Efficient image registration for the analysis of different phases of - PowerPoint PPT Presentation

Efficient image registration for the analysis of different phases of contrast-enhanced liver CT data R.Verdú, J. Larrey, J.Morales F. López, V. Naranjo , M. Alcañiz, R.López, Universitat Politècnica de València Universidad Politécnica de Cartagena. Unidad de Cirugía y Trasplante Hepático, Hospital La Fe, Valencia.

02 Variational Image Registration Conclusions Introduction Outline Results

03 Variational Image Registration Conclusions Introduction Outline Results

04 1. Introduction 1. Introducció 2. Variational Image Registration 3. Results 4. Conclusions Introduction � Hepatocellular carcinoma is responsible for a large proportion of liver cancers. � HepaPlan is a research and development project which is intended to assist the physician in � the diagnosis, � plan the treatment, � track the evolution of the pathology over time. � HepaPlan � quantitatively classify the diseases with the highest degree of objectivity � a 3-D view of the liver structure

05 1. Introduction 1. Introducció 2. Variational Image Registration 3. Results 4. Conclusions Introduction

06 1. Introduction 1. Introducció 2. Variational Image Registration 3. Results 4. Conclusions Introduction

07 1. Introduction 1. Introducció 2. Variational Image Registration 3. Results 4. Conclusions Introduction

08 1. Introduction 1. Introducció 2. Variational Image Registration 3. Results 4. Conclusions Introduction

09 1. Introduction 1. Introducció 2. Variational Image Registration 3. Results 4. Conclusions Introduction

010 1. Introduction 1. Introducció 2. Variational Image Registration 3. Results 4. Conclusions Introduction

011 1. Introduction 1. Introducció 2. Variational Image Registration 3. Results 4. Conclusions Introduction � Tri-phase 3D CT under contrast agent injection. � Adquisition at different times depending on the arrival time of the contrast agent. � Three studies � Non contrast phase. � Arterial phase. � Venous phase.

012 1. Introduction 1. Introducció 2. Variational Image Registration 3. Results 4. Conclusions Introduction � No exact correspondence between the studies. � Non Rigid Deformation method in the frequency domain � At least two times faster than other approaches in the spatial domain

Variational Image Registration Conclusions Introduction Outline Results

014 1. Introduction 1. Introducció 2.Variational Image Registration 2. Variational Image Registration 3. Results 4. Conclusions Variational Image Registration � Registration : correspondence between two images, template T and reference R . Initial problem: arterial and portal phase WITHOUT registration (checkerboard). Representation of [- 50,350] H.U.

015 1. Introduction 1. Introducció 2.Variational Image Registration 2. Variational Image Registration 3. Results 4. Conclusions Variational Image Registration � To minimize the joint energy functional � measures the distance between the deformed template dataset and the reference dataset: correlation ratio. � is a regularizer and determines the smoothness of the displacement field: diffusion term. � weights the influence of the regularization.

016 1. Introduction 1. Introducció 2.Variational Image Registration 2. Variational Image Registration 3. Results 4. Conclusions Variational Image Registration � Parseval’s theorem � Minimization: Euler-Lagrange equation

017 1. Introduction 1. Introducció 2.Variational Image Registration 2. Variational Image Registration 3. Results 4. Conclusions Variational Image Registration � Iteration scheme iterations

018 Variational Image Registration Conclusions Introduction Outline Results

019 1. Introducci ón 2. Variational Image Registration 3. Results 4. Conclusions 3. Results Results � Three experiments: portal-arterial phase, arterial-non Contrast phase and portal-portal phase for registration. � Registration parameters:

020 1. Introducci ón 2. Variational Image Registration 3. Results 4. Conclusions 3. Results Results � Three experiments: portal-arterial phase, arterial-non Contrast phase and portal-portal phase for registration. � Registration parameters: EXPERIMENT #1 • Arterial phase and portal phase of the same patient (256 X 256 X 75 voxels) • 30-60 seconds between the two acquisitions

021 1. Introducci ón 2. Variational Image Registration 3. Results 4. Conclusions 3. Results Results � Three experiments: portal-arterial phase, arterial-non Contrast phase and portal-portal phase for registration. � Registration parameters: EXPERIMENT #2 • non-contrast phase and arterial phase of the same patient (256 x 256 x 79 voxels) • several minutes between the two acquisitions

022 1. Introducci ón 2. Variational Image Registration 3. Results 4. Conclusions 3. Results Results � Three experiments: portal-arterial phase, arterial-non Contrast phase and portal-portal phase for registration. � Registration parameters: EXPERIMENT #3 • two arterial phases of the same patient (256 x 256 x 67 voxels) • several weeks between the two acquisitions

023 1. Introducci ón 2. Variational Image Registration 3. Results 4. Conclusions 3. Results Results Without registration With registration

024 Variational Image Registration Conclusions Introduction Outline Results

025 1. Introducci ón 2. Variational Image Registration 3. Results 4. Conclusions 4. Conclusions Conclusions � Efficient implementation of variational image registration of liver CT data. � The method is based on an efficient implementation of diffusion registration in frequency domain. � If results are analysed, method achieves the ability and the accuracy to estimate the deformation existing in these 3D acquisitions (PSNR > 27 dB and CR>95%).

LabHuman 2012 – Thank you for your attention Valery Naranjo LabHuman - Human Centered Technology Universitat Politècnica de València vnaranjo@labhuman.i3bh.es www.labhuman.com