Medical Imaging Research and Regulatory Concerns at the FDA

1
Medical Imaging Research and Regulatory Concerns
at the FDA

• David G. Brown, Ph.D.
• 11 November 2008
• david.brown_at_fda.hhs.gov

2
MI at FDA Overview

• FDA
• What, where, who, why, how
• (Is medical imaging a food or a drug?)
• Medical Imaging
• From safety to effectiveness
• Clinical trials and computer assists
• A sample of current projects

3
Division of Imaging and Applied Mathematics
OSEL, CDRH, FDA
4
Center for Devices and Radiological Health
5
Office of Science and Engineering Laboratories
6
Engineering and Physics Building FDA White Oak MD
Campus
7
Division of Imaging and Applied Mathematics

• Performance assessment for diagnostic medical
  imaging and related systems

8
(No Transcript)
9
FDA Mission

• The FDA is responsible for protecting the public
  health by assuring the safety, efficacy, and
  security of human and veterinary drugs,
  biological products, medical devices, our
  nations food supply, cosmetics, and products
  that emit radiation. The FDA is also responsible
  for advancing the public health by helping to
  speed innovations that make medicines and foods
  more effective, safer, and more affordable and
  helping the public get the accurate,
  science-based information they need to use
  medicines and foods to improve their health.

10
How Big is FDA?

• The FDA regulates more than 1 trillion worth of
  consumer goods, about 25 percent of consumer
  expenditures in the United States. This includes
  466 billion in food sales, 275 billion in
  drugs, 60 billion in cosmetics and 18 billion
  in vitamin supplements. Much of the expenditure
  is for goods imported into the United States the
  FDA is responsible for monitoring a third of all
  imports. (Wikipedia)
• 3 billion budget
• 15,000 staff

11
What FDA Regulates

• FDA is the federal agency responsible for
  ensuring that foods are safe, wholesome and
  sanitary human and veterinary drugs, biological
  products, and medical devices are safe and
  effective cosmetics are safe and electronic
  products that emit radiation are safe. FDA also
  ensures that these products are honestly,
  accurately and informatively represented to the
  public.

12
What FDA Does Not Regulate

• Advertising (FTC)except prescription drugs and
  medical devices
• Alcohol (BATF)
• Consumer Products (CPSC)except medical and
  radiation emitting
• Drugs of Abuse (DEA)
• Health Insurance (CMS)
• Meat and Poultry (USDA)
• Pesticides (USDA/EPA)
• Restaurants and Grocery Stores (local)
• Water (EPA)except bottled water

13
Miscellaneous FDA Activities

• MedWatch -- FDA provides safety information on
  drugs and other FDA-regulated products, and
  allows for adverse event reporting.
• Recalls -- FDA posts significant product actions
  of the last 60 days.
• Inspections - FDA inspects processing plants and
  other agency-regulated facilities.
• Advisory Committees - FDA convenes public
  meetings with outside experts for advice on
  making key public health decisions.

14
CDRH Regulation

• Medical Devices
• By risk classification
• Clearance vs. approval
• Safety and effectiveness
• Least burdensome
• Radiation Products
• Consumer as well as medical

15
What is a medical device?

• "an instrument, apparatus, implement, machine,
  contrivance, implant, in vitro reagent, or other
  similar or related article, including a component
  part, or accessory which is
• recognized in the official National Formulary, or
  the United States Pharmacopoeia, or any
  supplement to them,
• intended for use in the diagnosis of disease or
  other conditions, or in the cure, mitigation,
  treatment, or prevention of disease, in man or
  other animals, or
• intended to affect the structure or any function
  of the body of man or other animals, and which
  does not achieve any of it's primary intended
  purposes through chemical action within or on the
  body of man or other animals and which is not
  dependent upon being metabolized for the
  achievement of any of its primary intended
  purposes."

