Development and validation of a pathomics model for predicting CXCL8 expression and prognosis with machine learning in head and neck cancer.

PURPOSE: It is imperative to investigate a method to indicate prognosis and to seek new biological markers for personalized medicine for head and neck squamous cell carcinoma (HNSCC) patients. Pathomics based on quantitative medical imaging analysis has emerged lately. CXCL8, a crucial inflammatory cytokine, is reflected in overall survival (OS). This study aimed to explore associations of the CXCL8 mRNA expression with pathom ics features and investigate the underlying biology of CXCL8.

METHODS: Clinical information and TPM(transcripts per million) mRNA sequencing data were obtained from the TCGA TCGA- HNSCC data set. The correlation between CXCL8 mRNA expression and patients' survival rate was identified using the Kaplan Kaplan-Meier survival curve. We retrospectively analyzed 313 samples diagnosed with HNSCC in the TCGA database. Pathomics features were extracted from HE images, then using mRMR_RFE method and screened by the Logistic Regression(LR) algorithm.

RESULTS: According to Kaplan Kaplan-Meier curves, the high expression of CXCL8 had a significant association with OS deterioration. The LR pathomics model integrated 16 radiomics features identified by the mRMR_RFE method in the training set and performed well in the testing set. The calibration plots revealed that the high gene expression probability predicted by the pathomics model had a decent agreement with actual observation. And the model has high clinical practicability.

CONCLUSION: The pathomics model reflected by CXCL8 mRNA expression is an effective tool for predicting the prognosis in HNSCC patients and can assist clinical decisiondecision-making. High CXCL8 expression may contribute to decreased DNA damage and associate with the pro pro-inflammatory microenvironment, providing a potential target for tumor therapy.