Files
Ghislain_19721700_Godelaine_28631700_2022.pdf
Open access - Adobe PDF
- 18.48 MB
Details
- Supervisors
- Faculty
- Degree label
- Abstract
- Mobile tumors represent a major challenge for the treatment of liver and lung cancers. Radiotherapy and protontherapy are methods to treat these types of cancer. However, for the treatment to be effective, the position of the tumors needs to be known, which is difficult due to the motion induced by the respiration. To take it into account, the most used solution is the ITV-based PTV. In this context, we test to track the tumor on MRI images by first approximating it with an ellipse characterized by five parameters. These parameters are then predicted with a sinusoidal model and a Kalman filter. Segmentation algorithms (Region growing, Canny edge and Snake) are then applied when the MRI image is available to know precisely the contour of the tumor. That allows to determine whether the treatment has been correctly delivered and whether it has to be adapted due to a bad prediction. The precision of prediction and segmentation algorihtms are evaluated with the Dice coefficient. The gain is also determined by calculating the proportion of tumor and the surface of healthy cells irradiated. The principal results indicate that the Kalman filter has the best performances, i.e. the highest Dice coefficient (above 0.7). The best segmentation algorithms are the Canny edge (mean Dice from 0.7 to 0.81) and Snake (with mean Dice from 0.75 to 0.83) when the segmentation algorithm is based on a prediction returned by the Kalman filter. The main conclusion is that there is a gain in terms of surface of healthy cells irradiated when predicting the tumor (and by dilating the ellipse found) compared to the use of an ITV-based PTV treatment. The minimum gain found is around 200 mm² of healthy cells which are not irradiated.