16
Classification of Medical Devices

• Class I (low risk)e.g., stethoscopes
• Class II (moderate risk)e.g., most imaging
  devices such as CT, MRI and US scanners
• Class III (high risk)e.g., pacemakers. Other
  devices may be Class III because
• Used in conjunction with Class III devices
• Has a large potential effect on the public health
  
• Scientific principles of the device are not
  well-known

17
510(k)

• Most Class II devices are cleared for marketing
  via a 510(k) (named after the section of the
  statutory law) in which the sponsor demonstrates
  SUBSTANTIAL EQUIVALENCE to another legally U.S.
  marketed device (referred to as a predicate
  device).

18
510(k)

• A device is substantially equivalent if, compared
  to a predicate it
• has the SAME INTENDED USE as the predicate AND
• has the SAME TECHNOLOGICAL CHARACTERISTICS as the
  predicate
• OR

19
510(k)

• has the SAME INTENDED USE as the predicate
  device AND

20
510(k)

• Has different technological characteristics
  (e.g., change in material, design, energy source,
  software) AND the information submitted to FDA
• Does not raise different (i.e., new) types of
  questions of safety and effectiveness AND
• Demonstrates that the device is as safe and
  effective as the predicate device.

21
Pre-Market Approval Application (PMA)

• Most Class III devices are approved for marketing
  via a PMA
• Unlike a 510(k), a PMA is not typically a
  comparison to other legally marketed devices but
  must stand on its own to demonstrate the safety
  and effectiveness of the device for its intended
  use.

22
Safety

• There is reasonable assurance that a device is
  safe when it can be determined, based upon valid
  scientific evidence, that the probable benefits
  to health from use of the device for its intended
  uses and conditions of use, when accompanied by
  adequate directions and warnings against unsafe
  use, outweigh any probable risks

23
Safety

• The valid scientific evidence used to determine
  the safety of a device must adequately
  demonstrate the absence of unreasonable risk of
  illness or injury associated with the use of the
  device for its intended uses and conditions of
  use

24
Safety

• Among the types of evidence that may be required,
  when appropriate, to determine that there is
  reasonable assurance that a device is safe are
  investigations using laboratory animals,
  investigations involving human subjects, and
  non-clinical investigations including in vitro
  studies

25
Effectiveness

• There is reasonable assurance that a device is
  effective when it can be determined, based upon
  valid scientific evidence, that in a significant
  portion of the target population, the use of the
  device for its intended uses and conditions of
  use, when accompanied by adequate directions for
  use and warnings against unsafe use, will provide
  clinically significant results

26
Effectiveness

• The valid scientific evidence used to determine
  the effectiveness of a device shall consist
  principally of well-controlled investigations, as
  statutorily defined, unless the Commissioner
  authorizes reliance upon other valid scientific
  evidence which the Commissioner has determined is
  sufficient evidence from which to determine the
  effectiveness of a device, even in the absence of
  well-controlled investigations.

27
Least Burdensome

• A central purpose of the Food and Drug
  Administration Modernization Act of 1997 (FDAMA)
  is to ensure the timely availability of safe and
  effective new products that will benefit the
  public

28
Imaging Research in DIAM

• Emphasis on performance evaluation
• Study complete imaging chain from radiation
  production to image display and human observation
• Development of task-based methodology
• Reliance on physics, statistics, simulation

29
The typical image processing paper has six pages
of math and three pictures 1. Original image
30
2. Noisy image
31
3. Processed image
32
Task Selection

• Meaningful imaging performance assessment is task
  dependent

33
Physician Selection

• Not all Drs. were created equal

34
Use CAD Tools

• Reduce physician variability
• Increase accuracy

35
How good is your CAD?

• Medipatterns B-CAD(R) increases diagnostic
  accuracy on small breast cancer lesions by 44
• New applications will enable radiologists to
  analyze and track lymph nodes of interest in an
  accurate and consistent manner.
• Az values for radiologist alone .85, with CADx
  .89, with CALMA .91. Area under the ROC curve
  for unaided mammography reading and reading using
  two different CAD systems
• Az with CAD .96 and for junior radiologists .94

36
The dimension problem

• Many, many featuresToo few patients
• Genomic microarray data 10,000 or more
  gene-expression values per measurement
• Tens of patients, poor quality data, poor quality
  ground truth

37
DIAM projects

• Improved X-ray Imaging for Women
• Coronary applications
• How to improve imaging protocols and image-guided
  interventions for women?
• Breast applications emphasis on tomosynthesis
• What are the optimal imaging techniques,
  reconstruction algorithms, and display methods?

