Clinical Scorecard: Radiomics Model Utilizing CT Imaging to Forecast Disease Progression and Enhance Clinical Relevance in Locally Advanced Head and Neck Cancer

At a Glance

Key Highlights

• Over 60% of new HNSCC cases present as locally advanced disease with significant recurrence despite optimal therapy.
• Radiomics extracts quantitative imaging features from CT scans to improve prognostic stratification beyond TNM8 and HPV status.
• SHAP methodology enhances interpretability of machine learning models by explaining feature contributions to individual patient risk.

Guideline-Based Recommendations

Diagnosis

• Use contrast-enhanced CT imaging for detailed tumour delineation and staging in LAHNSCC.
• Determine HPV status via p16 staining and HPV DNA genotype analysis to inform prognosis.
• Apply TNM8 staging system for initial clinical stratification.

Management

• Standard definitive treatment includes high-dose cisplatin with concurrent radiotherapy or radiotherapy with cetuximab for unfit patients.
• Multidisciplinary tumor board decisions guide treatment selection considering tumour location, stage, age, and comorbidities.

Monitoring & Follow-up

• Perform baseline CT scans within 28 days before treatment initiation for radiomics analysis.
• Monitor progression-free survival through clinical follow-up and imaging for locoregional recurrence or distant metastases.

Risks

• High recurrence rates remain despite optimal therapy in LAHNSCC.
• Visual interpretation of CT scans post-treatment is challenging due to difficulty distinguishing inflammation/fibrosis from active disease.

Patient & Prescribing Data

LAHNSCC patients undergoing definitive concurrent radiation with cisplatin or cetuximab

Radiomics models integrating clinical and biological data may improve risk stratification and personalize treatment decisions beyond current staging and HPV status.

Clinical Best Practices

• Use standardized contrast-enhanced CT imaging protocols with consistent slice thickness for radiomics feature extraction.
• Employ expert radiologist review and manual 3D tumour segmentation to ensure accurate volume of interest delineation.
• Incorporate machine learning interpretability tools such as SHAP to provide transparent, patient-specific prognostic information.
• Adhere to transparent reporting guidelines for multivariable prediction models to enhance reproducibility and clinical applicability.

References

• Global prevalence of HNSCC
• Risk factors: tobacco and alcohol
• HPV infection and HNSCC
• HPV status determination methods
• TNM8 staging system
• Standard treatment guidelines for LAHNSCC
• Radiomics in head and neck cancer
• SHAP interpretability method
• TRIPOD guidelines for prediction models
• Image Biomarker Standardisation Initiative (IBSI)