Session Item

Quantitative imaging analysis aims to exploit the large amounts of data available in routine medical images, allowing for opportunities to identify biomarkers through non-invasive means.  Combining patient data along with radiomic features extracted from these images may support clinical decisions and enhance personalised care in radiation oncology.

For patients with cancers of the head and neck, CT and MRI based studies have demonstrated the value of radiomics in predicting histological grade and HPV status of the primary tumour, as well as distinguishing nodal metastases.  Knowing these tumour characteristics are associated with poorer clinical outcomes, it is natural researchers have explored the value of radiomics in predicting survival, local/locoregional control, and disease progression.  In the majority of these studies, the integration of radiomics improved the performance of existing models, thereby highlighting its potential in risk stratification for patients with head and neck cancer.  

Beyond the tumour volume, radiomics has also been used to predict the likelihood and severity of toxicities associated with radiation therapy to the head and neck region.  Despite advances in treatment planning and delivery normal tissue damage is inevitable, and some patients are at a higher risk of developing debilitating side effects due to inherent personal characteristics.  Isolating imaging biomarkers which correlate with adverse events may assist in identifying patients most likely to need additional supportive care.  Head and neck radiomics research in this regard has focused primarily on predicting xerostomia; with results consistently highlighting the discriminative power of these imaging biomarkers.  Some researchers have also attempted to identify imaging biomarkers indicative of trismus and hearing loss; however, as with many radiomics studies, additional research with larger patient cohorts and external validation is needed.


Whilst the clinical value of radiomics is easy to appreciate, there are various technical and methodological aspects within the radiomics pipeline which influence feature values and the generalizability of results.  Image acquisition and reconstruction protocols vary amongst institutions, and these factors have been shown to influence texture feature values. Likewise, observer variability in delineation of the region of interest has also shown to impact on feature reproducibility. Additionally, spatial resampling and intensity discretization also impacts results, and greater understanding of optimal settings are needed.  Studies which interrogate these aspects of radiomics methodology are primarily, but not exclusively, either phantom studies or focus on lung tumours.  Translating conclusions from these studies to patients with head and neck cancer is not appropriate, and more research is needed to address nuances specific to this cancer site.  Furthermore, models incorporating radiomics need to be sufficiently powered and should include external validation to identify possible overfitting.


The application of radiomics to the management of patients with head and neck cancer continues to grow, as does the quantity and quality of studies.  Identifying robust and informative imaging biomarkers is exciting, but greater standardisation of methodology is needed to realise their full potential.