Cardiovascular disease mortality trends for
males and females (US 1979-2002).
Tomosynthesis system
detector

x
-
scanning ray tube

q

planes of reconstruction
38
DIAM projects

• Improved Disease Quantitation via 3D Imaging
• Lung CT applications
• How can CAD algorithms be efficiently validated
  for changes in CT scanners, different imaging
  techniques, and CAD algorithm enhancements?
• What imaging-based measures are appropriate
  indicators of drug response?

Accelerate bench-to-bedside availability of new
technologies!
39
DIAM products

• Training for outstanding post-docs
• Training a new cadre of researchers who have been
  exposed to the regulatory environment and
  understand the safety and effectiveness questions
  that regulators must ask
• Lab tomosynthesis/
• cone-beam CT system for support of multiple
  projects
• Scientific presentations
• Journal publications
• Posters
• Regulatory support

40
A system for computer based Coronary Artery
Disease imaging clinical trials

• Iacovos Kyprianou

41
Heart muscle supplies blood to tissues,
organs. Coronary artery disease affects the blood
supply to the heart.
Coronary Artery Disease (CAD) is diagnosed with
coronary angiogram
Plaque is formed from fatty deposits in the
coronary arteries
Catheter transported to the coronaries through
the femoral artery
Over time arteries are clogged and hardened
(atherosclerosis)?
Arteries show up on screen as dark trees.
Stenosis shows up as narrowing
42
Coronary angiogram
43
Treatment example
A catheter is transported through the femoral
artery to the stenosis location
A stent is transported to the lesion site and
balloon expands
Catheter is removed, arterial pathway is free.
44
Public health concernCardiovascular disease
largest, most expensive killer disease

• Coronary artery disease
• 38 of all deaths
• 400B cost
• Many open questions
• Gender differences
• Anatomical, physiological, socioeconomic
• Differences affect diagnosis treatment, and
  outcomes
• Currently women are not served by angiography

Data American Heart Association 2005
45
Clinical study development

• Determine gender differences that affect imaging
• Build male and female virtual patients with
  differences incorporated
• Build a virtual imaging system/cath-lab
• Develop a virtual physician to detect disease
• Use models to determine best imaging protocol to
  image each gender (each patient?)?

46
Gender differencesGenerally women

• More unusual symptoms of a heart attack
• Longer to get to the hospital
• Less likely to be admitted to the intensive
  cardiac care, get electrocardiograms,
  clot-busting drugs, or cardiac catheterization
• More likely to die from a heart attack or get
  second
• Older, sicker (more diabetes, high blood
  pressure, other)
• More likely to die in the hospital during bypass
  or angioplasty
• Less likely to have better quality of life after
  bypass surgery
• Less likely to be directed to a cardiac
  rehabilitation program, or finish one
• Less likely to get counseling about nutrition,
  exercise, weight loss to prevent heart disease

47
Gender differencesSpecific to imaging
Differences in plaque deposition, size and
morphology.
Hemodynamic differences (different blood
viscosity).
Coronary calcium detection lower scores in
women (smaller arteries, smaller plaque)?
Coronary angiography less likely to succeed
women are older, frequent failure to cross
lesion, smaller vessel caliber
Higher prevalence of contrast-induced nephropathy
in women
48
Why clinical trials in-silico?

