MRI based unsuperviced clustering on MIBC reveals intratumor heterogeneity phenotypes and neoadjuvant chemotherapy efficacy

Abstract

Background

To develop and validate a non-invasive imaging-based intratumor heterogeneity (IITH) framework for risk stratification in muscle-invasive bladder cancer (MIBC) and to investigate its association with survival outcomes and molecular subtypes.

Methods

In this retrospective multicenter study, 437 MIBC patients were enrolled (237 patients of discovery cohort and 102 patients of validation cohort predominantly underwent radical cystectomy (RC); 98 patients of NAC cohort received three to four cycles of gemcitabine-cisplatin chemotherapy prior to surgery). From pretreatment MRI scans, 42 radiomic features were extracted. Unsupervised consensus clustering was performed using the “ConsensusClusterPlus” R package, a resampling-based method which generates a consensus matrix to robustly stratify cohorts into stable IITH subtypes. Prognostic significance for OS, CSS, RFS, PFS was then evaluated based on data reviewed up to July 30, 2024. Molecular correlates were assessed via immunohistochemistry and transcriptomic data from The Cancer Genome Atlas.

Results

The IITH framework significantly stratified patient survival in the discovery cohort (OS: HR = 1.646, p < 0.05; CSS: HR = 1.792, p < 0.05; RFS: HR = 2.06, p < 0.01; PFS: HR = 1.721, p = 0.01), with consistent results in external validation cohorts (OS: HR = 2.604, p < 0.05; RFS: HR = 2.422, p < 0.05; PFS: HR = 2.72, p < 0.05). High-IITH tumors demonstrated greater histological diversity and enriched mixed subtypes. In the NAC cohort, high IITH was associated with markedly worse outcomes (CSS: HR = 7.795, p < 0.05; RFS: HR = 9.529, p < 0.001; PFS: HR = 8.893, p < 0.001). Transcriptomic analysis revealed elevated inflammatory activity and enhanced proliferative/migratory potential in high-IITH groups.

Conclusions

The MRI-based IITH framework effectively stratified MIBC patients by prognosis and treatment response, with high heterogeneity being associated with adverse outcomes and aggressive molecular tumor biology.

Supplementary Information

The online version contains supplementary material available at 10.1186/s40644-026-01007-x.

Introduction

Intratumor heterogeneity (ITH) has been widely documented across various malignancies, significantly influencing both diagnosis and treatment strategies [1–3]. The need to characterize tumor heterogeneity in clinical practice stems from its central role in driving therapeutic resistance. Within neoadjuvant chemotherapy (NAC), where pathological response serves as a surrogate endpoint, quantifying this heterogeneity can help tailor therapy according to individual risk [4, 5].

Muscle-invasive bladder cancer (MIBC) displays pronounced heterogeneity. Genomic instability and a high tumor mutational burden contribute to its diverse clinical behavior and variable responses to chemotherapy [6–8]. While interpatient and intertumoral heterogeneity are recognized contributors, ITH itself is not well characterized in bladder cancer. Therefore, the potential clinical utility of therapeutic strategies directed by ITH warrants further and more systematic study.

Methodologically, conventional approaches for detecting ITH primarily rely on whole-genome sequencing or histopathological analysis. These techniques are invasive and may not comprehensively capture heterogeneity across multiple tumor regions [9]. This underscores the urgent need for non-invasive, imaging-based methods to evaluate ITH across the whole tumor. MRI, with its superior soft-tissue contrast resolution over CT, is well-suited for this purpose as it enables the quantitative extraction of radiomic features [10–13]. Notably, some radiomic features that depict gray-level distribution and texture variation patterns correlate strongly with tumor heterogeneity and hold significant prognostic value in oncology [14, 15]. More broadly, radiomics as a technique enables the non-invasive, whole-tumor assessment of intratumoral heterogeneity [16–18]. Its integration with molecular profiles - including histopathology and gene expression patterns - provides biological context for radiomic features [19].

