Life Cycle of Medical Imaging Data

Life Cycle of Medical Imaging Datahttp://catalyst.harvard.eduValerie Humblet, PhDHarvard Catalyst Imaging Consortium

What is an image ?

What is an image ? 2D array of pixels

Pixel dimensionsThe pixel size is the dimension in millimeters of the pixels

Life cycleClinical needAcquisitionStorageVisualizationAnalysisRegistrationSegmentation

Clinical needClinical needDiagnosis:Heart diseaseStrokeCancer Decision maker:Patient with myocardial infarction, MRI to assess viability for bypass surgery

AcquisitionMagnetic Resonance ImagingUltrasoundComputedTomographyImage: BWHPET/CTImage: Philips

Data representationResult of acquisition is a 3D volume of data related to the patientThe 3D volume is sampled on a 3D grid in the acquisitioncoordinate system (I,J,K).

StorageAll medical imaging data is collected and stored in standard radiological file format

Digital Imaging and Communications in MedicineDICOM 3.0 (1993)File has complex header that contains fields critically important to image analysis as well as fields with IHI

DICOM 3.0

DICOM 3.00002,0000,File Meta Elements Group Len=1480002,0001,File Meta Info Version=2560002,0002,Media Storage SOP Class UID=1.2.840.10008.5.1.4.1.1.4.0002,0003,Media Storage SOP Inst UID=0.0.0.0.0002,0010,Transfer Syntax UID=1.2.840.10008.1.2.1.0008,0060,Modality=MR0008,0070,Manufacturer=GE MEDICAL SYSTEMS0008,0080,Institution Name=18527965130008,0081,City Name=18527965130008,0090,Referring Physician's Name=18527965130008,0092,?=18527965130008,0201,?=-0500 0008,1010,Station Name=18527965130008,1030,Study Description=anon0008,103E,Series Description=anon0008,1040,Institutional Dept. Name=18527965130008,1050,Performing Physician's Name=18527965130008,1060,Name Phys(s) Read Study=18527965130008,1070,Operator's Name=anon0008,1080,Admitting Diagnosis Description=18527965130008,1090,Manufacturer's Model Name=GENESIS.SIGNA ....0028,0010,Rows=2560028,0011,Columns=2560028,0030,Pixel Spacing=0.937500 0.937500 0028,0100,Bits Allocated=160028,0101,Bits Stored=160028,0102,High Bit=150028,0103,Pixel Representation=1.7FE0,0010,Pixel Data=131072Data Header

DICOM 3.00002,0000,File Meta Elements Group Len=1480002,0001,File Meta Info Version=2560002,0002,Media Storage SOP Class UID=1.2.840.10008.5.1.4.1.1.4.0002,0003,Media Storage SOP Inst UID=0.0.0.0.0002,0010,Transfer Syntax UID=1.2.840.10008.1.2.1.0008,0060,Modality=MR0008,0070,Manufacturer=GE MEDICAL SYSTEMS0008,0080,Institution Name=18527965130008,0081,City Name=18527965130008,0090,Referring Physician's Name=18527965130008,0092,?=18527965130008,0201,?=-0500 0008,1010,Station Name=18527965130008,1030,Study Description=anon0008,103E,Series Description=anon0008,1040,Institutional Dept. Name=18527965130008,1050,Performing Physician's Name=18527965130008,1060,Name Phys(s) Read Study=18527965130008,1070,Operator's Name=anon0008,1080,Admitting Diagnosis Description=18527965130008,1090,Manufacturer's Model Name=GENESIS.SIGNA ..0010,0010,Patient's Name=anon0010,0020,Patient ID=anon0010,0030,Patient Date of Birth=000000000010,0032,Patient Birth Time=0000000010,0040,Patient Sex=O 0010,1010,Patient Age=000Y..0028,0010,Rows=2560028,0011,Columns=2560028,0030,Pixel Spacing=0.937500 0.937500 0028,0100,Bits Allocated=160028,0101,Bits Stored=160028,0102,High Bit=150028,0103,Pixel Representation=1Physician and Study informationData Header

DICOM 3.00002,0000,File Meta Elements Group Len=1480002,0001,File Meta Info Version=2560002,0002,Media Storage SOP Class UID=1.2.840.10008.5.1.4.1.1.4.0002,0003,Media Storage SOP Inst UID=0.0.0.0.0002,0010,Transfer Syntax UID=1.2.840.10008.1.2.1.0008,0060,Modality=MR0008,0070,Manufacturer=GE MEDICAL SYSTEMS0008,0080,Institution Name=18527965130008,0081,City Name=18527965130008,0090,Referring Physician's Name=18527965130008,0092,?=18527965130008,0201,?=-0500 0008,1010,Station Name=18527965130008,1030,Study Description=anon0008,103E,Series Description=anon0008,1040,Institutional Dept. Name=18527965130008,1050,Performing Physician's Name=18527965130008,1060,Name Phys(s) Read Study=18527965130008,1070,Operator's Name=anon0008,1080,Admitting Diagnosis Description=18527965130008,1090,Manufacturer's Model Name=GENESIS.SIGNA ..0010,0010,Patient's Name=anon0010,0020,Patient ID=anon0010,0030,Patient Date of Birth=000000000010,0032,Patient Birth Time=0000000010,0040,Patient Sex=O 0010,1010,Patient Age=000Y..0028,0010,Rows=2560028,0011,Columns=2560028,0030,Pixel Spacing=0.937500 0.937500 0028,0100,Bits Allocated=160028,0101,Bits Stored=160028,0102,High Bit=150028,0103,Pixel Representation=1Data HeaderPatient information

DICOM 3.00002,0000,File Meta Elements Group Len=1480002,0001,File Meta Info Version=2560002,0002,Media Storage SOP Class UID=1.2.840.10008.5.1.4.1.1.4.0002,0003,Media Storage SOP Inst UID=0.0.0.0.0002,0010,Transfer Syntax UID=1.2.840.10008.1.2.1.0008,0060,Modality=MR0008,0070,Manufacturer=GE MEDICAL SYSTEMS0008,0080,Institution Name=18527965130008,0081,City Name=18527965130008,0090,Referring Physician's Name=18527965130008,0092,?=18527965130008,0201,?=-0500 0008,1010,Station Name=18527965130008,1030,Study Description=anon0008,103E,Series Description=anon0008,1040,Institutional Dept. Name=18527965130008,1050,Performing Physician's Name=18527965130008,1060,Name Phys(s) Read Study=18527965130008,1070,Operator's Name=anon0008,1080,Admitting Diagnosis Description=18527965130008,1090,Manufacturer's Model Name=GENESIS.SIGNA ..0010,0010,Patient's Name=anon0010,0020,Patient ID=anon0010,0030,Patient Date of Birth=000000000010,0032,Patient Birth Time=0000000010,0040,Patient Sex=O 0010,1010,Patient Age=000Y..0028,0010,Rows=2560028,0011,Columns=2560028,0030,Pixel Spacing=0.937500 0.937500 0028,0100,Bits Allocated=160028,0101,Bits Stored=160028,0102,High Bit=150028,0103,Pixel Representation=1Data HeaderImage information

DICOM 3.0

0002,0003,Media Storage SOP Inst UID=0.0.0.0.0002,0010,Transfer Syntax UID=1.2.840.10008.1.2.1.0008,0060,Modality=MR0008,0070,Manufacturer=GE MEDICAL SYSTEMS0008,0080,Institution Name=18527965130008,0081,City Name=18527965130008,0090,Referring Physician's Name=18527965130008,0092,?=18527965130008,0201,?=-0500 0008,1010,Station Name=18527965130008,1030,Study Description=anon0008,103E,Series Description=anon0008,1040,Institutional Dept. Name=18527965130008,1050,Performing Physician's Name=18527965130008,1060,Name Phys(s) Read Study=18527965130008,1070,Operator's Name=anon0008,1080,Admitting Diagnosis Description=18527965130008,1090,Manufacturer's Model Name=GENESIS.SIGNA ..0010,0010,Patient's Name=anon0010,0020,Patient ID=anon0010,0030,Patient Date of Birth=000000000010,0032,Patient Birth Time=0000000010,0040,Patient Sex=O 0010,1010,Patient Age=000Y..0028,0010,Rows=2560028,0011,Columns=2560028,0030,Pixel Spacing=0.937500 0.937500 0028,0100,Bits Allocated=160028,0101,Bits Stored=160028,0102,High Bit=150028,0103,Pixel Representation=1.7FE0,0010,Pixel Data=131072

Data HeaderThe data

Data compression2 types of algorithmsThe lossy compression techniques deliberately discard information that is not diagnostically important

Ex: JPEGThe lossless compression techniques allow the exact original data to be reconstructed from the compressed data.

Ex: JPEG-LS

PACS*Picture Archiving and Communication System Hospital based computer networked storage of images DICOM images available immediatelyUltrasound, CT, X-ray, MRI, PET, Nuclear medicineArchiving system, replacing hardcopies, CDs, DVDs, accessible through network

VisualizationInteract in 3D to enhance data interpretation

Anatomical planesThe 3D viewer displays a model of the headThe 2D viewer display the three anatomical planes (axial, sagittal,coronal)

Space direction

Voxel ordering

Axes for spatial coordinatesThe index i in the file increases fromthe Left to the Right side of the patient.The index j in the file increases fromPosterior to Anterior.The index k in the file increases fromInferior to Superior.

RAS: Right-Anterior-Superior

Axes for spatial coordinatesThe index i in the file increases fromthe Right to the Left side of the patient.The index j in the file increases fromAnterior to Posterior.The index k in the file increases fromInferior to SuperiorLPS: Left-Posterior-Superior

RegistrationRegistration is the process of transforming three different spaces into a common reference frame

Use for image registrationWithin or Intra-subject:Purpose: To combine functional and anatomical info of different imaging modalitiesExamples: CT and MR for surgical planningMR and PET/SPECT images of tracer uptake for localization of functional activityAcross or Inter-subject:Purpose: To assess individual or group variability in some anatomical or functional measureExamples: Responders vs. non-responders to treatment

Use for image registrationSerial or Longitudinal: a sequence of images collected over time of the same subject(s) Purpose: To assess change within a subject or group over time due to development, aging, disease progression and/or to monitor response to treatment.Examples: Deformation based morphometry (months), fMRI (minutes)

Subject to atlas:Purpose: To use population based information as priors to inform labeling or registrationExamples: Brain segmentation and parcellation, Boundary based registration

Why use registration?Same subject, two time points

How a registration algorithm works

Compensate for global patient repositioningPreserve distances, planes and angles

Appropriate for Bones, brain, optically tracked surgical instrumentOften use to initialize non-rigid registrationTransformation

Non-rigid transformationCompensate for deformation of the subject

Example of non-rigid registration

SegmentationGoal: identify or label structures present in the image.Applications:Quantitative measurement of volume, shape or location of structures,Provides boundary for visualization by surface rendering.

Segmentation methodsInteractive or manual delineationSupervised approaches with user initializationThresholdingClusteringRegion growingEdge detection Atlas based alignment with a template Statistical pattern recognition

Image segmentationSegmentation issues:Interactive segmentation:time consuming.significant intra-rater and inter-rater variability (Warfield et al. 1995).Automatic segmentation:Challenges.Imaging artifacts.Normal and pathological variability.Prospects:Objective assessment of imaging data.

Validation of registration, segmentation or any image processing methodPrior to entering clinical practice:Technical validationSpeed, robustness, accuracy, reliabilityClinical validationUtility, improved diagnosis and patient managementFDA approval, incorporation into commercial system Liability

*AcknowledgementsNIH award #UL1 RR 025758

Documents

Life Cycle of Medical Imaging Data