• Can help develop and evaluate 21st century
  devices with 21st century methodologies
• Can simulate a multitude of imaging systems
• Infinite number of patients
• Target and minimize real clinical trials
• Reduce trial costs
• Allow for more specific, stronger claims and
  labeling
• Faster times from concept to clinic
• Improved clinical diagnostic and therapeutic
  outcomes
• Patients benefit
• Increase profit margin

49
Clinical trial in-silico

• Input
• Full body CT
• High res. CT angio
• Probabilistic pathology model

• Output 4th generation Phantom
• Accurate anatomy

50
Clinical trial in-silico

• Input
• Full body CT
• High res. CT angio
• Probabilistic pathology model

• Output 4th generation Phantom
• Accurate anatomy

51
Clinical trial in-silico

• Input
• Full body CT
• High res. CT angio
• Probabilistic pathology model

• Output 4th generation Phantom
• Accurate anatomy

52
Clinical trial in-silico

• Input
• Full body CT
• High res. CT angio
• Probabilistic pathology model

• Output 4th generation Phantom
• Accurate anatomy
• Realistic coronary pathology
• Physiology

53
Clinical trial in-silico

• Input
• Full body CT
• High res. CT angio
• Probabilistic pathology model

Incorporating a realistic coronary artery disease
model

• Output 4th generation Phantom
• Accurate anatomy
• Realistic coronary pathology
• Physiology

Gensini severity scores
54
Clinical trial in-silico

• Input
• Full body CT
• High res. CT angio
• Probabilistic pathology model

Incorporating a realistic coronary artery disease
model
Prevalence of significant stenoses (gt50) by
location and age

• Output 4th generation Phantom
• Accurate anatomy
• Realistic coronary pathology
• Physiology

Kyriakides et. al. J Clin Epidemiol Vol.48
55
Clinical trial in-silico

• Input
• Full body CT
• High res. CT angio
• Probabilistic pathology model

• Output 4th generation Phantom
• Accurate anatomy
• Realistic coronary pathology
• Physiology

56
Clinical trial in-silico

• Input
• Full body CT
• High res. CT angio
• Probabilistic pathology model

• LAMIS Virtual Cath. Lab.
• X-ray physics (Monte Carlo)?
• Virtual patient
• Virtual radiologist

• Output 4th generation Phantom
• Accurate anatomy
• Realistic coronary pathology Physiology

57
Clinical trial in-silico

• Input
• Full body CT
• High res. CT angio
• Probabilistic pathology model

• LAMIS Virtual Cath Lab
• X-ray physics (Monte Carlo)?
• Virtual patient
• Virtual radiologist

• Output 4th generation Phantom
• Accurate anatomy
• Realistic coronary pathology
• Physiology

58
Clinical trial in-silico

• Input
• Full body CT
• High res. CT angio
• Probabilistic pathology model

• LAMIS Virtual Cath Lab
• X-ray physics (Monte Carlo)?
• Virtual patient
• Virtual radiologist

• Output 4th generation Phantom
• Accurate anatomy
• Realistic coronary pathology
• Physiology

59
Clinical trial in-silico

• Input
• Full body CT
• High res. CT angio
• Probabilistic pathology model

• LAMIS Virtual Cath Lab
• X-ray physics (Monte Carlo)?
• Virtual patient
• Virtual radiologist

• Output 4th generation Phantom
• Accurate anatomy
• Realistic coronary pathology
• Physiology

60
Clinical trial in-silico

• Input
• Full body CT
• High res. CT angio
• Probabilistic pathology model

Patient Information (Blood work, imaging)?

• LAMIS Virtual Cath Lab
• X-ray physics (Monte Carlo)?
• Virtual patient
• Virtual radiologist

Virtual Cath Lab
Select optimized parameters in real time

• Output 4th generation Phantom
• Accurate anatomy
• Realistic coronary pathology
• Physiology

61
Clinical trial in-silico

• Input
• Full body CT
• High res. CT angio
• Probabilistic pathology model

Patient Information (Blood work, imaging)?