Accordingly, this study sought to develop an imaging-based ITH (IITH) framework for the molecular stratification of MIBC patients, and to delineate its clinical and biological relevance.

Methods

The cohort

We conducted a multicenter retrospective study of 437 MIBC patients, with the discovery cohort (n = 237) and NAC cohort (n = 98) from Jiangsu Province People’s Hospital. Validation cohorts (n = 102) were obtained from multiple centers. All patients in the discovery cohort and validation cohort predominantly underwent RC, while a minority who desired bladder preservation received transurethral resection of bladder tumor (TURBT) followed by adjuvant therapy. The NAC cohort received three to four cycles of gemcitabine-cisplatin chemotherapy prior to RC or TURBT. Forty-six NAC patients also received combined immunotherapy. We additionally analyzed TCGA data to explore transcriptomic-imaging correlations. All enrolled patients were followed up until July 30, 2024.

MRI segmentation and feature extraction

All MRI scans were acquired with patients in supine position. Participants followed standardized bladder preparation: complete voiding 2 h pre-scan, followed by fluid restriction and no further urination (detailed parameters in Table S1). All images underwent standardized preprocessing, including N4 bias field correction using the RIAS software, to mitigate potential batch effects arising from different scanner platforms and acquisition parameters. Two experienced radiologists manually segmented tumors, defining regions of interest (ROIs) extending 5 mm beyond visible margins while excluding bladder lumen to capture peritumoral regions. Using PyRadiomics, we extracted three feature classes: (a) wavelet, (b) first-order, and (c) texture features [20]. The reliability of the extracted features was quantitatively evaluated by calculating the intraclass correlation coefficient (ICC) for both intra- and inter-observer reproducibility.

Heterogeneity-related radiomic features

Radiomic features demonstrating the strongest association with imaging heterogeneity were identified as heterogeneity-related features based on PyRadiomics criteria. First-order features were selected based on their ability to quantify intensity value distributions, while textural features were chosen according to their capacity to characterize tissue pattern heterogeneity. Ultimately, 42 radiomic features (9 first-order and 33 texture features) were classified as heterogeneity-related (Table S2).

Imaging heterogeneity clustering

To optimize clustering performance, all features were normalized prior to applying the median absolute deviation (MAD) algorithm for feature selection. Features demonstrating MAD values > 1 were retained for subsequent unsupervised clustering analysis. Unsupervised clustering was performed using ConsensusClusterPlus (R package) on MAD-filtered features [21]. Cluster numbers (k) ranging from 2 to 5 were evaluated, with optimal k determined by maximal consensus matrix stability and clearest cluster segregation. According to their definitions, most radiomic features were positively associated with high IITH, whereas a minority showed an inverse relationship, corresponding to low IITH. The IITH score was computed as: Σ(Z-score normalized positive features) - Σ(Z-score normalized negative features), with higher scores indicating greater imaging heterogeneity.

Prognostic significance of the imaging subtypes

The prognostic value of imaging subtypes was initially assessed in the discovery cohort for overall survival (OS), cancer-specific survival (CSS), recurrence-free survival (RFS), and progression-free survival (PFS). Subsequently, the prognostic significance of these subtypes was validated using independent multicenter cohorts. A multivariate Cox proportional hazards model was constructed incorporating both imaging subtypes (derived from unsupervised clustering) and relevant clinical covariates.

IHC and RNA-seq data analysis

IHC analysis was performed on a subset of the discovery cohort, with positive staining defined as > 25% tumor cell expression for quality control purposes. RNA sequencing data for 15 TCGA-BLCA samples were obtained from the UCSC Xena browser (https://xena.ucsc.edu). DESeq2 was used to analyze the differentially expressed genes(|log2FC|>1,adjP < 0.05)between the two IITH groups. Gene Set Enrichment Analysis (GSEA) was employed to identify significantly enriched biological pathways associated with IITH stratification in the TCGA cohort. The protein-protein interaction (PPI) network was constructed using data extracted from the STRING database (https://cn.string-db.org/) and visualized/analyzed through Cytoscape software (v3.10.0). Molecular subtyping was performed using the BLCAsubtyping (https://github.com/cit-bioinfo/BLCAsubtyping).

