Mammography and DICOM
Adapting an Analog Modality
to the Digital World
Julian Marshall
R2 Technology, Inc.
Mammography
• Mammography is a film-based modality– Worldwide mammo machines:
• 25,500 film-screen
• 500 digital
98.1 %
1.9 %
Reading Mammograms
• ACR position:– Radiologist must read original image
• US clinical practice:– Read film-screen mammograms on film
• Do not digitize films and read softcopy
– Priors can be read softcopy
Digitized Film
• Mammograms are digitized– Wide variation– Scanners vary:
• Resolution
• Maximum O.D.
• Noise
RCC
RCC
FILM
SCANNER
IMAGE
Digital Mammography
• Mammograms are acquired digitally– Detectors do still vary:
• Resolution
• Bit depth (CR)
• Noise (CR)
RCC
IMAGE
Mammography
• Imaging demands are extreme:– Typical resolutions:
• Film: 43 to 50 microns x 12 bits
• Digital: 50 to 100 microns x 14 bits
– Typical image sizes:• 18x24 cm 85%
• 24x30 cm 15%
Mammography
• Imaging demands are extreme:– Typical data volume:
• 4 film case: 180 MB avg
• 100 case/day: 18.0 GB/day
• 250 days/yr: 4.5 TB/year
– Film scanner will generate:• 45 MB per minute, all day long!
Mammography and PACS
• Images are recalled regularly
• Scheduled pre-fetching is easy
• But … each image is accessed each year!
RCCRCC
RCCRCC
RCCRCC
RCCRCC
RCCRCC
RCCRCC
RCCRCC
RCCRCC
RCCRCC
RCCRCC
RCCRCC
RCCRCC
RCCRCC
RCCRCC
RCCRCC
RCCRCC
RCCRCC
RCCRCC
1998 2000 2002
Computer-Aided Detection
• Use a computer to look for regions-of-interest that might be overlooked by a radiologist
• Simple example: Count the ‘F’s
Computer-Aided Detection
• Simple example: Count the ‘F’s
FINISHED FILES ARE THE RE-SULT OF YEARS OF SCIENTIF-IC STUDY COMBINED WITHTHE EXPERIENCE OF YEARS
Computer-Aided Detection
• Most people find these three
FINISHED FILES ARE THE RE-SULT OF YEARS OF SCIENTIF-IC STUDY COMBINED WITHTHE EXPERIENCE OF YEARS
Computer-Aided Detection
• Many people do not find all six!
FINISHED FILES ARE THE RE-SULT OF YEARS OF SCIENTIF-IC STUDY COMBINED WITHTHE EXPERIENCE OF YEARS
Computer-Aided Detection
• Mammography CAD first became available:– 1998 Film-screen mammography– 2000 Digital mammography
• At that time:– DICOM support for images– No DICOM support for CAD output
DICOM WG 15
• Standards development:– Digital X-ray (includes mammo)
1998– Mammography CAD SR 2001
Mammography CAD SR
• Allows encoding of ACR’s BI-RADSTM reporting structure via an inference tree
• “Simple” CAD devices can create “simple” Mammo CAD objects
• “Complex” CAD devices can create full mammography report inference tree
Mammography CAD SR
• Single image finding – found in one image
• Composite object – findings correlated in one or more images:– Temporal – comparison over time– Spatial – e.g. mass behind the nipple, or
mammo/ultrasound correlation– Contra-laterally – e.g. left/right comparison
Inference Tree
SingleImage
Finding
CompositeObject
SingleImage
Finding
CompositeObject
IndividualImpression /
RecommendationSP
T
CL
Related Spatially
Related Temporally
Related Contra-Laterally
O
I Individual
Overall
C Individual Calcification
CC Calcification Cluster
D Density
OverallImpression /
Recommendation
CC D D D D
SP SP
SP T
D
I
SP
I
O
C C C
CC
Three individual calcifications are detected in a single image
Individual Calcification:–Location of center
–Outline of individual calcification
–Size
Inference Tree
SingleImage
Finding
CompositeObject
SingleImage
Finding
CompositeObject
IndividualImpression /
RecommendationSP
T
CL
Related Spatially
Related Temporally
Related Contra-Laterally
O
I Individual
Overall
C Individual Calcification
CC Calcification Cluster
D Density
OverallImpression /
Recommendation
CC D D D D
SP SP
SP T
D
I
SP
I
O
C C C
CC
The three are grouped together as a cluster of calcifications
Calcification cluster:–Location of center
–Outline of cluster
–Size
–No. of individual calcifications
Inference Tree
SingleImage
Finding
CompositeObject
SingleImage
Finding
