Riassunto analitico
Deep learning (DL)-based automated contouring offers a promising tool to support radiotherapy, significantly reducing the time and variability associated with manual delineation of target volumes and organs at risk (OARs). However, clinical validation of these systems is essential to ensure their reliability. This thesis aims to validate the performance of DL-based automated contouring systems by comparing them with expert manual contouring.
The study utilizes a dataset of radiotherapy planning images and applies deep learning models trained for anatomical segmentation. The evaluation is carried out using quantitative metrics, such as the Dice Similarity Coefficient (DSC) and Hausdorff Distance, alongside qualitative review by radiation oncologists. Furthermore, the impact of automated contours on treatment plan quality and dosimetry is analyzed to assess potential effects on clinical outcomes.
Results demonstrate a high degree of agreement between automated and manual contours, though some anatomical structures present greater challenges.
|