Statistical analysis

Statistical analyses and model construction were performed using R software (version 4.3.0; http://www.R-project.org). A two-sided p-value < 0.05 was considered statistically significant. The normality of distribution was evaluated using the Kolmogorov-Smirnov (K-S) test. Normally distributed continuous variables are expressed as mean ± standard deviation (SD), while non-normally distributed variables were compared using the Mann-Whitney U test. Categorical variables were analyzed using the chi-square test.

Results

Overview of study design and cohort

This four-phase study (Fig. 1) established IITH subtypes through multicenter MRI cohorts (Phase I-II), validated their prognostic utility across treatment settings (Phase III), and linked IITH to histological heterogeneity via integrated IHC/RNA-seq profiling (Phase IV). We compiled a multicenter radiomics dataset consisting of three independent cohorts: (1) discovery cohort (2), validation cohort, and (3) NAC cohort. For all cohorts, first-order and texture features were extracted from both tumor and peritumoral ROIs. The discovery cohort comprised 237 patients with complete clinical, pathological, and follow-up data; IHC slides were available for 67 of these cases (Table 1). This cohort served three aims: (1) to identify IITH and its prognostic associations; (2) to compare pathological characteristics across IITH-based clusters; and (3) to validate correlations between IITH and histological heterogeneity. The validation cohort included 102 patients with complete pathological and follow-up data, among whom RNA sequencing data were available for a subset (n = 15). This cohort was used to validate the prognostic value of IITH and to perform transcriptomic comparisons across IITH clusters. Finally, the NAC cohort (n = 98) included complete data on treatment response, pathology, and follow-up, and was employed to assess the potential of IITH for predicting response to neoadjuvant chemotherapy (Table 2).

Fig. 1.

Flowchart of this study

Table 1.

Clinicopathological characteristics of the discovery and validation cohorts

Characteristics	Discovery cohort	Validation Cohort (Outercenter+16TCGA)
		Outercenter
n	237	86
Gender (%)
Male	199 (84.0)	70 (81.4)
Female	38 (16.0)	16 (18.6)
Age (mean (SD))	69.29 (9.72)	66.95 (11.01)
VI.RADS score (%)
1	1 (0.4)	1 (1.2)
2	12 (5.1)	7 (8.1)
3	64 (27.0)	16 (18.6)
4	76 (32.1)	20 (23.3)
5	84 (35.4)	42 (48.8)
VUC (%)
No	170 (71.7)	63 (73.3)
Yes	67 (28.3)	23 (26.7)
Pathological garde (%)
Low	4 (1.7)	2 (2.3)
High	233 (98.3)	84 (97.7)
pT.stage (%)
T2	116 (48.9)	41 (47.7)
T3	88 (37.1)	38 (44.2)
T4	33 (13.9)	7 (8.1)
pN.stage (%)
N0	177 (74.7)	67 (77.9)
N1	60 (25.3)	19 (22.1)
CIS (%)
No	219 (92.4)	72 (83.7)
Yes	18 (7.6)	14 (16.3)
Lymphovascular invasion (%)
No	161 (67.9)	53 (61.6)
Yes	76 (32.1)	33 (38.4)
Surgery(%)
TURBT	17(7.2)	8(9.3)
RC	220(92.8)	78(90.7)

Abbreviations: VUC: variant urothelial carcinoma; pT.stage: pathological T stage; pN.stage: pathological N stage; CIS: crinoma in situ; pCR: pathological complete response; pDS: pathological downstaging; TURBT: transurethral resection of bladder tumor; RC: radical cystectomy

Table 2.

