411goruntuanalizi mumcuoglu

TIBBİ GÖRÜNTÜ ANALİZİ : 

TIBBİ GÖRÜNTÜ ANALİZİ Erkan Mumcuoğlu, Yrd. Doç. Dr. Enformatik Enstitüsü Orta Doğu Teknik Üniversitesi

Introduction: 

Introduction Anatomical imaging modalities: Radiography, MRI, CT, US Functional imaging modalities: NM (SPECT/PET), fMRI, EEG, MEG, etc. 100s of slices each  clinical analysis is not easy! (w/o SW) Image Analysis: feature extraction, boundary/edge detection, pattern recognition, segmentation, etc. Medical Image Analysis: registration, subtraction, fusion, segmentation, ROI analysis, visualization, etc.

Well-known Softwares: 

Well-known Softwares SPM Analyze MRIcro AFNI Vista MEDx Nomos

Image Analysis Tool: 

Image Analysis Tool

Preprocessing: 

Preprocessing Free rotation (rotation, scale, translate, shear)

Image Registration: 

Image Registration Information from two image sets are complementary. Proper integration of this information is desired. 1st step: Spatial alignment (Registration) M. Fitzpatrick defines registration as “the determination of a geometrical transformation that aligns points in one view of an object with corresponding points in another view of that object or another object”. 2nd step: Fusion

Image Registration: 

Image Registration The inputs of registration: two image sets (views) The output: a geometrical transformation which is a mathematical mapping from points in one view to points in the second. (The registration is accepted as successful to the extent that corresponding points are mapped together.)

Slide8: 

Top row shows orthogonal slices through the patients MRI scan, middle row shows SPECT scan 'A' (aligned with MRI), lower row shows SPECT scan 'B' (here 'A' and 'B' are displayed with different color scales to help distinguish between the two). The lower row (SPECT can ‘B’) is aligned with the other scans A project on MRI-DUAL-SPECT imaging for epilepsy case, Image Processing and Analysis Group of Departments of Diagnostic Radiology and Electrical Engineering, Yale School of Medicine

Slide9: 

Orthogonal slices through registered MRI (left), SPECT scan 'A' (middle column) and SPECT scan 'B' (right column). The two SPECT scans are displayed with the same colour table and intensity range to illustrate the overall difference in tracer reaching the brain during the two studies Orthogonal slices through registered MRI (left), SPECT scan 'A', SPECT scan 'B' (normalized to 'A') and their difference (displayed with a green-red color table to illustrate increase and decrease respectively). Note: Regions of increase in the temporal lobe and also reconstruction artifacts in the last slice of SPECT 'A'. A project on MRI-DUAL-SPECT imaging for epilepsy case, Image Processing and Analysis Group of Departments of Diagnostic Radiology and Electrical Engineering, Yale School of Medicine

Registration: 

Registration SPM: “Mutual Information” co-registration

Slide11: 

ROI analysis of the change in uptake in each region The regions of change A project on MRI-DUAL-SPECT imaging for epilepsy case, Image Processing and Analysis Group of Departments of Diagnostic Radiology and Electrical Engineering, Yale School of Medicine

Classification of Registration Methods : 

Classification of Registration Methods Dimensionality Nature of registration basis Domain and Nature of transformation Interaction Optimization procedure Modalities involved Subject Object The classification of registration methods described here is based on the criteria formulated by van den Elsen, Pol & and Viergever (1993).

I. Dimensionality: 

I. Dimensionality Spatial dimensions only 2D/2D (e.g. separate slices from tomographic data) 2D/3D (e.g. pre-operative CT image to an intra-operative X-ray image ) 3D/3D (e.g. two tomographic datasets ) Time series (more than two images) with spatial dimensions 2D/2D 2D/3D 3D/3D

II. Nature of registration basis: 

II. Nature of registration basis Extrinsic Invasive Stereotactic frame Fiducials (screw markers) Non-invasive Mould, frame, dental adapter, etc. Fiducials (skin markers) Non-image based (calibrated coordinate systems) Intrinsic Landmark based Anatomical Geometrical Segmentation based Rigid models (points, curves, surfaces) Deformable models (snakes, nets) Intensity (voxel) based Reduction to scalars/vectors (moments, principal axes) Using full image content

III.Domain and Nature of transformation : 

III.Domain and Nature of transformation

IV. Interaction: 

IV. Interaction Interactive Initialization supplied No initialization supplied Semi-automatic User initializing User steering/correcting Both Automatic

