Construction of an enhanced computed tomography radiomics model for non-invasively predicting granzyme A in head and neck squamous cell carcinoma by machine learning.

PURPOSE

Classical prognostic indicators of head and neck squamous cell carcinoma (HNSCC) can no longer meet the clinical needs of precision medicine. This study aimed to establish a radiomics model to predict Granzyme A (GZMA) expression in patients with HNSCC.

METHODS

We downloaded transcriptomic data of HNSCC patients from The Cancer Genome Atlas for prognosis analysis and then used corresponding enhanced computed tomography (CT) images from The Cancer Imaging Archive for feature extraction and model construction. We explored the influence of differences in GZMA expression on signaling pathways and analyzed the potential molecular mechanism and its relationship with immune cell infiltration. Subsequently, non-invasive CT radiomics models were established to predict the expression of GZMA mRNA and evaluate the correlation with the radiomics-score (Rad-score), related genes, and prognosis.

RESULTS

We found that GZMA was highly expressed in tumor tissues, and high GZMA expression was a protective factor for overall survival. The degree of B and T lymphocyte and natural killer cell infiltration was significantly correlated with GZMA expression. The receiver operating characteristic curve showed that the Relief GBM and RFE_GBM radiomics models had good predictive ability, and there were significant differences in the Rad-score distribution between the high- and low-GZMA-expression groups.

CONCLUSIONS

GZMA expression can significantly affect the prognosis of patients with HNSCC. Enhanced CT radiomics models can effectively predict the expression of GZMA mRNA.