Clinicopathological characteristics of the NAC cohort

Characteristics	NAC cohort
n	98
Age (mean (SD))	64.47 (9.54)
pCR (%)
No	73 (74.5)
Yes	25 (25.5)
pDS (%)
No	40 (40.8)
Yes	58 (59.2)
Primary tumor (%)
No	21 (21.4)
Yes	77 (78.6)
Immutherapy combination (%)
No	52 (53.1)
Yes	46 (46.9)
Surgery(%)
TURBT	18(18.4)
RC	80(81.6)

Abbreviations: pCR: pathological complete response; pDS: pathological downstaging; TURBT: transurethral resection of bladder tumor; RC: radical cystectomy

Identification of IITH

Our initial objective was to develop a radiomic signature capturing imaging intratumor heterogeneity for population clustering. On the basis of the arithmetic formula and explicit definitions of all extracted radiomic features, we extracted 42 heterogeneity related first-order and textual radiomic features to assess the IITH of distinct samples (Table S2). Thus, by including the wavelet-based features, a total of 559 features were initially included. In the discovery cohort, the median absolute deviation (MAD) was calculated to assess the variance in radiomic features among samples, and features with MAD > 1 were regarded as highly variable heterogeneity-related radiomic features. A total of 85 features, comprising 18 first-order and 67 textural features, were ultimately selected for clustering (Table S3). Subsequently, utilizing Consensus Clustering, we clustered the discovery cohort based on these features and extrapolated this process to validation cohort and NAC cohort (Fig. 2A and S1). The IITH score for each sample was calculated using the formula: Σ(Z-score of pro-heterogeneity features) - Σ(Z-score of anti-heterogeneity features). Correspondingly, a greater IITH indicated greater imaging heterogeneity. Using this framework, MIBC patients were effectively stratified by heterogeneity patterns. One cluster with significantly higher IITH scores was designated the high-IITH group, and the other as the low-IITH group (Fig. 2D). As expected, high-IITH clusters showed significantly elevated expression of pro-heterogeneity features compared to low-IITH clusters (Fig. 2B).

Fig. 2.

Identification and prognostic stratification capability of IITH. (A) Consensus matrix showing unsupervised clustering of heterogeneity-related radiomic features. (B) Heatmap showing heterogeneity-related radiomic features between high IITH cluster and low IITH cluster. (C) Representative MRI demonstrating high versus low IITH patients. (D) Comparative analysis of IITH scores after clustering in discovery cohort. (E to H) Kaplan-Meier survival curves stratified by IITH clusters for OS, CSS, RFS, and PFS in discovery cohort. (I) Comparative analysis of IITH scores after clustering in validation cohort. (J to M) Kaplan-Meier survival curves stratified by IITH clusters for OS, CSS, RFS, and PFS in validation cohort. (N) Comparative analysis of IITH scores after clustering in NAC cohort. (O to T) Kaplan-Meier survival curves stratified by IITH clusters for OS, CSS, RFS, and PFS in NAC cohort

Exploration of the prognostic stratification value of IITH in the MIBC cohort

Extended follow-up revealed a significant prognostic association between the two IITH clusters. In the discovery cohort, patients in the high-IITH group had significantly worse outcomes across all endpoints: OS(HR = 1.646; 95%CI: 1.046–2.59), CSS(HR = 1.792; 95%CI: 1.008–3.184), RFS(HR = 2.06; 95%CI: 1.182–3.587), PFS(HR = 1.721; 95%CI: 1.133–2.615) (Fig. 2, E to H). A multivariate Cox proportional hazards models revealed that, among numerous clinical and pathological characteristics, high IITH cluster serves as an independent predictor of adverse prognosis (Figure S1O). This prognostic value was externally validated in the validation cohort. Consistent with the discovery phase, patients stratified into the high-IITH subgroup showed significantly inferior OS (HR = 2.604; 95%CI: 1.205–5.627), RFS (HR = 2.422; 95%CI: 1.031–5.689), and PFS (HR = 2.72; 95%CI: 1.184–6.249) compared to those with low IITH (Fig. 2, J to M). Collectively, these results establish IITH as a robust and independent prognostic biomarker for stratifying MIBC patients across multiple endpoints and cohorts.

