Identifying Biological Subtypes Of Head And Neck Squamous Cell Carcinoma (Hnscc) From Contrast Enhanced Ct Scans Using Radiomic And The Cancer Genome Atlas (Tcga)

Contrast enhanced computed tomography (CT) scans are routinely used for anatomic staging of HNSCC, however at present provide little information regarding biologic subtypes of HNSCC. We sought to identify whether radiomic analyses of pre-therapy staging CT scans could identify biologically relevant information regarding subtypes of HNSCC. We analyzed HNSCC patients from The Cancer Imaging Atlas (TCIA) and correlated imaging features with genomic data from the TCGA. A retrospective analysis of individual patient imaging data and genomic data (n=188) from matched TCIA and TCGA portals respectively was performed. DICOM images of all available pre-treatment CT scans were imported into computational environment for radiation therapy research (CERR), the primary GTV was manually segmented, and contours were reviewed by two experienced neuro-radiologists. Poor quality scans were excluded, leaving a final cohort of 81 patients for analysis. Radiomic features were extracted using CERR and unsupervised hierarchical & consensus clustering were performed. Genomic data was extracted from our previous work and the TCGA data portal. The relationship between resulting radiomic clusters and tumor subsite, gene expression sub-type, intra-tumoral genetic heterogeneity, and immune microenvironment was examined. The cohort consists of tumors from the oral cavity (n=40), larynx (n=28) and oropharynx (n=13). In total, 676 radiomic features were extracted. Consensus clustering on these radiomic features yielded two stable and coherent clusters with 41 and 40 samples in cluster 1 and 2, respectively. The radiomic clusters differed by primary tumor subsite with all oropharyngeal tumors clustering together in cluster 1 (p=0.0002), 65% of oral cavity in cluster 2, and laryngeal evenly distributed. Principal component analysis demonstrated 7 principal components explain 90% of the variance of the radiomic data. Of recurrently mutated genes in HNSCC, there was a significant difference in FAT1 mutation1 between cluster 1 (14.6%; 6/41) and cluster 2 (37.5%; 15/40). Evaluation of gene-expression subtypes of HNSCC demonstrated that atypical immunoreactive (cluster1) and mesenchymal cell motility (cluster 2) are driving the differences between the two radiomic clusters (p=0.055). Analysis of genetic intra-tumoral heterogeneity revealed a strong correlation with two radiomic features (p=0.0035 & p=0.0054 respectively; Spearman test). Evaluation of the immune micro-environment and radiomic features is ongoing and will be presented at the meeting. Radiomic analyses of pre-therapy CT scans of HNSCC patients can help identify biologic information about the underlying malignancy including gene-expression subtypes and degree of intra-tumoral heterogeneity. In the emerging era of molecularly guided therapy, further development of radiomic biomarkers may be able to identify relevant molecular subtypes to guide therapeutic selection.