CompositeObject
IndividualImpression /
RecommendationSP
T
CL
Related Spatially
Related Temporally
Related Contra-Laterally
O
I Individual
Overall
C Individual Calcification
CC Calcification Cluster
D Density
OverallImpression /
Recommendation
CC D D D D
SP SP
SP T
D
I
SP
I
O
C C C
CC
Densities and other clusters are detected, some from priors
Density–Center of density
–Outline
–Size
–Description of margin
Inference Tree
SingleImage
Finding
CompositeObject
SingleImage
Finding
CompositeObject
IndividualImpression /
RecommendationSP
T
CL
Related Spatially
Related Temporally
Related Contra-Laterally
O
I Individual
Overall
C Individual Calcification
CC Calcification Cluster
D Density
OverallImpression /
Recommendation
CC D D D D
SP SP
SP T
D
I
SP
I
O
C C C
CC
Densities become masses if spatially related
Inference Tree
SingleImage
Finding
CompositeObject
SingleImage
Finding
CompositeObject
IndividualImpression /
RecommendationSP
T
CL
Related Spatially
Related Temporally
Related Contra-Laterally
O
I Individual
Overall
C Individual Calcification
CC Calcification Cluster
D Density
OverallImpression /
Recommendation
CC D D D D
SP SP
SP T
D
I
SP
I
O
C C C
CC
Other findings may also be spatially related
Inference Tree
SingleImage
Finding
CompositeObject
SingleImage
Finding
CompositeObject
IndividualImpression /
RecommendationSP
T
CL
Related Spatially
Related Temporally
Related Contra-Laterally
O
I Individual
Overall
C Individual Calcification
CC Calcification Cluster
D Density
OverallImpression /
Recommendation
CC D D D D
SP SP
SP T
D
I
SP
I
O
C C C
CC
Calcs within a mass are related spatially
Inference Tree
SingleImage
Finding
CompositeObject
SingleImage
Finding
CompositeObject
IndividualImpression /
RecommendationSP
T
CL
Related Spatially
Related Temporally
Related Contra-Laterally
O
I Individual
Overall
C Individual Calcification
CC Calcification Cluster
D Density
OverallImpression /
Recommendation
CC D D D D
SP SP
SP T
D
I
SP
I
O
C C C
CC
Objects found in priors are temporally related to currents
Inference Tree
SingleImage
Finding
CompositeObject
SingleImage
Finding
CompositeObject
IndividualImpression /
RecommendationSP
T
CL
Related Spatially
Related Temporally
Related Contra-Laterally
O
I Individual
Overall
C Individual Calcification
CC Calcification Cluster
D Density
OverallImpression /
Recommendation
CC D D D D
SP SP
SP T
D
I
SP
I
O
C C C
CC
Objects can also be related contra-laterally (not shown here)
Inference Tree
SingleImage
Finding
CompositeObject
SingleImage
Finding
CompositeObject
IndividualImpression /
RecommendationSP
T
CL
Related Spatially
Related Temporally
Related Contra-Laterally
O
I Individual
Overall
C Individual Calcification
CC Calcification Cluster
D Density
OverallImpression /
Recommendation
CC D D D D
SP SP
SP T
D
I
SP
I
O
C C C
CC
Individual Impressions and Recommendations are formed
Inference Tree
SingleImage
Finding
CompositeObject
SingleImage
Finding
CompositeObject
IndividualImpression /
RecommendationSP
T
CL
Related Spatially
Related Temporally
Related Contra-Laterally
O
I Individual
Overall
C Individual Calcification
CC Calcification Cluster
D Density
OverallImpression /
Recommendation
CC D D D D
SP SP
SP T
D
I
SP
I
O
C C C
CC
Overall Impression and Recommendation is formed
A Vast Array of Adjectives
• Every Single Image Finding and Composite Object has a set of common descriptors:– Rendering intent– Certainty of finding– Probability of cancer
• Plus a variety of context-specific descriptors:– Calcs: rod-like, pleomorphic, etc.
Other Information
• Breast outline (border)
• Pectoral muscle outline
• Nipple location
• Other findings:– BBs– J-wires
Other Information
• Image quality findings– Motion blur– Artifacts
Coming Soon
• Breast Imaging Report SR
• Relevant Patient History Query
Summary
• Mammography is almost entirely a film-based modality
• Slowly this is changing
• And with that change comes DICOM!