The implications of IITH for neoadjuvant chemotherapy

Neoadjuvant chemotherapy followed by radical cystectomy constitutes the standard therapeutic regimen for MIBC. We evaluated the clinical utility of IITH stratification in NAC-treated patients. Using our predefined heterogeneity-related radiomic features, we calculated IITH scores to classify the NAC cohort (Fig. 2N). Patients with high IITH exhibited significantly worse CSS (HR = 7.795, 95% CI: 0.935–64.96), RFS (HR = 9.529, 95% CI: 2.106–43.12), and PFS (HR = 8.893, 95% CI: 1.945–40.67) (Fig. 2, O to T). This association remained significant in multivariate analysis, establishing IITH as an independent prognostic factor (Figure S2A). The results imply that IITH may help identify NAC beneficiaries, with low-IITH patients gaining more survival benefit from therapy. Strikingly, despite similar pathological response rates between clusters (Figure S2, J and K), non-responders within the high-IITH cluster had especially poor outcomes. Consequently, low IITH appears predictive of a more favorable prognosis, especially in the absence of a complete pathological response (Figure S2, L to N).

Similarly, we assessed the stratification efficacy of IITH in the NAC-immunotherapy combination cohort. Although no significant differences in prognostic outcomes were observed between subgroups, the overall trend aligned with previous research findings (Figure S2, B to E). Finally, in the discovery cohort, patients undergoing adjuvant chemotherapy (AC) were selected, demonstrating that IITH maintained excellent stratification efficacy within the AC-treated patients (Figure S2, F to I).

Correlation of IITH with pathological features and histological subtypes

To further investigate the relationship between imaging heterogeneity and the intrinsic histological characteristics of tumors, we initially analyzed the differential pathological manifestations between the two clusters stratified by IITH. Compared to the low IITH cluster, the high IITH cluster exhibited significantly higher pathological T stages and pathological N stages (Fig. 3, A and B). These findings support a potential imaging-histology correlation, where macroscopic heterogeneity may reflect underlying microscopic diversity in tumor composition. To further characterize this relationship, we performed IHC profiling for CD20, GATA3, CK5/6, and CK14 in a representative subset (n = 67) from the discovery cohort (Fig. 3C). Utilizing these four markers, we categorized the tissues into four distinct subtypes: luminal subtype, basal subtype, mixed subtype, and double-negative subtype [22–24]. The high IITH cluster displayed a significantly higher prevalence of the mixed subtype, while the low IITH cluster was predominantly characterized by the luminal subtype (Fig. 3D). Additionally, the high IITH cluster exhibited a markedly increased proportion of cases with double positivity for CK5/6 and GATA3 (Fig. 3E). This finding indicates that MIBC tumors demonstrating high radiological heterogeneity present a more complex and heterogeneous histological composition. A heterogeneous histological composition is often associated with poorer therapeutic efficacy and prognosis, which is consistent with previous findings.

Fig. 3.

Association between IITH and histopathological heterogeneity. (A and B) Comparative analysis of pathological characteristics (including T stage, N stage) between IITH clusters. (C) Representative IHC staining sections demonstrating high versus low IITH patients. (D) IHC heatmap demonstrating the expression profiles of molecular subtype biomarkers in MIBC. (E) Comparative analysis of the proportion of mixed subtype across IITH clusters

Transcriptomic alterations associated with IITH

To investigate the biological basis of the prognostic disparity between high and low IITH tumors, we performed RNA - seq analysis to uncover their molecular distinction. In the validation cohort, we extracted the RNA - seq data of samples from TCGA and performed differential expression analysis according to the IITH cluster. When setting |log₂FC| > 1 and padj < 0.05, a total of 44 differentially expressed genes were selected, among which 15 genes were up - regulated in the high IITH cluster and 29 genes were down - regulated in it (Fig. 4A). Subsequently, Protein-protein interaction (PPI) network analysis revealed core interacting differentially expressed genes, including PROM1, CEACAM family members, and COL22A1, potentially associated with the enhanced proliferative and migratory capacities observed in high IITH tumors. (Fig. 4D). Then, we used gene set enrichment analysis (GSEA) to explore the enriched pathways in high IITH including cancer hallmark pathways and KEGG pathways (Fig. 4, B and C). In the KEGG pathways, pathways related to substance metabolism, such as Ascorbate and aldarate metabolism, Cholesterol metabolism, Porphyrin metabolism, and pathways related to inflammatory responses, such as NOD - like receptor signaling pathway, TNF signaling pathway, Toll - like receptor signaling pathway, are up - regulated in the high IITH cluster. Furthermore, among the cancer hallmark pathways, INTERFERON_ALPHA_RESPONSE, INTERFERON_GAMMA_RESPONSE, and TNFA_SIGNALING_VIA_NFKB are up - regulated in the high IITH cluster, while KRAS_SIGNALING_DN is down - regulated therein (Fig. 4, E to J).

