This document summarizes a master's thesis on segmenting the left ventricle boundary in tagged magnetic resonance images. The student developed a two-step algorithm using semantic descriptors of tissue and prior shape knowledge to automatically segment the left ventricle boundary. Results showed high accuracy of the automatic segmentation compared to manual segmentation, with mean errors less than 0.2 degrees for global rotation and 1 degree for regional rotation. This promising algorithm could serve as a support tool for clinical analysis of heart function. Future work further validates the approach on additional image slices and pathological cases.
Automated quantitative assessment of left ventricular functions by MR image s...An-Cheng Chang
Thesis (English):
http://handle.ncl.edu.tw/11296/ndltd/08095156356149825986
General Information:
This is the slides for my MSc thesis defense, which introduces an algorithm that automatically analyzes cardiac magnetic resonance scans for extracting left ventricular cardiac parameters. These parameters are crucial for determining whether or not a patient suffers from certain forms of cardiovascular diseases, such as ischaemic heart disease and hypertrophy.
The performance of the algorithm in terms of segmentation accuracy (APD) outperforms all other similar reported algorithms (as of 2014) that also use the same CMR scan database* by at least 15%. This means more accurate cardiac parameters can be obtained using my proposed algorithm.
*Evaluation is done using the Cardiac MR Database provided by Sunnybrook Health Science Center, Toronto, Canada.
Sparsity Based Spectral Embedding: Application to Multi-Atlas Echocardiograph...Kevin Keraudren
Slides from Ozan Oktay at the MICCAI workshop on Sparsity Techniques in Medical Imaging (STMI2014), presenting one of the methods we used in the CETUS challenge (http://www.creatis.insa-lyon.fr/Challenge/CETUS/index.html).
Marker Controlled Segmentation Technique for Medical applicationRushin Shah
Medical image segmentation is a very important field for the medical science. In medical images, edge detection is an important work for object recognition of the human organs such as brain, heart or kidney etc. and it is an essential pre-processing step in medical image segmentation.
Medical images such as CT, MRI or X-Ray visualizes the various information’s of internal organs which is very important for doctors diagnoses as well as medical teaching, learning and research.
It is a tough job to locate the internal organs if images contains noise or rough structure of human body organs.
Automated quantitative assessment of left ventricular functions by MR image s...An-Cheng Chang
Thesis (English):
http://handle.ncl.edu.tw/11296/ndltd/08095156356149825986
General Information:
This is the slides for my MSc thesis defense, which introduces an algorithm that automatically analyzes cardiac magnetic resonance scans for extracting left ventricular cardiac parameters. These parameters are crucial for determining whether or not a patient suffers from certain forms of cardiovascular diseases, such as ischaemic heart disease and hypertrophy.
The performance of the algorithm in terms of segmentation accuracy (APD) outperforms all other similar reported algorithms (as of 2014) that also use the same CMR scan database* by at least 15%. This means more accurate cardiac parameters can be obtained using my proposed algorithm.
*Evaluation is done using the Cardiac MR Database provided by Sunnybrook Health Science Center, Toronto, Canada.
Sparsity Based Spectral Embedding: Application to Multi-Atlas Echocardiograph...Kevin Keraudren
Slides from Ozan Oktay at the MICCAI workshop on Sparsity Techniques in Medical Imaging (STMI2014), presenting one of the methods we used in the CETUS challenge (http://www.creatis.insa-lyon.fr/Challenge/CETUS/index.html).
Marker Controlled Segmentation Technique for Medical applicationRushin Shah
Medical image segmentation is a very important field for the medical science. In medical images, edge detection is an important work for object recognition of the human organs such as brain, heart or kidney etc. and it is an essential pre-processing step in medical image segmentation.
Medical images such as CT, MRI or X-Ray visualizes the various information’s of internal organs which is very important for doctors diagnoses as well as medical teaching, learning and research.
It is a tough job to locate the internal organs if images contains noise or rough structure of human body organs.
Biomedical Image Processing
Topics covered: Biomedical imaging, Need of image processing in medicine, Principles of image processing, Components of image processing, Application of image processing in different medical imaging systems
Adaptive K-Means Clustering Algorithm for MR Breast Image Segmentation
3D Brain Tumor Segmentation Scheme using K-mean Clustering and Connected Component Labeling Algorithms
Volume Identification and Estimation of MRI Brain Tumor
MRI Breast cancer diagnosis hybrid approach using adaptive Ant-based segmentation and Multilayer Perceptron NN classifier
Biomedical Image Processing
Topics covered: Biomedical imaging, Need of image processing in medicine, Principles of image processing, Components of image processing, Application of image processing in different medical imaging systems
Adaptive K-Means Clustering Algorithm for MR Breast Image Segmentation
3D Brain Tumor Segmentation Scheme using K-mean Clustering and Connected Component Labeling Algorithms
Volume Identification and Estimation of MRI Brain Tumor
MRI Breast cancer diagnosis hybrid approach using adaptive Ant-based segmentation and Multilayer Perceptron NN classifier
1. Master in Computer Vision and Artificial Intelligence LV Contour Segmentation in TMR Images Using Semantic Description of Tissue and Prior Knowledge Correction Student: Albert AndaluzGonzález Advisors: Débora Gil Resina & Jaume Garcia iBarnés
3. Clinical problem ♥ The problem: 30% of global deaths are caused by heart diseases Treatment: diagnose of the heart function Method: Extraction of clinical scores from regional wall motion Requirements: accurate contour estimation ●●○○○○○○○○○○○○○○○○○○○○○○
4. Contour Segmentation Manual: slow Inter/intra observer variability -> requires high expertise. Automatic Objective for all subjects Requires accurate estimation of the LV boundaries ●●●○○○○○○○○○○○○○○○○○○○○○
5. Tagged Magnetic Resonance LV(*) TMR in SA view Long Axis (LA) Base Mid Short Axis (SA) cuts Apex (*)LV = LeftVentricle ●●●●○○○○○○○○○○○○○○○○○○○○
6. Problems in Tagged Magnetic Resonance Blood pool and tissue appearance is similar at time=0 … Tag pattern misleads some common image descriptors… Contours LV Boundaries …and few algorithms for TMR automatic segmentation currently exist ●●●●●○○○○○○○○○○○○○○○○○○○
8. Our Contribution Goal: Algorithm for automatic segmentation for the extraction of clinical regional scores of the LV Semantic definition of the LV Two steps: Shape approximation Shape correction ●●●●●●●○○○○○○○○○○○○○○○○○
9.
