DATA Visualization

DATA Visualization-ppt Download

  • Date:10 Jul 2020
  • Views:17
  • Downloads:0
  • Size:7.22 MB

Share Presentation : DATA Visualization

Download and Preview : DATA Visualization

Report CopyRight/DMCA Form For : DATA Visualization


Transcription:

MEDICAL IMAGING INFORMATICS Lecture 6SegmentationNorbert SchuffProfessor of Radiology.
VA Medical Center and UCSFNorbert schuff ucsf eduMedical Imaging Informatics 2009 N Schuff UCSF VACourse 170 03 Department ofSlide 1 67 Radiology Biomedical Imaging.
Overview Definitions Role of Segmentation Segmentation methods Intensity based.
Shape based Texture based Summary Conclusion LiteratureMedical Imaging Informatics 2009 N Schuff UCSF VA.
Course 170 03 Department ofSlide 2 67 Radiology Biomedical Imaging The Concept Of SegmentationIdentify classes features that characterize this image Intensity Bright dark.
Shape Squares spheres trianglesTexture homogeneous speckledConnectivity Isolated connectedTopology Closed openMedical Imaging Informatics 2009 N Schuff UCSF VA.
Course 170 03 Department ofSlide 3 67 Radiology Biomedical Imaging More On The Concept Of SegmentationCan you still identify multiple classes in each image Medical Imaging Informatics 2009 N Schuff UCSF VA.
Course 170 03 Department ofSlide 4 67 Radiology Biomedical Imaging Segmentation Of ScenesSegment this scene Hint Use color composition and spatial features.
By J Chen and T Pappas 2006 SPIE DOI 10 1117 2 1200602 0016Medical Imaging Informatics 2009 N Schuff UCSF VACourse 170 03 Department ofSlide 5 67 Radiology Biomedical Imaging Examples Intensity and Texture.
Gray mattersegmentationBy intensitySegmentation of abdominal CT scanBy texture.
image at www ablesw com 3d doctor 3dseg... Stephen Cameron Oxford U Computing LaboratoryMedical Imaging Informatics 2009 N Schuff UCSF VACourse 170 03 Department ofSlide 6 67 Radiology Biomedical Imaging Definitions.
Segmentation is the partitioning of an image intoregions that are homogeneous with respect to somecharacteristics In medical context Segmentation is the delineation of anatomical.
structures and other regions of interest i e lesions Medical Imaging Informatics 2009 N Schuff UCSF VACourse 170 03 Department ofSlide 7 67 Radiology Biomedical Imaging Formal Definition.
If the domain of an image is then the segmentationproblem is to determine sets classes Zk whose unionrepresent the entire domain Z kSets are connected .
Z k Z j k jMedical Imaging Informatics 2009 N Schuff UCSF VACourse 170 03 Department ofSlide 8 67 Radiology Biomedical Imaging More Definitions.
When the constraint of connected regions is removed then determining the sets Zk is termed pixelclassification Determining the total number of sets K can be achallenging problem .
In medical imaging the number of sets is often basedon a priori knowledge of anatomy e g K 3 gray white CSF for brain imaging Medical Imaging Informatics 2009 N Schuff UCSF VACourse 170 03 Department of.
Slide 9 67 Radiology Biomedical Imaging Labeling Labeling is the process of assigning a meaningfuldesignation to each region or pixel This process is often performed separately from.
segmentation Generally computer automated labeling is desirable Labeling and sets Zk may not necessarily share a one to one correspondenceMedical Imaging Informatics 2009 N Schuff UCSF VA.
Course 170 03 Department ofSlide 10 67 Radiology Biomedical Imaging Dimensionality Dimensionality refers to whether the segmentationoperates in a 2D or 3D domain .
Generally 2D methods are applied to 2D images and3D methods to 3D images In some instances 2D methods can be appliedsequentially to 3D images Medical Imaging Informatics 2009 N Schuff UCSF VA.
Course 170 03 Department ofSlide 11 67 Radiology Biomedical Imaging Characteristic and Membership A characteristic function is an indicator whether a pixel at location jbelongs to a particular class Zk .
1 if j Z k k j 0 otherwise This can be generalized to a membership function which does nothave to be binary valued .
