Head and neck cancer is the seventh most common cancer worldwide. Radiotherapy is often the preferred treatment because it helps preserve nearby organs. Treatment planning for radiotherapy relies heavily on multi-modality medical images like CT, PET and MRI, creating a vast amount of information with significant potential for further analysis. Recently, Artificial Intelligence (AI) has shown promise in analyzing these complex datasets, which could lead to a more effective treatment and better outcomes.
This thesis of Alessia de Biase developed deep learning (DL) techniques for two key tasks: tumor contouring and prediction of tumor-related endpoints. For tumor contouring, DL methods were developed to assist radiation oncologists in accurately identifying head and neck tumors on PET/CT. Tumor probability maps were introduced to visually show how confident the DL model is in its predictions at the voxel level. An interactive user interface was designed to integrate these tools into clinical practice, and its effectiveness was tested by radiation oncologists.
Furthermore, the thesis explored predicting tumor-related endpoints using DL. By analyzing clinical data and imaging features from scans taken before and during treatment, models were developed to predict outcomes (local control, regional control, distant metastasis-free survival, overall survival, and disease-free survival). The goal was to identify factors that could help tailor treatment strategies and improve patient outcomes.
Overall, this research highlights the transformative potential of AI in radiotherapy. It demonstrates how advanced computational methods can enhance the analysis of medical images to optimize radiotherapy, ultimately leading to more personalized treatment for head and neck cancer patients.