V. Optimization procedure: 

V. Optimization procedure Parameters computed Parameters searched for (iteratively)

VI. Modalities involved: 

VI. Modalities involved Monomodal (intra-modal) (MRI, PET, SPECT, US, etc.) Multimodal (inter-modal) (CT-MRI, CT-PET, CT-SPECT, DSA-MRI, etc.) Modality to Model (CT, MRI, SPECT,etc. to Atlas) Modality to Patient (CT, MRI, PET, etc. to Patient)

VII. Subject: 

VII. Subject Intra-subject (single patient) Inter-subject (different patients or patient and a model) Atlas (patient and a constructed image from an image information database)

VIII. Object: 

VIII. Object Head Brain or skull  Eye Dental Thorax Entire Cardiac Breast Abdomen General Kidney Liver Pelvis and perineum Limbs General Femur Humerus Hand Spine and vertebrae

Intrinsic Methods: 

Intrinsic Methods Anatomical landmark based methods: usually labor-intensive accuracy depends on the accurate indication of corresponding landmarks in all modalities Segmentation (surface) based registration: generally highly data and application dependent surfaces not easily identified in functional modalities (e.g.PET) Intensity (voxel) based registration methods: operate directly on the image grey values, without prior data reduction by the user or segmentation. uses all available information throughout the registration process. feature calculation is straightforward or even absent the accuracy of these methods is not limited by segmentation errors.

Intensity-Based Registration Methods: 

Intensity-Based Registration Methods Registration transformation, f, is determined by iteratively optimizing some “similarity measure” calculated from all pixel or voxel values: Sum of Squares of Intensity Difference (SSD) Correlation Coefficient (CC) Ratio-Image Uniformity (RIU) Partitioned Intensity Uniformity (PIU) Joint histograms and joint probability distributions  Joint entropy  Mutual Information (MI) Normalization of Mutual Information (NMI)  Knowledge of a 1-to-1 relationship between the grey value images Voxel grey value p in image A, implies that the corresponding voxel in image B has a grey value q=f (p) and vice versa.

MI Image Registration: 

MI Image Registration MI(A,B’) =  p(a,b’) log[p(a,b’) / p(a)p(b’)] a,b’

Nonrigid 3-D Registration: 

Nonrigid 3-D Registration Non-rigid transformations are important for: applications to nonrigid anatomy, Inter-patient registration of rigid anatomy Intra-patient registration of rigid anatomy when there are nonrigid distortions in the image acquisition procedure.

Interpatient Registration of Rigid Anatomy: 

Interpatient Registration of Rigid Anatomy comparison of brains of different individuals assessment of normal and abnormal anatomical variability between subjects building an electronic brain atlas Multi-subject registration (Courtesy of D. Vandermeulen, KUL)

Intra-patient Registration of Rigid Anatomy When there are Nonrigid Distortions : 

Intra-patient Registration of Rigid Anatomy When there are Nonrigid Distortions Shape changes of the brain (brain shift) during neurosurgery caused by the intervention and physiological changes. Elastic matching applied to MR scan of the brain obtained during neurosurgery. (a) Slice from an early stage of the surgery; (b)Slice after craniotomy; (c) Deformed image; (d) Difference image before alignment; (e) Difference image after alignment.

Image Segmentation: 

Image Segmentation Automatic techniques Semi-automatic techniques Model based techniques Data driven techniques Hybrid techniques (thresholding, region-growing, classifier, clustering, Markov Random Field, deformable models, model fitting (atlas), morhology, etc.)

Segmentation: 

Segmentation Brain Extraction (“BET”) Tissue Classification (white matter, gray m., CSF) Initial step for surface rendering

Visualization: “Marching Cubes” A High Resolution 3D Surface Construction Algorithm: 

Visualization: “Marching Cubes” A High Resolution 3D Surface Construction Algorithm Goal: create a constant density surface from a 3D array of data Idea: create a triangular mesh that will approximate the iso-surface calculate the normals to the surface at each vertex of the triangle. Algorithm: locate the surface in a cube of eight pixels, calculate normals, march to the next cube Afterwards, can map functional SPECT/PET/fMRI data on the surface

Marching Cubes Method: 

Marching Cubes Method

3-D Visualization of SISCOM:(focal activation region in epilepsy): 

3-D Visualization of SISCOM: (focal activation region in epilepsy)