0 k j 1 for all j k j 1 for The characteristic function describes a deterministic segmentationprocess whereas the membership function describes a probabilistic Medical Imaging Informatics 2009 N Schuff UCSF VA.
Course 170 03 Department ofSlide 12 67 Radiology Biomedical Imaging Segmentation Has An Important RoleComputational diagnosticSurgical planning.
Database storage retrievalSEGM Image registrationinformaticsImage transmissionQuantification.
Partial volume correctionSuper resolutionMedical Imaging Informatics 2009 N Schuff UCSF VACourse 170 03 Department ofSlide 13 67 Radiology Biomedical Imaging.
Segmentation MethodsMedical Imaging Informatics 2009 N Schuff UCSF VACourse 170 03 Department ofSlide 14 67 Radiology Biomedical Imaging Threshold Method.
Angiogram showing a right MCA aneurysmHistogram fictitious TA TB Dr Chris Ekong 0 5 10 15 20 25 30www medi fax com atlas brainan... .
Medical Imaging Informatics 2009 N Schuff UCSF VACourse 170 03 Department ofSlide 15 67 Radiology Biomedical Imaging Threshold MethodThreshold min max Threshold standard deviation.
Medical Imaging Informatics 2009 N Schuff UCSF VACourse 170 03 Department ofSlide 16 67 Radiology Biomedical Imaging Threshold Method Applied To Brain MRIWhite matter segmentation.
Major failures Anatomically non specific Insensitive to global signalinhomogeneityMedical Imaging Informatics 2009 N Schuff UCSF VA.
Course 170 03 Department ofSlide 17 67 Radiology Biomedical Imaging Threshold Principle Limitations Works only for segmentation based on intensities Robust only for images with global uniformity and high.
contrast to noise Local variability causes distortions Intrinsic assumption is made that the probability offeatures is uniformly distributedMedical Imaging Informatics 2009 N Schuff UCSF VA.
Course 170 03 Department ofSlide 18 67 Radiology Biomedical Imaging Region Growing Edge DetectionSeed point Region growing groups pixelsor subregions into larger.
A simple procedure is pixelaggregation It starts with a seed point andprogresses to neighboringpixels that have similar.
properties Region growing is better thanGuided e g by energy potentials edge detection in noisyIi I jSimilarity V i j .
Edges V i j Ii I jMedical Imaging Informatics 2009 N Schuff UCSF VACourse 170 03 Department ofSlide 19 67 Radiology Biomedical Imaging.
Region Growing Watershed TechniqueGray scale intensity a WS over segmentationCT of different types of b WS conditioned by regional density mean valuesbone tissue femur area c WS conditioned by hierarchical ordering of regional density.
mean valuesM Straka et al Proceedings of MIT 2003Medical Imaging Informatics 2009 N Schuff UCSF VACourse 170 03 Department ofSlide 20 67 Radiology Biomedical Imaging.
Region Growing Principle Limitations Segmentation results dependent on seed selection Local variability dominates the growth process Global features are ignored Generalization needed .
Unsupervised segmentation i e insensitive to selection of Exploitation of both local and global variabilityMedical Imaging Informatics 2009 N Schuff UCSF VACourse 170 03 Department ofSlide 21 67 Radiology Biomedical Imaging.
Clustering Generalization using clustering Two commonly used clustering algorithms K mean Fuzzy C mean.
Medical Imaging Informatics 2009 N Schuff UCSF VACourse 170 03 Department ofSlide 22 67 Radiology Biomedical Imaging Definitions Clustering Clustering is a process for classifying patterns in such a way that the.
samples within a class Zk are more similar to one another than samplesbelonging to the other classes Zm m k m 1 K The k means algorithm attempts to cluster n patterns based onattributes e g intensity into k classes k n The objective is to minimize total intra cluster variance in the least .
square sense k 1 j Skj k for k clusters Zk k 1 2 K i is the mean point centroid of all.
pattern values j Zk Medical Imaging Informatics 2009 N Schuff UCSF VACourse 170 03 Department ofSlide 23 67 Radiology Biomedical Imaging Fuzzy Clustering.
