Radiomics for the prediction of Ki-67 expression in rectal cancer patients: a systematic review and meta-analysis

Ki-67 is a marker of tumor proliferation with potential value for risk stratification and personalized management in rectal cancer. However, conventional immunohistochemical assessment requires tissue sampling and may be affected by intra-tumoral heterogeneity. MRI-based radiomics offer non-invasive approaches for predicting Ki-67 expression. This systematic review and meta-analysis evaluated its diagnostic performance. This review followed PRISMA 2020 and searched PubMed, Scopus, Embase, and Web of Science through August 30, 2026. Studies evaluating radiomics or deep learning models based on pretreatment MRI for predicting Ki-67 expression in rectal cancer were included. Methodological quality was assessed using QUADAS-2 and the Radiomics Quality Score. Training and validation cohorts were analyzed separately. For radiomics studies providing 2 × 2 data, sensitivity and specificity were pooled using a bivariate random-effects model; positive and negative likelihood ratios, diagnostic odds ratios, and summary receiver operating characteristic curve areas were also calculated. Because only four studies were eligible for quantitative synthesis, the pooled analyses were considered exploratory. Out of 8218 records, 5 retrospective studies comprising 1585 rectal cancer patients were included. Four studies used radiomics, and one was based on a deep learning pipeline. An exploratory quantitative synthesis of four radiomics studies yielded a validation sensitivity of 0.85 (95% CI, 0.80–0.89), specificity of 0.60 (95% CI, 0.40–0.77), and SROC AUC of 0.842. The deep learning study reported AUCs of 0.97 and 0.98 in its training and validation sets; however, these figures require caution as this single study carries a high risk of bias, suggesting potential overfitting rather than clinical superiority. Methodological quality was limited, with a mean RQS score of 16.25/36 and a notable risk of bias in the deep learning study. Radiomics models applied to preoperative MRI yielded exploratory diagnostic estimates for the non-invasive prediction of Ki-67 expression in rectal cancer. Deep-learning findings were limited to non-comparable descriptive observations and do not establish superiority over radiomics. Because the available evidence is retrospective and methodologically limited, and no study evaluated effects on clinical decisions or patient outcomes, further prospective external validation is required before clinical application.

Authors

Institutions

Publication Details

Journal
International Journal of Colorectal Disease
Published
2026-11-19
DOI
https://doi.org/10.1007/s00384-026-05262-z
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

Radiomics for the prediction of Ki-67 expression in rectal cancer patients: a systematic review and meta-analysis

Sina Delazar, Mojtaba Sedaghat, Faeze Salahshour, Alireza Kazemeini et al.
International Journal of Colorectal Disease
Radiomics and Machine Learning in Medical Imaging
article

Radiomics for the prediction of Ki-67 expression in rectal cancer patients: a systematic review and meta-analysis

Sina Delazar, Mojtaba Sedaghat, Faeze Salahshour, Alireza Kazemeini, Mohammad Sadegh Fazeli, Amir Keshvari, Behnam Behboudi, Yassin Rahnama, Mohammad Reza Keramati, Mahdiar Pirzadeh, Seyed Mohsen Ahmadi-Tafti
article en

Abstract

Ki-67 is a marker of tumor proliferation with potential value for risk stratification and personalized management in rectal cancer. However, conventional immunohistochemical assessment requires tissue sampling and may be affected by intra-tumoral heterogeneity. MRI-based radiomics offer non-invasive approaches for predicting Ki-67 expression. This systematic review and meta-analysis evaluated its diagnostic performance. This review followed PRISMA 2020 and searched PubMed, Scopus, Embase, and Web of Science through August 30, 2026. Studies evaluating radiomics or deep learning models based on pretreatment MRI for predicting Ki-67 expression in rectal cancer were included. Methodological quality was assessed using QUADAS-2 and the Radiomics Quality Score. Training and validation cohorts were analyzed separately. For radiomics studies providing 2 × 2 data, sensitivity and specificity were pooled using a bivariate random-effects model; positive and negative likelihood ratios, diagnostic odds ratios, and summary receiver operating characteristic curve areas were also calculated. Because only four studies were eligible for quantitative synthesis, the pooled analyses were considered exploratory. Out of 8218 records, 5 retrospective studies comprising 1585 rectal cancer patients were included. Four studies used radiomics, and one was based on a deep learning pipeline. An exploratory quantitative synthesis of four radiomics studies yielded a validation sensitivity of 0.85 (95% CI, 0.80–0.89), specificity of 0.60 (95% CI, 0.40–0.77), and SROC AUC of 0.842. The deep learning study reported AUCs of 0.97 and 0.98 in its training and validation sets; however, these figures require caution as this single study carries a high risk of bias, suggesting potential overfitting rather than clinical superiority. Methodological quality was limited, with a mean RQS score of 16.25/36 and a notable risk of bias in the deep learning study. Radiomics models applied to preoperative MRI yielded exploratory diagnostic estimates for the non-invasive prediction of Ki-67 expression in rectal cancer. Deep-learning findings were limited to non-comparable descriptive observations and do not establish superiority over radiomics. Because the available evidence is retrospective and methodologically limited, and no study evaluated effects on clinical decisions or patient outcomes, further prospective external validation is required before clinical application.

International Journal of Colorectal Disease
University of Tehran (IR), Imam Khomeini Hospital (IR), Tehran University of Medical Sciences (IR)
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