2.2. Patient selection

The study used 82 CT scans to train and validate the model. The training set consisted of 64 CT scans, of which 53 were from Charité-Universitätsmedizin Berlin, Germany, and 11 were from MVision AI Oy, Helsinki, Finland, in compliance with the General Data Protection Regulation. The validation set included 18 scans, 15 from Charité and three from MVision AI Oy. After the development phase, the model underwent testing on an additional set of 20 independent CT scans obtained exclusively from Charité. The scans from Charité were from patients diagnosed with head and neck cancer and undergoing (chemo)radiation therapy. The group comprised 72 % males and 28 % females, with an average age of 66.8 years. The scans were collected between January 2013 and October 2020 using Siemens Somatom Sensation Open scanners. The median slice thickness was 2 mm (range 1–3 mm), and metal artifact reduction was not applied. Scans provided by MVision included five males, two females, and four individuals of unspecified gender, with specific age information available for three individuals: 42, 50, and 82 years. Siemens Sensation Open and Siemens Somatom Definition AS scanners were used for imaging, with a slice thickness ranging from 2 to 3 mm. Imaging data was missing for three cases. All scans had a slice size of 512 x 512 pixels, except for one which had a size of 256 x 256 pixels. CT scans with osseous destruction of the jawbone and HNC patients with a tumor site in the oral cavity were excluded.