The fuzzy C means algorithm is a generalization of K Rather than assigning a pattern to only one class thefuzzy C means assigns the pattern a number m 0 m 1 described as membership function Medical Imaging Informatics 2009 N Schuff UCSF VA.
Course 170 03 Department ofSlide 24 67 Radiology Biomedical Imaging K meansThree classes 10 0 10 20 30 40 50.
Medical Imaging Informatics 2009 N Schuff UCSF VACourse 170 03 Department ofSlide 25 67 Radiology Biomedical Imaging K meansFour classes.
10 0 10 20 30 40 50Medical Imaging Informatics 2009 N Schuff UCSF VACourse 170 03 Department ofSlide 26 67 Radiology Biomedical Imaging K means.
0 20 40 60 80Medical Imaging Informatics 2009 N Schuff UCSF VACourse 170 03 Department ofSlide 27 67 Radiology Biomedical Imaging K means.
2 clusters0 20 40 60 80Medical Imaging Informatics 2009 N Schuff UCSF VACourse 170 03 Department ofSlide 28 67 Radiology Biomedical Imaging.
K meansThree classes0 20 40 60 80Medical Imaging Informatics 2009 N Schuff UCSF VACourse 170 03 Department of.
Slide 29 67 Radiology Biomedical Imaging K means TRAPPED Five classes0 20 40 60 80Medical Imaging Informatics 2009 N Schuff UCSF VA.
Course 170 03 Department ofSlide 30 67 Radiology Biomedical Imaging Fuzzy C meansFour classesComponent 2.
40 20 0 20 40 60Component 1These two components explain 100 of the point variability Medical Imaging Informatics 2009 N Schuff UCSF VACourse 170 03 Department of.
Slide 31 67 Radiology Biomedical Imaging Fuzzy C Means Segmentation ITwo classesOriginal Class I Class 2Medical Imaging Informatics 2009 N Schuff UCSF VA.
Course 170 03 Department ofSlide 32 67 Radiology Biomedical Imaging Fuzzy Segmentation IIFour classesClass I Class 2.
Class 3 Class 4Medical Imaging Informatics 2009 N Schuff UCSF VACourse 170 03 Department ofSlide 33 67 Radiology Biomedical Imaging Brain Segmentation With Fuzzy C Means.
4T MRI bias field inhomogeneity contributes to the problem of poor segmentationMedical Imaging Informatics 2009 N Schuff UCSF VACourse 170 03 Department ofSlide 34 67 Radiology Biomedical Imaging Clustering Principle Limitations.
Convergence to the optimal configuration is notguaranteed Outcome depends on the number of clusters chosen No easy control over balancing global and localvariability.
Intrinsic assumption of a uniform feature probability isstill being made Generalization needed Relax requirement to predetermine number of classes Balance influence of global and local variability.
Possibility to including a priori information such as non uniformdistribution of features Medical Imaging Informatics 2009 N Schuff UCSF VACourse 170 03 Department ofSlide 35 67 Radiology Biomedical Imaging.
Segmentation As Probabilistic Problem Treat both intensities Y and classes Z as random distributions The segmentation problems is finding the classes thatmaximize the likelihood to represent the image Segmentation in Bayesian formulation becomes .
p Z f Y Z p Z Y where p Y Y is the observed image values y1 yn Z is the segmented image classes z1 zK .
p Z is the prior probabillity p Y Z the observation probability and p Y is the observation and hence stableMedical Imaging Informatics 2009 N Schuff UCSF VACourse 170 03 Department of.
Slide 36 67 Radiology Biomedical Imaging Treat As Energy Minimization Problem Since p Y is stable it follows ln p Z Y ln p Z ln f Y Z The goal is to find the most probably distribution of p Z Y given the data.
Since the log probabilities are all additive they are equivalent todistribution of energyEZ Y EZ EY Z segmentation becomes an energy minimization problem This means in particular that no probabilistic point of view is finally.
Medical Imaging Informatics 2009 N Schuff UCSF VACourse 170 03 Department ofSlide 37 67 Radiology Biomedical ImagingDepartment of Radiology & Biomedical Imaging Medical Imaging Informatics 2009, N.Schuff Course # 170.03 Slide */67 Segmentation Has An Important Role SEGM Quantification Surgical planning Image registration Computational diagnostic Partial volume correction Super resolution Atlases Database storage/retrieval informatics Image transmission ...

Related Presentations