Response prediction with artificial intelligence in inoperable head and neck cancer patients treated with curative radiotherapy

In this study, we investigated the potential of artificial intelligence (AI) models to predict treatment response in inoperable head and neck cancer (HNC) patients treated with curative radiotherapy (RT). Radiomic features extracted from pre-treatment planning computed tomography (CT) images, together with dosiomic features derived from RT dose distributions and relevant clinical parameters, were analyzed using AI-based methods. A total of 129 patients who underwent curative RT for HNC between September 2015 and February 2024 were retrospectively evaluated. Treatment response was assessed 3–6 months after RT using CT, MRI, and PET-CT. Patients showing complete or partial response were classified as responders, while those with stable or progressive disease were considered non-responders. Data were randomly divided into training (70%) and test (30%) sets. Model performance was evaluated using accuracy, precision, recall, and the F1-score, representing the harmonic mean of precision and recall. Radiomic, dosiomic, and clinical features associated with treatment response were analyzed using 11 AI algorithms based on machine learning (ML) and deep learning (DL) approaches. All algorithms showed strong predictive performance in both training and test sets. Feature importance ranking was performed with the XGBoost algorithm, and the most significant parameters were compared with previous studies. In conclusion, radiomic and dosiomic features derived from planning CT images, when combined with clinical data, can effectively predict treatment response using AI models. Validation of these results in larger, multicenter studies is needed to confirm their clinical utility.