Soutenance de thèse : Mélanie MOUCHET

Breathing motion detection in computed tomography using data consistency conditions

Doctorante : Mélanie MOUCHET

Laboratoire INSA : CREATIS
Ecole doctorale : EDA160 : Électronique, Électrotechnique, Automatique

Four-dimensional (4D) computed tomography (CT) images are used for the radiother- apy planning of thoracic and upper-abdominal tumors to account for breathing motion. It assumes that the breathing cycle is periodic which is often not the case, especially for patients suffering from respiratory diseases, resulting in artifacts in tomographic images which may impact the radiotherapy. The aim of this thesis is to detect motion in helical CT using data consistency conditions (DCC). DCC are mathematical equations characterizing the redundancy in the projection data. They have been developed for several source and detector configurations but the literature is scarce on helical CT with a cylindrical detector which is the most used geometry in 4D CT. The first contribution is the development of two approaches to apply DCC to pairs of cone-beam projections. The first approach resamples two projections onto a virtual detector which rows are parallel  to the line connecting the source positions to apply a fan-beam DCC along each row. The second approach computes DCC in the cylindrical detector coordinates to increase the number of DCC in a helical acquisition. The second contribution studies the ability of DCC to locally detect motion. A consistency metric is introduced to account for the acquisition noise in the DCC by computing the variance of the DCC. Lastly, a graph approach is proposed to detect mo- tion during an acquisition, in which the vertices are the projections and two vertices are connected if a DCC exists between the projections. All contributions are evaluated using simulated and real data. In both cases, DCC allow the detection of motion for most pairs, but not those for which DCC are heavily impacted by noise. Constructing the graph with pairs of projections robust to noise, the graph approach provides an overview of motion for an entire helical CT.