Radiomics-based model for prediction of TGF-β1 expression in head and neck squamous cell carcinoma.

Am J Nucl Med Mol Imaging 2024;14(4):239-252 www.ajnmmi.us /ISSN:2160-8407/ajnmmi0154301 Original Article Radiomics-based model for prediction of TGF-β1 expression in head and neck squamous cell carcinoma Kai Qin1*, Chen Gong1*, Yi Cheng1, Li Li2, Chengxia Liu2, Feng Yang1, Jie Rao1, Qianxia Li1 Department of Oncology, Tongji Hospital, Tongji Medical College, Huazhong University of Science and Technology, Wuhan 430030, Hubei, China; 2Department of Radiology, Tongji Hospital, Tongji Medical College, Huazhong University of Science and Technology, Wuhan 430030, Hubei, China. *Equal contributors. 1 Received November 7, 2023; Accepted August 8, 2024; Epub August 25, 2024; Published August 30, 2024 Abstract: Objective: To explore the connection between TGF-β1 expression and the survival of patients with head and neck squamous cell carcinoma (HNSCC), as well as whether non-invasive CT-based Radiomics can predict TGF-β1 expression in HNSCC patients. Methods: Data on transcriptional profiling and clinical information were acquired from the TCGA database and subsequently categorized based on the TGF-β1 expression cutoff value. Based on the completeness of enhanced arterial phase CT scans, 139 HNSCC patients were selected. The PyRadiomics package was used to extract radiomic features, and the 3D Slicer software was used for image segmentation. Using the mRMR_RFE and Repeat LASSO algorithms, the optimal features for establishing the corresponding gradient enhancement prediction models were identified. Results: A survival analysis was performed on 483 patients, who were divided into two groups based on the TGF-β1 expression cut-off. The Kaplan-Meier curve indicated that TGF-β1 was a significant independent risk factor that reduced patient survival. To construct gradient enhancement prediction models, we used the mRMR_RFE algorithm and the Repeat_LASSO algorithm to obtain two features (glrlm and ngtdm) and three radiation features (glrlm, first order_10percentile, and gldm). In both the training and validation cohorts, the two established models demonstrated strong predictive potential. Furthermore, there was no statistically significant difference in the calibration curve, DCA diagram, or AUC values between the mRMR_RFE_GBM model and the LASSO_GBM model, suggesting that both models fit well. Conclusion: Based on these findings, TGF-β1 was shown to be significantly associated with a poor prognosis and to be a potential risk factor for HNSCC. Furthermore, by employing the mRMR_RFE_GBM and Repeat_LASSO_GBM models, we were able to effectively predict TGF-β1 expression levels in HNSCC through non-invasive CT-based Radiomics. Keywords: TCGA, TCIA, head and neck squamous cell carcinoma, Radiomics, TGF-β1 expression, prediction model Introduction Head and neck squamous cell carcinoma (HNSCC) is a malignant tumor with a high incidence that develops in the mucous epithelium of the mouth, pharynx, and larynx. The standard treatment for HNSCC is currently a combination of surgery and chemoradiotherapy in clinics; however, patients’ survival rates within 5 years are still unsatisfactory, only reaching up to 34% [1-3]. Traditional prognostic indicators of HNSCC, such as clinical stages, p16, human papillomavirus (HPV) status, and Programmed cell death ligand 1 (PD-L1) expression [4], can no longer meet the clinical needs of precision medicine. As a result, more research is required to identify new prognostic indicators for personalized stratified patient care. Due to the unsatisfactory outcomes of the standard treatment options, immunotherapy has been observed as another therapeutic strategy and has been widely used for the treatment of HNSCC. However, only 15-20% of patients have benefited from this treatment [5], highlighting the need to investigate immune-resistant mechanisms in the immune microenvironment and provide evidence for HNSCC treatment. The transforming growth factor beta (TGF-β) family is a key immunosuppressive gene in HNSCC and is linked to a poor prognosis [6]. TGF-β signal dysfunction promotes tumor progression and metastasis by regulating epithelial cell proliferation, inhibiting cell apoptosis, and inducing genomic instability of tumor cells [7]. PD-L1 is a protein that is crucial to the regulation of the immune system. Similar to TGF-β, PD-L1 is highly expressed in patients with HNSCC. PD-L1 can inhibit lymphocyte activation and induce apoptosis by binding to the PD-1 receptor on lymphocytes’ surfaces, allowing tumor cells to escape the immune system. TGF-β promotes tumorigenesis and contributes to drug resistance against PD-L1 monoclonal antibodies. Blocking both PD-L1 and TGF-β signals can enhance a synergistic anti-tumor effect and response rate of PD-1 inhibitors [8]. TGF-β1 is highly expressed in the majority of HNSCCs [9], however, whether it is an independent factor that can predict the survival of HNSCC patients remains unclear. Radiomics is a high-throughput “image sequencing” that can acquire a large number of image parameters by noninvasive, dynamic, and quantitative detection of tumor features [10]. Radiomics has been regarded as an effective technology in HNSCC for guiding early diagnosis and classification, assessing tumor heterogeneity, and identifying cell constituents in the tumor microenvironment [11]. In addition, radiomics has a high potential for overcoming the limitations of traditional tumor markers [12], https://doi.org/10.62347/JMKV7596 Radiomics prediction model of TGF-β1 expression in HNSCC Figure 1. The scheme of patients’ selection (A) and radiomics modeling (B). TCGA: The Cancer Genome Atlas, TCIA: The Cancer Imaging Archive, HNSCC: Head and Neck squamous cell carcinoma, ICC: the intraclass correlation coefficient, GBM: Gradient Boosting Machine, mMRM: maximum relevance minimum redundancy, RFE: recursive feature elimination. because it provides complete three-dimensional information about tumors and allows for non-invasive repetitive analysis using follow-up images. In this study, we used CT-based radiomics to determine the prognostic value of TGF-β1 in HNSCC. We also investigated its potential molecular mechanism and relationship with immune cell constituents using integrated bioinformatics analysis. Methods data and images are anonymous and public, so they are exempt from ethics and informed consent once approved by the unit ethics committee. TCIA images were used to identify radiomic features and establish models, whereas TCGA data was used for prognosis analysis. Sample selection criteria include preoperative samples, complete clinical data (survival time greater than 30 days), transcriptome sequencing data, arterial phase enhanced CT images, and TCGA-TCIA intersection data. Figure 1A shows a brief flow chart. Sources of data & images Analysis of TGF-β1 expression and patients’ survival Transcriptome profiling data and clinical information of 528 HNSCC patients were collected from the TCGA database (https://portal.gdc.cancer.g (...truncated)