• LAMIS Virtual Cath Lab
• X-ray physics (Monte Carlo)?
• Virtual patient
• Virtual radiologist

Virtual Cath Lab
Select optimized parameters in real time

• Output 4th generation Phantom
• Accurate anatomy
• Realistic coronary pathology
• Physiology

62
Clinical trial in-silico

• Input
• Full body CT
• High res. CT angio
• Probabilistic pathology model

• LAMIS Virtual Cath Lab
• X-ray physics (Monte Carlo)?
• Virtual patient
• Virtual radiologist

• Output 4th generation Phantom
• Accurate anatomy
• Realistic coronary pathology
• Physiology

Virtual coronary angiogram
63
Conclusions

• Identified public health concern
• Identified gender differences that affect CAD
  imaging
• Built and validated virtual cath-lab
• Biological model In-silico patients
• High resolution heart models
• Statistical biological function model of heart
• Method for plaque placement
• Virtual patient
• Generated realistic virtual angiograms
• Project at final stage
• generate data and develop imaging protocol

64
Quantitative assessment of tumor drug response
with thoracic CT imaging

• Project team at Division of Imaging and Applied
  Mathematics, OSEL/CDRH/FDA
• Marios A Gavrielides
• Lisa M Kinnard
• Rongping Zeng
• Kyle J Myers
• Nicholas Petrick
• Support by FDA, NIBIB, NCI

65
Quantitative assessment of drug response with
thoracic CT imaging
66
Quantitative assessment of drug response with
thoracic CT imaging

• Biomedical imaging can provide an early
  indication of drug response, but
• many sources of uncertainty that limit the
  quantitative use of imaging as a biomarker
• variables in the image data collection platform
• robustness of software tools required for
  reliable, quantitative measurement of subtle
  changes

67
Quantitative assessment of drug response with
thoracic CT imaging

• Quantitative imaging research
• Shift from binary tasks (malignant vs benign) to
  estimation tasks (i.e. decrease in tumor size by
  25 in 3 months)
• Measurement science of estimation tasks not
  mature
• Further research is needed

68
Quantitative assessment of drug response with
thoracic CT imaging

• Estimation task
• Drug response of lung nodules
• Surrogate endpoint
• Size change estimation using thoracic CT imaging

69
Background

• Lung nodule change analysis w/thoracic CT to
• verify diagnosis, especially for small nodules
• estimate response to drug therapy
• Metrics used to describe lung nodule size
• 1-D analysis (max dimension) using RECIST
  criteria
• (replaced older 2-D analysis using WHO criteria)
• 3-D volumetric analysis

70
Background- RECIST criteria
Diameter17.7 mm
Diameter17.1 mm
Vol525.4 mm3
MSKCC DATA
Courtesy of Larry Clarke, NCI
71
Background- nodule size metrics

• RECIST criteria
• based on assumption of spherical nodules
• but majority of nodules grow irregularly
• Volumetric 3D analysis may be more accurate

72
Background- nodule size metrics

• New generation CT scanners
• Fast gantry rotation
• Acquire full scan in single breath hold at lt1mm
• Near-isotropic data
• Improved volumetric analysis

73
Diameter17.7 mm
Vol525.4 mm3
Diameter17.1 mm
MSKCC DATA
Courtesy of Larry Clarke, NCI
74
Volumetric analysis of lung nodules

• Potential for quantitative volumetric analysis of
  CT imaging to impact
• accurate assessment of change
• patient management
• reduce time needed to assess change
• Key issues
• accuracy/reliability of volumetric measurements
• is a 20 decrease in lung nodule volume after a
  week of drug therapy due to actual nodule
  response or measurement error?

75
Volumetric analysis of lung nodules

• Factors influencing volumetric measurement
  accuracy
• Image acquisition and reconstruction parameters
• Slice thickness, slice collimation, tube current,
  pitch, reconstruction algorithm/kernel
• Inherent variability of scanner
• Nodule characteristics
• Size, shape, density, location, vascular/pleura
  attachments
• Volume measurement method
• Manually-drawn boundaries, different segmentation
  algorithms
• Which parameters contribute the most to
  measurement error?

