics prognostic model for patients with proximal esophageal cancer after definitive chemoradiotherapyLinrui Li0Zhihui Qin1Juan Bo2Jiaru Hu3Yu Zhang4Liting Qian5Jiangning Dong6Department of Radiation Oncology, Anhui Provincial Hospital Affiliated to Anhui Medical UniversityDepartment of Radiation Oncology, The First Affiliated Hospital of Anhui Medical UniversityDepartment of Radiation Oncology, Anhui Provincial Hospital Affiliated to Anhui Medical UniversityDepartment of Radiation Oncology, West Branch of the First Affiliated Hospital of University of Science and Technology of ChinaDepartment of Radiation Oncology, The First Affiliated Hospital of Anhui Medical UniversityDepartment of Radiation Oncology, Anhui Provincial Hospital Affiliated to Anhui Medical UniversityDepartment of Radiation Oncology, Anhui Provincial Hospital Affiliated to Anhui Medical UniversityAbstract Objectives To explore the role of radiomics in predicting the prognosis of proximal esophageal cancer and to investigate the biological underpinning of radiomics in identifying different prognoses. Methods A total of 170 patients with pathologically and endoscopically confirmed proximal esophageal cancer from two centers were enrolled. Radiomics models were established by five machine learning approaches. The optimal radiomics model was selected using receiver operating curve analysis. Bioinformatics methods were applied to explore the potential biological mechanisms. Nomograms based on radiomics and clinical–radiomics features were constructed and assessed by receiver operating characteristics, calibration, and decision curve analyses net reclassification improvement, and integrated discrimination improvement evaluations. Results The peritumoral models performed well with the majority of classifiers in the training and validation sets, with the dual-region radiomics model showing the highest integrated area under the curve values of 0.9763 and 0.9471, respectively, and outperforming the single-region models. The clinical–radiomics nomogram showed better predictive performance than the clinical nomogram, with a net reclassification improvement of 34.4% (p = 0.02) and integrated discrimination improvement of 10% (p = 0.007). Gene ontology enrichment analysis revealed that lipid metabolism-related functions are potentially crucial in the process by which the radiomics score could stratify patients. Conclusions A combination of peritumoral radiomics features could improve the predictive performance of intratumoral radiomics to estimate overall survival after definitive chemoradiotherapy in patients with proximal esophageal cancer. Radiomics features could provide insights into the lipid metabolism associated with radioresistance and hold great potential to guide personalized care. Critical relevance statement This study demonstrates that incorporating peritumoral radiomics features enhances the predictive accuracy of overall survival in proximal esophageal cancer patients after chemoradiotherapy, and suggests a link between radiomics and lipid metabolism in radioresistance, highlighting its potential for personalized treatment strategies. Key Points Peritumoral region radiomics features could predict the prognosis of proximal esophageal cancer. Dual-region radiomics features showed significantly better predictive performance. Radiomics features can provide insights into the lipid metabolism associated with radioresistance. Graphical Abstracthttps://doi.org/10.1186/s13244-024-01853-yRadiomicsMachine learningProximal esophageal cancerTumor microenvironment
spellingShingle Linrui Li
Zhihui Qin
Juan Bo
Jiaru Hu
Yu Zhang
Liting Qian
Jiangning Dong
Machine learning-based radiomics prognostic model for patients with proximal esophageal cancer after definitive chemoradiotherapy
Insights into Imaging
Radiomics
Machine learning
Proximal esophageal cancer
Tumor microenvironment
title Machine learning-based radiomics prognostic model for patients with proximal esophageal cancer after definitive chemoradiotherapy
title_full Machine learning-based radiomics prognostic model for patients with proximal esophageal cancer after definitive chemoradiotherapy
title_fullStr Machine learning-based radiomics prognostic model for patients with proximal esophageal cancer after definitive chemoradiotherapy
title_full_unstemmed Machine learning-based radiomics prognostic model for patients with proximal esophageal cancer after definitive chemoradiotherapy
title_short Machine learning-based radiomics prognostic model for patients with proximal esophageal cancer after definitive chemoradiotherapy
title_sort machine learning based radiomics prognostic model for patients with proximal esophageal cancer after definitive chemoradiotherapy
topic Radiomics
Machine learning
Proximal esophageal cancer
Tumor microenvironment
url https://doi.org/10.1186/s13244-024-01853-y
work_keys_str_mv AT linruili machinelearningbasedradiomicsprognosticmodelforpatientswithproximalesophagealcancerafterdefinitivechemoradiotherapy
AT zhihuiqin machinelearningbasedradiomicsprognosticmodelforpatientswithproximalesophagealcancerafterdefinitivechemoradiotherapy
AT juanbo machinelearningbasedradiomicsprognosticmodelforpatientswithproximalesophagealcancerafterdefinitivechemoradiotherapy
AT jiaruhu machinelearningbasedradiomicsprognosticmodelforpatientswithproximalesophagealcancerafterdefinitivechemoradiotherapy
AT yuzhang machinelearningbasedradiomicsprognosticmodelforpatientswithproximalesophagealcancerafterdefinitivechemoradiotherapy
AT litingqian machinelearningbasedradiomicsprognosticmodelforpatientswithproximalesophagealcancerafterdefinitivechemoradiotherapy
AT jiangningdong machinelearningbasedradiomicsprognosticmodelforpatientswithproximalesophagealcancerafterdefinitivechemoradiotherapy