MétaCan
Menu
Back to cohort
Record W4416262808 · doi:10.1016/j.eclinm.2025.103640

A non-invasive MRI-based multimodal fusion deep learning model (MF-DLM) for predicting overall survival in bladder cancer: a multicentre retrospective study

2025· article· en· W4416262808 on OpenAlexaff
Lingkai Cai, Rongjie Bai, Qiang Cao, Weijie Sun, Fei Wang, Xiaotong Liu, Bo Liang, Meihua Jiang, Gongcheng Wang, Qiang Shao, Xuping Jiang, Chenghao Wang, Chang Chen, Meiling Bao, Hao Yu, Pengchao Li, Xiao Yang, Qiang Lü

Bibliographic record

VenueEClinicalMedicine · 2025
Typearticle
Languageen
FieldMedicine
TopicBladder and Urothelial Cancer Treatments
Canadian institutionsCanadian VIGOUR CentreUniversity of Alberta
FundersChina Postdoctoral Science FoundationNational Natural Science Foundation of China
KeywordsFoundation (evidence)Retrospective cohort studyChinaOverall survivalDeep learning

Abstract

fetched live from OpenAlex

Background: Accurate prognosis prediction in bladder cancer (BCa) is crucial for personalized treatment. This study aimed to develop and validate a non-invasive model using magnetic resonance imaging (MRI) for predicting the overall survival (OS) in patients with BCa. Methods: This retrospective multicentre study included 1131 patients with BCa from eight institutions in China from June 2011 to March 2024. 871 patients were enrolled from one centre, who were randomly divided (8:2) into training (n = 697) and internal validation (n = 174) sets. For the external test set, 260 patients with BCa from seven centres were retrospectively included. We developed a multimodal fusion deep learning model (MF-DLM), leveraging a cross-attention mechanism to integrate four key preoperative data modalities: three-dimensional (3D) deep learning features using a modified 3D ResNet50 network, 3D radiomics features, morphological MRI features, and clinical features. Patients were stratified into low- and high-risk prognostic groups based on MF-DLM scores, and model interpretability was evaluated using Shapley additive explanations (SHAP) and Gradient-weighted class activation mapping (Grad-CAM). Findings: The median follow-up time for the training, validation, and external test sets are 38.0 months (interquartile ranges [IQR]: 22.0, 62.0), 40.5 months (IQR: 23.0, 71.0), and 38.5 months (IQR: 26.0, 50.0), respectively. The MF-DLM demonstrated excellent performance in predicting OS, achieving higher C-index values than pathological T stage (training: 0.902 vs. 0.793, p < 0.001; validation: 0.864 vs. 0.757, p = 0.014; external test: 0.841 vs. 0.760, p = 0.047). In addition, MF-DLM-based low-risk group demonstrated significantly longer OS in the training, validation, and external test sets (p < 0.001). In the adjuvant therapy (AT) cohort, high-risk patients had significantly worse prognosis compared with low-risk patients (p < 0.0001). Additionally, high-risk pathological T3/4 patients exhibited no statistically significant OS difference between those who received AT and those who did not (p = 0.18), whereas low-risk pathological T3/4 patients experienced significantly improved OS with AT (p = 0.0059). Besides, the low-risk group had better OS than the high-risk group in neoadjuvant therapy cohort (p = 0.0032). Interpretation: The MF-DLM can reliably predict OS in patients with BCa and provide additional prognostic stratification beyond pathological T and N stages. Furthermore, MF-DLM-based risk groups can identify patients most likely to benefit from perioperative therapy. Funding: The Noncommunicated Chronic Diseases-National Science and Technology Major Project (2024ZD0525700); National Natural Science Foundation of China (82273152, 82503879), Jiangsu Province Hospital (the First Affiliated Hospital of Nanjing Medical University) Clinical Capacity Enhancement Project (JSPH-MA-2022-5), China Postdoctoral Science Foundation funded project (2024M761211), and the Nanjing Postdoctoral Science Foundation funded project (2024BHS210).

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.

How this classification was reachedexpand

Full frame distilled prediction

Teacher imitation

Not calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.

metaresearch head score (Codex)0.001
metaresearch head score (Gemma)0.002
Version: codex-gemma-dda1882f352aValidation status: machine_predicted_unvalidated
Candidate categoriesMeta-epidemiology (narrow)
Consensus categoriesnone
DomainCandidate signal: none · Consensus signal: none
Study designCandidate signal: Observational · Consensus signal: none
GenreCandidate signal: Empirical · Consensus signal: Empirical
Teacher disagreement score0.529
Threshold uncertainty score1.000

Codex and Gemma teacher scores by category

CategoryCodexGemma
Metaresearch0.0010.002
Meta-epidemiology (narrow)0.0000.000
Meta-epidemiology (broad)0.0010.000
Bibliometrics0.0000.000
Science and technology studies0.0000.000
Scholarly communication0.0000.000
Open science0.0000.000
Research integrity0.0000.001
Insufficient payload (model declined to judge)0.0000.000

Machine scores (provisional)

The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.

Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.

Opus teacher head0.026
GPT teacher head0.357
Teacher spread0.331 · how far apart the two teachers sit on this one work
Validation statusscore_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from it

Classification

machine, unvalidated

Machine predicted; a candidate call from one teacher head, not a consensus.

Study designObservational
Domainnot available
GenreEmpirical

How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".

Quick stats

Citations3
Published2025
Admission routes1
Has abstractyes

Explore more

Same venueEClinicalMedicineSame topicBladder and Urothelial Cancer TreatmentsFrench-language works237,207