Fig. 4.

Exploration of the biological mechanisms underlying IITH. (A) Heatmap showing genes differentially expressed across IITH clusters in the TCGA dataset. (B) Bar plot showing the top enriched pathways ordered by the -log10p-value in the high-IITH group (red) and low-IITH group (blue) in the TCGA dataset. (C) Dot plot showing the activated and suppressed pathways in the high-IITH cluster. (D) PPI network of genes differentially expressed across IITH clusters. (E to J) GSEA plot demonstrating several pathways activated in the high-IITH cluster

Overall, via the transcriptome analysis of the two IITH clusters, we discovered that the high IITH cluster exhibits a higher level of inflammatory activation and more vigorous substance - metabolism activities. This might indicate that the intratumoral immune environment within tumors with high - imaging heterogeneity is also more intricate and more reliant on active substance - metabolism to supply the materials necessary for proliferation.

Discussion

ITH is a critical driver of therapeutic resistance and poor clinical outcomes. To this end, we developed a novel method termed IITH to assess ITH in bladder cancer using MRI. We demonstrated its robust capability for prognostic stratification and its correlation with histological heterogeneity. Furthermore, our findings indicate that IITH can function as an adjunctive biomarker to guide therapeutic decisions and evaluate treatment response in patients receiving neoadjuvant chemotherapy. Finally, by integrating transcriptomic data, we explored the biological mechanisms underlying IITH.

A refined subset of radiomic features for quantifying intratumoral heterogeneity was constructed based on PyRadiomics definitions [20]. After selecting features with high inter-sample variability, consensus clustering was employed to stratify samples into high- and low-heterogeneity subgroups. This classification proved prognostically significant, as the high-heterogeneity subgroup was associated with inferior outcomes across all survival endpoints (OS, CSS, RFS, PFS)—a result replicated in multicenter validation cohorts. To advance this field, future work could combine optimized textural features with tumor habitat analysis for more robust ITH assessment [25].

The clinical applicability of IITH was further investigated through pathological correlation and IHC in a discovery cohort subset. High IITH scores were associated with elevated heterogeneity in both histopathology and molecular subtypes, establishing a direct link between imaging phenotypes and tumor biology. These traditional methods, however, are constrained by two principal limitations: sampling bias due to regional analysis and invasiveness that limits clinical feasibility. Radiomics circumvents these issues by allowing non-invasive, quantitative evaluation of heterogeneity across the entire tumor [26].

Given the distinct histopathology between IITH subgroups, we hypothesized differential responses to NAC. Our exploratory analysis indicates that the IITH framework may offer prognostic stratification for NAC-treated MIBC patients. While its utility was limited in complete response (CR) patients—a subgroup with a favorable prognosis—it provided clearer stratification among non-CR patients. This may be due to the greater biological heterogeneity typically found in non-responding tumors, which could make them more discernible by imaging-based heterogeneity metrics. In line with this, our data suggest that even without a pathological complete response, low-IITH patients may still obtain a survival benefit from NAC. These preliminary findings highlight the need for validation in larger, prospective studies to determine IITH’s potential for refining patient stratification.

