Respiratory Motion Compensation in Emission Tomography
Compensation du mouvement respiratoire en tomographie d'émission
Résumé
This thesis work deals with the problem of respiratory motion correction in emission tomography imaging.
It has been proven that respiratory motion renders blurred reconstructed images, affecting lesions detection, diagnosis, treatment planning and following of lung cancer. While current motion correction methodologies are based on external breathing tracking devices or specific data acquisition modes. The proposed approach was designed to work without any external tracking devices, which occur on institutions not having access to such material or in cases where the data was already acquired and no tracking device was present at the moment of its capture. The proposed method presents a retrospective scheme of motion correction based on a motion model plugged to the image reconstruction step. The model takes into account displacements and elastic deformations of emission elements (voxels), which allows to consider the non-rigid deformations produced in the thorax during respiration. Furthermore, the chosen voxel modeling improves computations, outperforming classical methods of voxel/detector-tube.
The lack of specific patient respiratory information, two estimation models were investigated and developed. A first simplified model consists in adapting a known respiratory motion model, obtained from a single subject, to the patient anatomy. The initial known model describes by means of a displacement vector field, the lungs deformations produced between extremal respiratory states. This displacement vector field is further adapted by means of an affine transformation to the patient's anatomy, yielding a displacement vector field that matches the thoracic cavity of the patient. The second method deals with the possible lack of robustness caused by the fact of using a single subject when constructing the known displacement vector field of the simplified method. Incorporation of subject variability into a statistical respiratory motion model was developed. The statistical study served as well to highlight the main deformation modes of the breathing lungs.
The whole methodology was developed under a 3-D image reconstruction framework. The algorithm was parallelized and acceleration schemes are presented as well.
Simulations and phantom experiences were carried out. For the first, the SimSET library (Simulation System for Emission Tomography) was used along with the NCAT phantom, upon which a real respiratory motion was incorporated. For phantom experiences, the methodology
was tested against translational movements applied within the data acquisition. For both, simulations and phantom experiences, the results obtained show the ability of the proposed method to correct and compensate the effects of motion during data acquisition. For patient data, the methodology was tested against a dataset composed by five patients with lung cancer.
Although no ground truth was available, preliminary results on patient data are encouraging since improvements in contrast recovery and signal to noise ratios were found on each case.
It has been proven that respiratory motion renders blurred reconstructed images, affecting lesions detection, diagnosis, treatment planning and following of lung cancer. While current motion correction methodologies are based on external breathing tracking devices or specific data acquisition modes. The proposed approach was designed to work without any external tracking devices, which occur on institutions not having access to such material or in cases where the data was already acquired and no tracking device was present at the moment of its capture. The proposed method presents a retrospective scheme of motion correction based on a motion model plugged to the image reconstruction step. The model takes into account displacements and elastic deformations of emission elements (voxels), which allows to consider the non-rigid deformations produced in the thorax during respiration. Furthermore, the chosen voxel modeling improves computations, outperforming classical methods of voxel/detector-tube.
The lack of specific patient respiratory information, two estimation models were investigated and developed. A first simplified model consists in adapting a known respiratory motion model, obtained from a single subject, to the patient anatomy. The initial known model describes by means of a displacement vector field, the lungs deformations produced between extremal respiratory states. This displacement vector field is further adapted by means of an affine transformation to the patient's anatomy, yielding a displacement vector field that matches the thoracic cavity of the patient. The second method deals with the possible lack of robustness caused by the fact of using a single subject when constructing the known displacement vector field of the simplified method. Incorporation of subject variability into a statistical respiratory motion model was developed. The statistical study served as well to highlight the main deformation modes of the breathing lungs.
The whole methodology was developed under a 3-D image reconstruction framework. The algorithm was parallelized and acceleration schemes are presented as well.
Simulations and phantom experiences were carried out. For the first, the SimSET library (Simulation System for Emission Tomography) was used along with the NCAT phantom, upon which a real respiratory motion was incorporated. For phantom experiences, the methodology
was tested against translational movements applied within the data acquisition. For both, simulations and phantom experiences, the results obtained show the ability of the proposed method to correct and compensate the effects of motion during data acquisition. For patient data, the methodology was tested against a dataset composed by five patients with lung cancer.
Although no ground truth was available, preliminary results on patient data are encouraging since improvements in contrast recovery and signal to noise ratios were found on each case.
L'objectif de cette thèse sont les corrections liées aux problèmes des mouvements respiratoires en tomographie d'émission. Il a été prouve que les mouvements respiratoires produisent des images floues, ce qui affecte la détection des lésions, les diagnostics, les traitements, etc. La solution proposée a été conçue pour opérer sans aucun dispositif externe. Cette méthode présente un schéma de la correction du mouvement basée sur un modèle inclus dans la reconstruction d'image. Le modèle prend en compte les déplacements et déformations des éléments d'émissions (voxels), lequel permet de considérer les déformations non rigides produites dans le thorax pendant la respiration. De plus, le model de voxel choisit, permet une amélioration aux calculs par rapport aux méthodes classiques. Deux models d'estimation etaitent développes. Un premier model simplifie consiste a adapter un model de respiration connu sur l'anatomie du patient. Le model initial décrit a travers un champ de déplacement les déformations du poumon produit entre les états de respiration extreme.Ce champ de déplacement est ensuite adapte sur l'anatomie du patient. La deuxième méthode a été conçu pour prendre en compte le manque de robustesse provoque par l'utilisation d'un seul sujet quand on construit les champs de déplacement connus. Incorporation de la variation des sujets dans un model statistique de respiration a été développe. La méthodologie a été développe dans un cadre de reconstruction d'image 3D et a été teste avec des données simules et réels.
Domaines
Autre [cs.OH]
Loading...