M A Gavrielides, L M Kinnard, et. al, Volumetric
methods for the assessment of lung nodules with
thoracic CT, Radiology, (in press)
76
Image acquisition and reconstruction parameters

• Effect of slice thickness on measurement error
• For 38-mm diameter (synthetic)
• 11.2 (slice thickness 2 mm) to 16.4 (10 mm)
• For 13-mm diameter
• 13.0 (slice thickness 2 mm) to 28.0 (10 mm)
• Spherical, homogeneous synthetic nodules, no
  attachments

Winer-Muram, H.T., et. al Effect of varying CT
section width on volumetric measurement of lung
tumors and application of compensatory
equations. Radiology, 2003. 229(1) p. 184-194.
77
Nodule characteristics

• Decreasing size increases error
• Measurement error ranging from
• 1.0 for nodule volume of 268 cm3
• 28.6 for 3.5 cm3.
• Error also up for irregular shapes

Ko, J.P., H. Rusinek,et. al, Small pulmonary
nodules Volume measurement at chest CT- Phantom
study. Radiology, 2003. 228(3) p. 864-870.
78
Volume measurement method
Courtesy of Larry Clarke, NCI
Meyer, C.R.,.et. al, Evaluation of lung MDCT
nodule annotation across radiologists and
methods. Academic Radiology, 2006. 13(10) p.
1254-1265.
79
Volume measurement method

• Based on manually drawn boundaries
• large inter-observer variability exits
• Based on software tools
• reduced variance but bias present
• (bias not predictable across lesions, scans,
  etc.)
• Both bias and variance error analysis should be
  performed before choosing software method

Meyer, C.R.,.et. al, Evaluation of lung MDCT
nodule annotation across radiologists and
methods. Academic Radiology, 2006. 13(10) p.
1254-1265.
80
Experimental approaches to examine volumetric
estimation accuracy

• Phantom studies w/ synthetic nodules
• known volume truth
• can investigate range of variables such as tube
  current, w/out concern for patient exposure
• provide lower bound on volume estimation error
  (bias and variance)
• Simulation studies
• generation of images with realistic properties by
  tracking the transport of particles from the
  source to the object of interest.
• Analysis of clinical nodules
• the lunch break experiment on repeat scans
• provides estimates of measurement variance

Badano, A. and J. Sempau, MANTIS combined x-ray,
electron and optical Monte Carlo simulations of
indirect radiation imaging systems. Physics in
Med. and Biol
81
Project Objective

• Identify and quantify sources of nodule size
  estimation error
• 3D (volume), 1D (RECIST), and 2D (WHO)
  measurements
• Realistic phantom studies with systematic
  approach
• Knowledge of dependence of bias/variance to
  specific parameters will
• assist in development of standards for imaging
  protocols/software tools
• impact the adoption of quantitative imaging
  approaches in the design, development and
  evaluation of drugs

82
Project Plan

• Thoracic CT phantom studies
• Realistic phantom
• Multislice helical CT scanner
• Synthetic lung nodules of varying size, shape,
  density, location
• Different truthing approaches
• Extract estimates of nodule size
• Analyze bias/variance of estimation error

83
Lung CT phantom studies

• Anthropomorphic phantom from Kyotokagaku, Japan
• lung vasculature
• synthetic nodules
• allows for studies on realistic scenario of
  nodule attachments

84
Lung CT phantom studies

• CT scanner
• Philips 16-slice Mx8000
• FDA Center for Veterinary Medicine (CVM)
• Collaboration with Drs. Pritchard and Karanian
  (Dr. Wood/NIH)

85
Lung CT phantom studies

• Nodule sets
• Spherical nodules (Kyotokagaku)
• 0.3-1.2 cm
• -800, -630, 100 HU
• Larger spherical set from CIRS
• 2.0, 4.0, 6.0cm
• -800, -630, 100 HU

