Radiomics-based machine learning models decipher heterogeneity of anoikis-related molecular subtypes for prognostic stratification in head and neck squamous cell carcinoma.
Authors
Affiliations (2)
Affiliations (2)
- Department of Medical Ultrasound, Maoming People's Hospital, Maoming, China.
- Radiology Imaging Center, The Affiliated Hospital of Guangdong Medical University, Zhanjiang, China.
Abstract
Head and neck squamous cell carcinoma (HNSC) exhibits high molecular heterogeneity and frequent metastasis, which is strongly linked to anoikis dysregulation. This study aimed to investigate whether radiomics could offer a novel perspective for assessing the heterogeneity and risk stratification of HNSC associated with anoikis-related genes (ANRGs). A total of 118 HNSC patients with paired gene expression and imaging data were included. Radiomics features were extracted from computed tomography (CT) images using PyRadiomics. By employing consensus clustering methods, the heterogeneity of ANRGs-related HNSC was identified into different subtypes, and multi-machine learning model combinations were constructed based on radiomics features to predict the molecular subtypes. An ANRGs-related radiomics prognostic model was constructed to assess the association between subtypes and clinical outcomes. Three subtypes were identified via consensus clustering of 31 prognostic ANRG expression profiles: C1, C2, and C3. Compared to other subtypes, C1 exhibited significantly worse overall survival and distinct immune infiltration levels. Of the multi-combination machine learning models, 71% demonstrated robust generalization in predicting C1 [areas under the curve (AUC) >0.75). The least absolute shrinkage and selection operator-random forest combined model achieved optimal performance, with an AUC of 0.86. The ANRGs-associated radiomics prognostic model showed significant prognostic value and was closely linked to pathways involving enhanced tumor invasiveness and immunosuppressive microenvironment. This study demonstrated that radiomics enables stratified prediction of ANRGs-associated HNSC heterogeneity while providing molecular annotation and prognostic evaluation for ANRGs-related radiomics features, highlighting its translational value in precision oncology.