An exploratory PET/MR radiomics framework for preoperative prediction of HER2 status and metastatic burden in colorectal cancer

Introduction: Noninvasive assessment of human epidermal growth factor receptor 2 (HER2) status and metastatic burden in colorectal cancer remains challenging. Multimodal positron emission tomography (PET)/magnetic resonance (MR) radiomics offers an opportunity for integrated molecular and metastatic characterization. Objective: To develop and internally validate an exploratory PET/MR radiomics framework for the preoperative prediction of HER2 status and metastatic burden (non-regional lymph nodes) in colorectal cancer (CRC) patients. Methods: This retrospective study incorporated 61 samples with pathologically confirmed CRC who underwent preoperative 18F-fluorodeoxyglucose PET/MR imaging. HER2 positivity was defined according to the refined HER2 Amplification for CRC Enhanced Stratification criteria. Metastatic burden was categorized into three groups based on preoperative imaging: non-metastatic, regional lymph node metastasis, and distant metastasis. Tumor regions were manually segmented on PET, T1- and T2-weighted imaging (T1WI, T2WI), and 2,264 radiomic features were extracted from each imaging modality. Feature selection was performed using a comprehensive multi-step pipeline. An extreme gradient boosting (XGBoost)-based model was then developed and internally validated via 5-fold cross-validation, with performance metrics reported as mean ± standard deviation. Results: For HER2 status prediction, the multimodal combined XGBoost model demonstrated superior performance compared to single-modality models, yielding a mean AUC of 0.9214 ± 0.0779. For metastatic burden prediction, the combined model further excelled, yielding an AUC of 0.827 ± 0.136. SHapley Additive exPlanations analysis revealed imaging signatures for the two prediction tasks. HER2 status prediction was primarily driven by texture-related features, particularly those derived from the Gray-Level Run Length Matrix, Gray-Level Co-occurrence Matrix, and First-order matrices, whereas metastatic burden prediction relied mainly on multiscale texture descriptors from the first-order, Neighborhood Gray-Tone Difference Matrix, and Gray-Level Dependence Matrix. Conclusion: PET/MR radiomics offers an exploratory approach for future risk stratification in CRC, although independent validation is required to confirm clinical utility.

Authors

Publication Details

Journal
Eurasian Journal of Medicine and Oncology
Published
2026-11-16
DOI
https://doi.org/10.36922/ejmo026270322
Primary Topic
Radiomics and Machine Learning in Medical Imaging
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

An exploratory PET/MR radiomics framework for preoperative prediction of HER2 status and metastatic burden in colorectal cancer

Qingqing Huang, Zengbei Yuan, Weilun Wang, Wen Jiang et al.
Eurasian Journal of Medicine and Oncology
Radiomics and Machine Learning in Medical Imaging
article

An exploratory PET/MR radiomics framework for preoperative prediction of HER2 status and metastatic burden in colorectal cancer

Qingqing Huang, Zengbei Yuan, Weilun Wang, Wen Jiang, Kun Wang, Yihong Yang, Jun Zhao, Zhixian Tang, Zirong Zhou, Na Qi, Xing Chen
article en

Abstract

Introduction: Noninvasive assessment of human epidermal growth factor receptor 2 (HER2) status and metastatic burden in colorectal cancer remains challenging. Multimodal positron emission tomography (PET)/magnetic resonance (MR) radiomics offers an opportunity for integrated molecular and metastatic characterization. Objective: To develop and internally validate an exploratory PET/MR radiomics framework for the preoperative prediction of HER2 status and metastatic burden (non-regional lymph nodes) in colorectal cancer (CRC) patients. Methods: This retrospective study incorporated 61 samples with pathologically confirmed CRC who underwent preoperative 18F-fluorodeoxyglucose PET/MR imaging. HER2 positivity was defined according to the refined HER2 Amplification for CRC Enhanced Stratification criteria. Metastatic burden was categorized into three groups based on preoperative imaging: non-metastatic, regional lymph node metastasis, and distant metastasis. Tumor regions were manually segmented on PET, T1- and T2-weighted imaging (T1WI, T2WI), and 2,264 radiomic features were extracted from each imaging modality. Feature selection was performed using a comprehensive multi-step pipeline. An extreme gradient boosting (XGBoost)-based model was then developed and internally validated via 5-fold cross-validation, with performance metrics reported as mean ± standard deviation. Results: For HER2 status prediction, the multimodal combined XGBoost model demonstrated superior performance compared to single-modality models, yielding a mean AUC of 0.9214 ± 0.0779. For metastatic burden prediction, the combined model further excelled, yielding an AUC of 0.827 ± 0.136. SHapley Additive exPlanations analysis revealed imaging signatures for the two prediction tasks. HER2 status prediction was primarily driven by texture-related features, particularly those derived from the Gray-Level Run Length Matrix, Gray-Level Co-occurrence Matrix, and First-order matrices, whereas metastatic burden prediction relied mainly on multiscale texture descriptors from the first-order, Neighborhood Gray-Tone Difference Matrix, and Gray-Level Dependence Matrix. Conclusion: PET/MR radiomics offers an exploratory approach for future risk stratification in CRC, although independent validation is required to confirm clinical utility.

Eurasian Journal of Medicine and OncologyVol. 0(0)
Radiomics and Machine Learning in Medical Imaging
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon