Towards Clinically-relevant Standardization of Image Quality

1
Towards Clinically-relevant Standardization of
Image Quality

• Ehsan Samei, Duke University
• Alan Rowberg, University of Washington
• Ellie Avraham, Eastman Kodak Company
• Craig Cornelius, Eastman Kodak Company

2
Objectives

• Describe current medical image quality and
  consistency performance efforts
• Identify limitations in existing standards
• Outline 3 specific proposals
• Add new image quality factors to standards
• Update DICOM IQ performance services
• Research technical-clinical connection

3
DICOM Image Consistency Efforts

• Three DICOM initiatives (1998-99)
• Grayscale Standard Display Function (GSDF)
• Presentation LUT (P-LUT)
• Grayscale Softcopy Presentation State (GSPS)
• And in actual practice
• Are widely effectively implemented
• Via IHE Consistent Presentation of Images
  solution
• Promoted conformance testing
• Demonstrated inter-vendor consistency

4
DICOM GSDF Barten Curve
Absolute luminance increment L1 ltlt L2
Number of perceptual levels J1 J2
5
AAPM Task Group 18 Efforts

• Team of academic, clinical, and industry
  contributors
• Assessment of Display Performance for Medical
  Imaging Systems (2002)
• Practical guidelines for qualitative
  quantitative display assessment
• Includes all key aspects of display performance
• Defines IQ test patterns and procedures
• Recommends specific IQ acceptance criteria

6
Standardization gaps

• DICOM GSDF
• Pros
• Mathematical definition based on Human Visual
  System model
• Limitations
• Tonescale consistency only no other IQ factors
• No acceptance criteria / conformance procedures
• Only for grayscale images
• No Display Device Services (Capabilities)

• AAPM TG18 report
• Pros
• Provides Professional recommendations
• Covers all key display performance aspects
• Acceptance Criteria
• Quantitative measures
• Limitations
• Only guidelines
• Not a standard

7
The technical-clinical gap

• The connection between quantifiable IQ metrics
  clinical performance is unknown
• Luminance deviations from GSDF, number of gray
  levels displayed
• Spatial resolution, noise, geometric distortion
• Chromaticity variations
• Environmental ambient light, glare, reflection,
• Amount of acceptable variation is unknown

8
3 proposals to bridge the gaps

1. Extend DICOM standard beyond luminance response
2. Add and update DICOM Service Classes for image
   quality / performance
3. Promote research on clinical - technical image
   quality relationship

9
1. Display Image Quality (DIQ) Initiative

• Add measurable and quantifiable elements of AAPM
  display performance procedures
• Include testing methodologies defined limits
  for clinical / diagnostic performance
• Quantify visual performance using
• Simple test images
• Specific observer protocols
• Relative acceptance indicators
• For both softcopy and hardcopy presentation

10
DIQ Softcopy Examples

• Quantify deviation from GSDF curve
• Define criteria for min max luminance
• Define visual luminance evaluation
• Evaluate specific image quality factors
• Ambient light limits specular and diffuse
• Spatial resolution with TG18-QC/CX
• Check geometric distortion with TG18-QC

11
Contrast response comparison
GSDF contrast/- 10
Non-standardized display contrast
12
AAPM TG18 Patterns
TG18-MP bit-depth / continuous grayscale
TG18-CTcontrast / luminance response
13
Comprehensive TG18-QC

• The comprehensive TG18-QC test pattern for
  evaluation of key display characteristics
• Resolution
• Luminance
• Geometric distortion

14
Other DIQ Extensions

• Hardcopy quality metrics
• Media printer quality (e.g., visible coating
  variations, distortions, artifacts)
• GSDF compliance, of JNDs theory vs. actual
• Printable matrix size
• Spatial frequency response fidelity
• Color extensions
• Standardize for grayscale areas of color images
• Add descriptions for color image characterization
• Color display and print device calibration

15
2. DICOM Service Extensions

• Add Display Performance Service Class
• Query image quality / performance information
• Control / configure manageable settings
• Include new IQ factors, e.g., MTF, ambient,
• Extend Printer Configuration Retrieval Service
  Class
• Include access to additional IQ factors
• Note Measurables include both
  human-evaluated and automatically-measured values

16
Use cases Standardizing Output

• Printing application
• Retrieves matrix size and MTF of film printer
• Determines type of magnification, if any, to be
  applied to the image for smallest artifact
• Display Performance Service Class User
• Requests the luminance characteristic curve from
  its workstations display system
• Determines if the display is standardized
• If needed, computes an internal image tonescale
  correction, producing GSDF standardized result

17
Real device ? standardized performance
300.00
250.00
Target GSDF
200.00
Actual device
150.00
Luminance
100.00
50.00
0.00
0
100
200
300
400
500
600
700
Digital Value
18
Use Case Quality Control Management

• Centralized management application
• Queries devices for calibration date, luminance
  characteristics, ambient light settings, etc.
• Records the results in a central database
• Creates maintenance lists for displays and
  printers
• Reporting application uses database for
• Regulatory and management reports
• Stability and lifetime statistics on displays to
  support replacement schedules and budgets

19
Network-wide Quality Control
RegulatoryReports
Performance LifetimeStatistics
20
More use cases

• Consultation
• Maximize perceptual similarity to ensure What I
  see is what you will get!
• Display performance information gives confidence
• Capture Consoles
• The technologist sees what the doctor will get.
• Reduce errors, retakes, miscommunication
• Extend quality control program to consoles

21
What the Radiologist Displays
What the Reviewing Physician Sees
Consistent Presentation of Images On a Calibrated
Display
With Presentation State
The Radiologists Transformations Are Saved
Original Image
Inconsistent Presentation On an Uncalibrated
Display
Original Image
The Radiologists Transformations Are Lost
22
3. Clinical Significant of IQ Measures
Hmm
We assume there is a connection
What deviations matter, and how much?
Physical metrics
Clinical performance
23
Clinical impact unknown
Rendered with DICOM GSDF
Not rendered with DICOM GSDF
24
Goals of proposed research

• Determine the clinical consequence of variations
  in image quality metrics (e.g., GSDF conformance,
  MTF, noise, )
• Define what constitutes image quality from a
  diagnostic perspective
• Incorporate results into new joint standards that
  will utilize standardized test patterns,
  procedures, and clinical use cases

25
Suggested research approach

• Sample research
• Obtain images from 3 radiographic modalities
• Present images, simulating nonstandard display
  behavior
• Run observer performance experiments at major
  professional meetings and events
• Analyze by ROC methods

• Form inter-society committee to
• Design specific research projects
• Obtains review data sets
• Solicit encourage active participation by
  researchers radiologists, scientists,
• Arrange reporting of results
• Define recommendations to standards and
  professional groups

26
Conclusions

• Existing standards are insufficient to assure
  consistent, high quality medical image output.
• Steps are proposed to further the reach and
  impact of DICOM toward quality medicine.
• New directions will provide benefits for PACS
  users, administrators, vendors, and patients.

27
Contact Information

• Ehsan Samei, Duke University samei_at_duke.edu
• Alan Rowberg, M.D., University of Washington,
  arowberg_at_earthlink.com
• Ellie Avraham, Eastman Kodak Company, ellie.avrah
  am_at_kodak.com
• Craig Cornelius, Eastman Kodak Company, craig.cor
  nelius_at_kodak.com
• AAPM Task Group 18 web site http//deckard.duhs.
  duke.edu/tg18