86
Lung CT phantom studies

• Nodule sets
• Aspheric nodules (CIRS)
• Elliptical shapes
• Spiculated nodules
• Lobulated nodules

87
Lung CT phantom studies

• Volume truthing approaches
• Water displacement method
• Micro-CT imaging
• Collaboration with NIST (S. Lin-Gibson, C.
  Fenimore), U of Iowa (G. McClennan), NIH
• Collaboration with Wash U.
• Cartesian coordinate machine
• Collaboration with NIST (S. Philips, C. Fenimore)
• Robotic probe taking surface samples, surface fit

surface rendering of micro_CT scan
88
Lung CT phantom studies

• Volume measurement methods
• Use available 3D software tools
• publicly available OSIRIX
• commercial 3D Doctor
• research tools from collaborators level-set
  segmentation (NIH)
• All are 3D-segmentation based
• Difficult to assess effect of variables due to
  large segmentation errors

1L.M. Kinnard,et al Volume Error Analysis for
Lung Nodules Attached to Bronchial Vessels in an
Anthropomorphic Thoracic Phantom, SPIE Medical
Imaging 2008, vol. 6915, 69152Q, 2008.
89
Lung CT phantom studies

• Volume measurement methods
• Develop 3D matched filter
• Use simulation to derive models of system effect
  along object_to_image transformation
• Apply models to correct for acquisition/reconstruc
  tion parameters before measurement

90
Image acquisition and reconstruction parameters
variable tube current
20mAs Volume660mm3
200mAs Volume830mm3
10mm nodule Volume524mm3
91
Image acquisition and reconstruction variance on
repeat scans
92
Lung CT phantom studies

• Matching of scanned nodule volumes to banks of
  simulated 3D nodule templates to estimate volume

93
Nodule layouts

• Preliminary scans to determine materials to
  secure nodules
• surgical suture prolene 5.0

94
Phantom Study Deliverables 1

• Public Image Database
• Phantom thoracic scans
• Across range of imaging protocols, nodules sizes,
  shapes, attachments
• Associated truth of nodule volume
• Available to software developers and researchers
• Allow for determination of bias and variance
• Understanding/quantifying sources of error
• Related to size measures (volume, RECIST, WHO)
• Relative contribution to the error of different
  sources (imaging parameters, nodule
  characteristics, software tools)

95
Phantom Study Deliverables 2

• FDA Recommendations for guidance
• For techniques used to design clinical trials
  utilizing imaging as a biomarker
• For computer-aided analysis tools used in
  quantitative assessment of medical images (used
  by reviewers of submissions to CDRH/CDER)
• For phantoms and image acquisition protocols

96
Summary

• Potential for improved volumetric measurements
  tools
• Increased/improved use of quantitative imaging in
  drug trials and monitoring
• Recommendations to software developers and drug
  companies on which variables should be controlled

97
Summary

• Improving the precision and accuracy of drug
  response analysis
• could significantly reduce the size and cost of
  clinical trials for drug response
• could assist in increase the use of quantitative
  imaging in the drug development process
• Quantitative imaging would provide a step towards
  personalized medicine

98
Computer-aided assessment of HER2/neu
immunohistochemical expression in breast cancer
An observer study
99
Study objective

• The objective of this study is to assess the
  benefit of a computer-aid in reducing inter- and
  intra-observer variability in the evaluation of
  the biomarker HER2/neu for breast cancer

100
Background

• About 25 of breast cancer patients over-express
  the protein HER2/neu
• These patients respond well to adjuvant treatment
  with the drug herceptin which was shown to reduce
  risk of recurrence by 1/2 and mortality by 1/3
  (Wolff, 2007)
• However
• Need to consider side effects (including heart
  dysfunction) and cost (50K-100K/year)
• Need accurate and reproducible assessment of HER2
  expression to find good responders and guide
  treatment

101
Background

• Guidelines for HER2 management
• Examine HER2-stained tissue with
  immunohistochemistry (IHC)
• Categorize each case as 1, 2, or 3 (3 highest
  expression)
• Follow-up
• 3 positive gt herceptin treatment
• 2 ambiguous gt further testing (FISH imaging)
• 1 negative gt no herceptin treatment