10. Internal energy: avoids deviation from anatomical LV shapes●●●●●●●●○○○○○○○○○○○○○○○○
11. External energy computation Define contour for image potential LV Contour extraction Distance map to the LV contours ●●●●●●●●●○○○○○○○○○○○○○○○
13. Energy definition Amplitude 2 Motion module Amplitude1 * * mean mean mean * * * * * * mean * * * * ●●●●●●●●●●●○○○○○○○○○○○○○
14. Contour extraction 2 Detection 3 Selection 1 Clustering from E1|E2 K-Means Contours+ Binary Morphology Region area Filtering Epicardium (outerboundary of the LV) Endocardium (innerboundary of the LV) ●●●●●●●●●●●●○○○○○○○○○○○○
15. Distance map Epicardium Endocardium Distance to LV contours Distance map Vector Field ●●●●●●●●●●●●●○○○○○○○○○○○
16. Shape Correction Converged snake shape is corrected using PCA and GOPA: Incoming shape Mean shape Variation modes PCA eigenvalues then If for Correction applied No correction needed Externalenergy Converged snake Correction ●●●●●●●●●●●●●●○○○○○○○○○○
18. Test set 29 real cases from 15 healthy patients 2 cuts (basal and mid) 2 energies for automatic segmentation TMR sequences from the Clinica La Creu Blanca ●●●●●●●●●●●●●●●●○○○○○○○○
29. < 1º mean regional rotationPromising support tool for clinical analysis ●●●●●●●●●●●●●●●●●●●●●●●○
30. Future work Validate in apical cuts Use Mean shift for improved clustering Validate our method in pathologic patients ●●●●●●●●●●●●●●●●●●●●●●●●
31. “Coronary heart disease is a silent disease and the first manifestation frequently is sudden death.” Dr. Herman K.Hellerstein - Cardiologist 1916- 1993 (Ohio, USA)
Editor's Notes
There are severalreason f
There are severalreason f
There are severalreasonforblablaWewillnow show visuallythegeometry of theheart and theinnergroundsfortheissuespresententher
The vertical plane divides the heart in two parts. Whereas the three horizontal cuts (the sort axis) cretae the base mid and apex cutsThe videos show the evolution at those sections of the myocardium motion.
Classicsnakes are curves thatdeformundertheinfluence of twoenergies. Theexternalenergyattractsthesnakestowardsthe LV contours, whereastheinternalenergyavoids…
Asforthesemanticdescriptors, foreachframe of thetaggedsequence, we compute the amplitude3 of thegaborfiltger in twodirection. Wealso compute themotion module of theimage. Thus, allthreefeatureslackthetaggedgrid
We define twoenergyes. Thefirstisobtained as follows. Wecomptuethe mean of everysequence of imagedescitpros. Thethreemeans are combiendinto a single energypotential, namely E1In thesecondeneryg, foreachframewe combine alltheedescriptorsinto a single one and compute the mean. Weget a new sequence, whichisavergaged , into a global mean image
Theimagepotentialisclusteredby pixel intentsyusingkmenas. Next, we use commoncontourdetectionstodetectthecanditateregions. Finally, weselectedthe externa landinternalboundaries of the LV byusingregionareafiltering.
Foreachboundary, we compute thedistancemaptothecontours. Next, weextractthe vector fieldwhcihatttractsthesnakestowardsthedesiredcontours.
Tovalidateiourmethod, wedefinedtwocritera:Thesegmentation error isthedistancebetwweemourautomawticsegmentation and manual shapesbymedicalexperts.Moreover, oneachboundary, we compute the global and regional rotationforboth cases fortestingtheaccuracy
In vertical, we can seethetwo short axis cuts,. Eachcolumnrepresents a diferentenedy,. The manual and automaticaproximations are shown. Thetablesrefertothewhole set, values show similar behaviour in bouth cases
In the case of a goodaproxiamtion, the manual and automatic curves are veryclose, which shows thatouraproximationbehavessimilarytothe manual (groundtruth) segmentation,. Thisfactisfurtherconfirmedbythe global score, whichiscomputed in thewhole disc.
Whenthereis considerable deviation, the scores diverge in some sector (A, AS). In therestits more orlessthesmaAlkso, thegtlobal scores are afected in themiddle of thesystoliccycle. However, thedifferenceisnotthatgreat
Bothenergies show similar behaviour. In both cases, the mean rotationisbelow 1 degree in both cases forregionalscores. In the CASE of global scores, the mean rotationisbelwo 0.2 degreesw
There are severalreason f
Acceptablerotationvalues showbebelow 1 degreese
Beforeendingthispresentation, I wouldliketoquote a cardiologistfromthestaes. He saidthat