Subsequently, we explored the biological underpinnings of IITH. Phenotypically, high-IITH tumors were associated with larger dimensions and irregular, heterogeneous morphology, radiomorphologically suggestive of aggressive potential. Exploratory transcriptomic profiling in a subset indicated heightened anabolic and metabolic activity in these tumors, potentially reflecting bioenergetic adaptation for growth. Heterogeneity-enriched patterns correlated with more aggressive biology and adverse outcomes. Intriguingly, high-IITH tumors displayed immunologically complex tumor microenvironments, exhibiting hyperactivation of multiple inflammatory pathways compared to low-IITH tumors [27]. This chronic inflammatory milieu potentially enabled tumor cells to resist apoptosis and immune-mediated cytotoxicity [28]. While these analyses provide initial insights, the correlative findings, particularly from the transcriptomic data generated from a relatively small sample subset, require validation in larger cohorts.

Our study is subject to certain limitations. Although IITH was validated in large multicenter cohorts for robust quantification and stratification, the retrospective design ultimately requires prospective confirmation of clinical utility. However, rigorous radiomic feature extraction and data fidelity necessitate high-resolution MRI acquisitions with standardized imaging protocols. Finally, the transcriptomic correlation was exploratory, based on a limited TCGA subset; thus, the biological interpretations need verification in larger prospective cohorts.

Conclusions

In summary, this multicenter radiogenomic study established an MRI-based method for quantifying IITH that effectively stratified patients by prognosis and therapy response. By integrating histopathological, immunohistochemical, and transcriptomic analyses, we demonstrated bidirectional alignment between radiologically defined heterogeneity and molecular tumor behavior, thus bridging macroscopic imaging phenotypes with microscale oncobiological mechanisms.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

AC

Adjuvant Chemotherapy

BLCA

Bladder Urothelial Carcinoma

CI

Confidence Interval

CIS

Crinoma In Situ

CR

Complete Response

CSS

Cancer-Specific Survival

CT

Computed Tomography

DS

Downstaging

FC

Fold Change

GSEA

Gene Set Enrichment Analysis

HR

Hazard Ratio

IHC

Immunohistochemistry

IITH

Imaging-based Intratumor Heterogeneity

ITH

Intratumor Heterogeneity

MAD

Median Absolute Deviation

MIBC

Muscle-Invasive Bladder Cancer

MRI

Magnetic Resonance Imaging

NAC

Neoadjuvant Chemotherapy

OS

Overall Survival

PFS

Progression-Free Survival

PPI

Protein-Protein Interaction

RC

Radical Cystectomy

RFS

Recurrence-Free Survival

RNA-seq

RNA Sequencing

ROI

Region of Interest

SD

Standard Deviation

TCGA

The Cancer Genome Atlas

TURBT

Transurethral Resection of Bladder Tumor

VUC

Variant Urothelial Carcinoma

Author contributions

Yuheng Zhang and Lingkai Cai were involved in the study conceptualization, design, data analysis, and in drafting and revising the manuscript. Yuhan Chen was responsible for the acquisition and processing of IHC data. Rongjie Bai, Chenghao Wang, Zhengye Tan, Chang Chen, Xiaotong Liu, Bo Liang, Meihua Jiang, Gongcheng Wang, Qiang Shao, Xuping Jiang, and Meiling Bao contributed to the collection and curation of multi-center data. Hao Yu, Qiang Cao, Xiao Yang, and Qiang Lu provided supervision and overall guidance for the research project.

Funding

This work was supported by Noncommunicated Chronic Diseases-National Science and Technology Major Project (2024ZD0525700), Jiangsu Province Hospital (the First Affiliated Hospital of Nanjing Medical University) Clinical Capacity Enhancement Project (JSPH-MA-2022-5), National Natural Science Foundation of China (grants 82273152), China Postdoctoral Science Foundation funded project (2024M761211), and Nanjing Postdoctoral Science Foundation funded project (2024BHS210).

Data availability

The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

This study was approved by the ethics committee of the First Affiliated Hospital with Nanjing Medical University.

Consent for publication

Informed consent for publication was obtained from all participants (or their parents/legal guardians in the case of children) in this study.

Competing interests

The authors declare no competing interests.