A 2.5D transformer-based deep learning model for predicting metachronous liver metastasis in rectal cancer: a two-center study

To develop and validate a pretreatment magnetic resonance imaging (MRI)-based 2.5D Transformer deep learning model for predicting metachronous liver metastasis (MLM) in patients with rectal cancer. This retrospective two-center study included 503 patients with pathologically confirmed rectal adenocarcinoma. Patients from Center A were divided into training and internal validation cohorts, while patients from Center B served as an external test cohort. Three 2.5D slice-selection strategies based on T2-weighted imaging were evaluated: axial three-slice, orthogonal maximum, and orthogonal central schemes. DenseNet-201 was used for deep feature extraction, and multiple fusion strategies were compared, including Transformer-based fusion. Independent clinical predictors were integrated with the optimal deep learning signature to construct a combined nomogram. Model performance was assessed using receiver operating characteristic (ROC) analysis, calibration curves, decision curve analysis (DCA), net reclassification improvement (NRI), and integrated discrimination improvement (IDI). The orthogonal maximum strategy achieved the most stable performance among the slice-selection schemes. The Transformer model showed strong discrimination, with areas under the curve (AUCs) of 0.885, 0.864, and 0.858 in the training, internal validation, and external test cohorts, respectively. The combined model integrating the Transformer signature with MRI-reported extramural venous invasion (EMVI) and carbohydrate antigen 19 − 9 (CA19-9) yielded numerically higher AUCs of 0.903, 0.881, and 0.880 in the training, internal validation, and external test cohorts, respectively. The proposed 2.5D Transformer-based model may provide a non-invasive tool for individualized MLM risk stratification in rectal cancer.

Authors

Institutions

Publication Details

Journal
BMC Medical Imaging
Published
2026-11-19
DOI
https://doi.org/10.1186/s12880-026-03072-x
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

A 2.5D transformer-based deep learning model for predicting metachronous liver metastasis in rectal cancer: a two-center study

Xiangbing Tang, Ming Zhao, Hang Zhou, Xueli Zhang et al.
BMC Medical Imaging
Radiomics and Machine Learning in Medical Imaging
article

A 2.5D transformer-based deep learning model for predicting metachronous liver metastasis in rectal cancer: a two-center study

Xiangbing Tang, Ming Zhao, Hang Zhou, Xueli Zhang, Yujie Wang, Chunjing Wang, Mengyao Wang, Zhihao Yang
article en

Abstract

To develop and validate a pretreatment magnetic resonance imaging (MRI)-based 2.5D Transformer deep learning model for predicting metachronous liver metastasis (MLM) in patients with rectal cancer. This retrospective two-center study included 503 patients with pathologically confirmed rectal adenocarcinoma. Patients from Center A were divided into training and internal validation cohorts, while patients from Center B served as an external test cohort. Three 2.5D slice-selection strategies based on T2-weighted imaging were evaluated: axial three-slice, orthogonal maximum, and orthogonal central schemes. DenseNet-201 was used for deep feature extraction, and multiple fusion strategies were compared, including Transformer-based fusion. Independent clinical predictors were integrated with the optimal deep learning signature to construct a combined nomogram. Model performance was assessed using receiver operating characteristic (ROC) analysis, calibration curves, decision curve analysis (DCA), net reclassification improvement (NRI), and integrated discrimination improvement (IDI). The orthogonal maximum strategy achieved the most stable performance among the slice-selection schemes. The Transformer model showed strong discrimination, with areas under the curve (AUCs) of 0.885, 0.864, and 0.858 in the training, internal validation, and external test cohorts, respectively. The combined model integrating the Transformer signature with MRI-reported extramural venous invasion (EMVI) and carbohydrate antigen 19 − 9 (CA19-9) yielded numerically higher AUCs of 0.903, 0.881, and 0.880 in the training, internal validation, and external test cohorts, respectively. The proposed 2.5D Transformer-based model may provide a non-invasive tool for individualized MLM risk stratification in rectal cancer.

BMC Medical Imaging
Heilongjiang Provincial Hospital (CN), Second Affiliated Hospital of Harbin Medical University (CN)
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