102
HER-2/neu scoring using IHC

• Evaluation based on color stain assessment of
  the cell membrane
• membrane staining completeness
• is staining around the whole circumference?
• membrane staining intensity
• ranging from faint to strong brown

103
Guidelines for HER-2/neu scoring

• From College of American Pathologists
• Similar guidelines from staining kit
  manufacturers

based on whole slide
104
Examples of regions of interest extracted from
slides scored as 3 (positive)
Strong staining around the whole membrane for a
significant number of cells
Note evaluation is performed only on epithelial
cells not this area of connective tissue
105
Examples of regions of interest extracted from
slides scored as 3 (positive)
more positive ..? Membrane staining is more
intense and more complete than previous slide
106
Examples of regions of interest extracted from
slides scored as 1
Faint staining in a few cells
107
Issues related to HER2 evaluation

• Studies have shown significant inter- and
  intra-observer variability in HER2 interpretation
• 48 correct agreement , 5 observers (Hsu, Am J
  Clin Pathol, 2002)
• Sources of variability include
• subjective criteria for color staining assessment
• different ways observers combine scores from
  different fields of view
• differences in pathologist experience, training
• Computer-aided assessment has the potential to
  decrease observer variability in the evaluation
  of IHC

108
Computer-aided assessment of HER2

• We have developed a computer system for the
  automated quantitative assessment of HER21
• Provides quantitative measures of both
• membrane staining completeness and
• membrane staining intensity
• which are the main features used in HER2
  assessment
• Preliminary results1 showed good agreement with
  pathologist-provided HER2 scores

1H Masmoudi, S Hewitt, K J Myers, N Petrick, and
M A Gavrielides, Automated quantitative
assessment of HER-2/neu immunohistochemical
expression in breast cancer, IEEE Trans Med Imag
(accepted).
109
Observer study

• Main hypotheses to be tested
• computer-aided assessment of IHC can decrease
  intra- and/or inter-observer variability in the
  assessment of HER2 for breast cancer
• computer-aided assessment of IHC can be used as a
  training tool to increase agreement between
  novices and trained pathologists in the
  evaluation of HER2 microscopy images from breast
  cancer tissue specimens

110
Computer-output for observer study

• Prompt displays the computer-extracted membrane
  staining intensity and completeness values for
  the current ROI relative to values of training
  ROIs
• Location of prompt closer to the upper right
  corner (brown cluster) indicates higher HER2
  expression
• Location of prompt closer to the lower left
  corner (blue cluster) indicates lower HER2
  expression

• Color points display computer-extracted membrane
  staining intensity and completeness distribution
  of training images
• Note classification in 1, 2 and 3 categories
  based on provided slide scores

111
3D Breast Imaging

• This program studies the many clinical challenges
  in conventional mammography that may be addressed
  by 3D breast imaging
• Overlapping breast tissue can mask a tumor
  (false negative).
• Overlapping breast tissue can create the
  appearance of tumor (false positive).
• Lesion margins can be obscured, hindering
  differentiation between malignant and benign
  lesions.
• Precise localization of lesion can be difficult.

detector

• The potential benefits of 3D breast imaging
  include the following
• Revelation of overlapping structures
• Reduction of summation shadows
• Improved visualization of lesion margins
• Accurate 3D localization of lesions
• Reduced call-backs

breast
Multiple potential system configurations for
acquiring 3D breast data.

• Our project addresses these unanswered questions
• How to optimize the numerous system design
  parameters for the individual patient?
• How to best make the trade-off between
  diagnostic performance and radiation dose?
• How to process sequence of images to create 3D
  imagery?
• How to display the images?
• What are the best methods for 3D computer-aided
  diagnosis (CAD) compared with standard
  mammography?

112
In-Silico ImagingLet your computer do the walking

• Faster
• Cheaper
• Safer
• Better