Development of a Molecular-Subtype-Associated Immune Prognostic Signature That Can Be Recognized by MRI Radiomics Features in Bladder Cancer

Abstract

Background: Bladder cancer (BLCA) is highly heterogeneous with distinct molecular subtypes. This research aimed to investigate the heterogeneity of different molecular subtypes from a tumor microenvironment perspective and develop a molecular-subtype-associated immune prognostic signature that can be recognized by MRI radiomics features. Methods: Individuals with BLCA in The Cancer Genome Atlas (TCGA) and IMvigor210 were classified into luminal and basal subtypes according to the UNC classification. The proportions of tumor-infiltrating immune cells (TIICs) were examined using The Cell Type Identification by Estimating Relative Subsets of RNA Transcripts algorithm. Immune-linked genes that were expressed differentially between luminal and basal subtypes and associated with prognosis were selected to develop the immune prognostic signature (IPS) and utilized for the classification of the selected individuals into low- and high-risk groups. Functional enrichment analysis (GSEA) was performed on the IPS. The data from RNA-sequencing and MRI images of 111 BLCA samples in our center were utilized to construct a least absolute shrinkage and selection operator (LASSO) model for the prediction of patients’ IPSs. Results: Half of the TIICs showed differential distributions between the luminal and basal subtypes. IPS was highly associated with molecular subtypes, critical immune checkpoint gene expression, prognoses, and immunotherapy response. The prognostic value of the IPS was further verified through several validation data sets (GSE32894, GSE31684, GSE13507, and GSE48277) and meta-analysis. GSEA revealed that some oncogenic pathways were co-enriched in the group at high risk. A novel performance of a LASSO model developed as per ten radiomics features was achieved in terms of IPS prediction in both the validation (area under the curve (AUC): 0.810) and the training (AUC: 0.839) sets. Conclusions: Dysregulation of TIICs contributed to the heterogeneity between the luminal and basal subtypes. The IPS can facilitate molecular subtyping, prognostic evaluation, and personalized immunotherapy. A LASSO model developed as per the MRI radiomics features can predict the IPSs of affected individuals.

MRI-Based Radiomics in Bladder Cancer: A Systematic Review and Radiomics Quality Score Assessment

“…A total of 1092 studies were initially identified through the literature search. After removing duplicates and screening for eligibility, 26 studies were finally included in the analysis [ 12 , 13 , 14 , 15 , 16 , 17 , 18 , 19 , 20 , 21 , 22 , 23 , 24 , 25 , 26 , 27 , 28 , 29 , 30 , 31 , 32 , 33 , 34 , 35 , 36 , 37 ]. Figure 1 depicts the PRISMA flowchart.…”

“…Three studies used radiomics features to preoperatively identify genetic and immunologic markers of BCa, including Ki-67, CD-8A, and a constructed immune prognostic signature (IPS) consisting of five immune-related genes [ 29 , 34 , 37 ]. For the prediction of Ki-67, the data set was divided into a high Ki-67 expression group (>15% cells stained) and a low Ki-67 expression group (≤15%).…”

“…In the paper by Zheng et al [ 34 ], the authors demonstrated the potential of radiomics features to preoperatively predict the expression of CD8A in both the training (AUC = 0.857) and validation sets (AUC: 0.844). Liu et al [ 37 ] developed a radiomic model which efficiently predicted the five gene-based IPS in both the training (AUC = 0.839) and validation sets (AUC = 0.819). All of these three papers developed Rad-Score using features derived from T2WI and DCE images and used LASSO algorithm as an ML classifier.…”

“…In 23 articles, multiple segmentation was performed by two or three radiologists. However, the intraclass correlation coefficient was employed only in 14 studies [ 15 , 19 , 21 , 23 , 26 , 27 , 28 , 29 , 30 , 32 , 33 , 34 , 35 , 37 ] in order to select the most robust radiomics features, while in the remaining articles, a consensus between radiologists was made regarding the final segmented volume.…”

A computed tomography urography‐based machine learning model for predicting preoperative pathological grade of upper urinary tract urothelial carcinoma

“…The formula was as follows: where, x i is the normalized value of each selected feature and Coe f i is the coefficient corresponding to each selected feature. 19 …”