Interplay between MRI radiomics and immune gene expression signatures in oral squamous cell carcinoma

Abstract

With the advances in immunotherapy and the challenge of poor responsiveness in oral squamous cell carcinoma (OSCC) patients, understanding the tumor microenvironment is crucial. Radiogenomics offers the potential to provide pre-operative, non-invasive image-derived immune biomarkers. To this aim, the present study explores the capability of MRI-based radiomics to describe patients' immune state in OSCC. Seven MRI-based radiomic, 29 immune-related gene expression signatures were computed and deconvolution analysis was performed for a subset of OSCC from the BD2Decide database. A correlation-driven analysis identified key associations between radiomic and immune-related signatures and cell populations. Radiomic classifiers of the gene expression signatures were then developed to evaluate their capability to stratify patients based on immune status. MRI-based radiomic models showed promising results in predicting a gene expression signature associated with significant prognostic value for HNSCC patients who underwent radiotherapy (AUC = 0.92), suggesting these models' potential in distinguishing radioresistant from radiosensitive patients, aiding treatment decisions. Additionally, radiomic signatures reflected immune infiltrating cells in our cohort (M1, CD8 + T, B cells). MRI-radiomic signatures and associated models could become non-invasive methods to evaluate the prognosis and treatment choice in OSCC patients. Based on our promising results, and upon external validation, MRI-radiomics could enhance personalized medicine approaches.

Keywords: Cluster analysis; Head and neck cancer; Magnetic resonance imaging; Radiogenomic; Radiomic features; Transcriptomic signatures; Tumor microenvironment.

Fig. 1

Consort.

Fig. 2

Correlation-driven clustering and dendrogram between the 7 radiomic (R) signatures and the selected 15 gene expression (G) signatures. Only stable clusters were depicted. Spearman’s correlation coefficient (ρ) was computed between each R and G signature. The main characteristics of R and G signatures are respectively reported in Tables 2 and 3.

Fig. 3

Paired scatter plots between the 7 radiomic signatures (R1-R7) and correlation matrix heatmap of Spearman’s correlation coefficient (ρ).

Fig. 4

Paired scatter plots between the 15 gene expression (G) signatures and correlation matrix heatmap of Spearman’s correlation coefficient (ρ).

Fig. 5

Median balanced accuracy and area under the receiver operating characteristic curve (AUC) over 100 repetitions of the single radiomic models (top) and combined radiomic model (bottom) of each gene expression signature included in the study.

Fig. 6

Correlation between radiomic signatures (R1-R7) and immune populations identified by Cibersort in the 82 OSCC patients. Spearman’s correlation coefficient (ρ) was computed.