Dr. Thomas Yankeelov: Integrating Advanced Imaging and Biophysical Models to Predict Tumor Growth

Tom Yankeelov, Ph.D.

Ingram Associate Professor of Cancer Research Institute of Imaging Science

Departments of Radiology, Biomedical Engineering, Physics, and Cancer Biology

Vanderbilt University

19 June 2013

Integrating Advanced Imaging and Biophysical Models to Predict Tumor Growth

2

Weather models use observations of the atmosphere to predict how wind, temperature, and humidity evolve in time

DW-MRI Cell Number

DCE-MRI Perfusion

FDG PET Metabolism

Noninvasive, quantitative imaging enables the same approach for predicting how tumors evolve in time

Yankeelov et al. Sci Transl Med. 2013;5:187ps9

Visit 2 25 15 40

+5%

Visit 3 28 16 44

+16%

Visit 4 32 18 50

+32%

Visit 5 48 23 71

+89%

Baseline 24 14 38

T1: T2:

SLD: (% )

Courtesy of Rick Abramson, M.D.

8 weeks 16 weeks

24 weeks

T2 T1

3

4

Working hypothesis:

Readily-available, multi-scale imaging techniques can provide the data to initialize/constrain predictive models of tumor

growth and treatment response for clinical application.

5

Outline

1.  What can quantitative imaging offer oncology?

2.  Specialize to neoadjuvant therapy

3.  Lessons from weather forecasting

4.  Towards a science of tumor forecasting

6

Outline





7

Magnetic resonance imaging (MRI)

8

Diffusion weighted MRI

•  Boundaries may reduce distance molecules travel when compared to free molecules

•  Thus, the Apparent Diffusion Coefficient (ADC) is lowered

~√t

Distance from

original position

Free

Restricted

• Water molecules wander about randomly in tissue (Brownian Motion)

• In a free solution, after a time t, molecules travel (on average) a distance L from where they started

• But in tissue, compartment effects may hinder movement = restricted diffusion

9

• Increasing cell density (cellularity); more cell membranes per unit distance to hinder diffusion lower ADC

AD

C

•  ADC depends on cell volume fraction

•  Tumor cellularity may be monitored by DWI Hall et al. Clin Canc Res 2004;10:7852 Anderson et al. Magn Reson Imaging. 2000;18:689-95.

Diffusion weighted MRI

10 Arlinghaus et al.

Diffusion weighted MRI, clinical example

Responder ΔADC = 11.8%

Non-responder ΔADC = -11.6%

Pre-NAC Post-1 cycle

ROC Analysis Sensitivity = 0.64 Specificity = 0.93

AUC = 0.70

  Sensitivity = true positive rate = TP/(TP+FN)

  Specificity = true negative rate = TN/(FP + TN)

11

Dynamic contrast enhanced MRI (DCE-‐MRI)

•  Each voxel yields a signal intensity time course that can be analyzed with a mathematical model

•  By fitting data to model, extract parameters that report tissue characteristics

•  Serial acquisition of images before, after an injection of contrast agent (CA)

•  As CA perfuses into tissue, changes the measured signal intensity

plasma

space

tissue space Ktrans

Ktrans/ve

Cp(t)

Ct(t)

Ktrans = transfer rate constant

ve = extravascular extracellular volume fraction

Tofts, et al. J Magn Reson Imaging 1999;10:223-‐232.

12

Sens

itivi

ty

ROC Analysis Sensitivity = 0.81 Specificity = 0.75

AUC = 0.80

When combining DW-MRI & DCE-MRI data: Sensitivity = 0.88 Specificity = 0.82

AUC = 0.86

Pre-therapy Post-1 cycle Post therapy

Responder

Non-Responder

DCE-‐MRI, Clinical Example

Li et al. Magn Reson Med 2013 (epub ahead of print; Mani et al. J Am Med Inform Assoc. 2013;20:688-95.

13

Positron emission tomography (PET)

14

18FDG

(blood)

18FDG

(tissue)

18FDG-6-PO4

(cells) X

Pre-NAC Post-1 cycle Post-all NAC

Non-responder

pCR

Li et al. Eur J Nuc Med &Mol Imaging Res 2012;2:62

Pre-NAC

extravascular extracellular volume fraction

plasma volume fraction

cellularity

FDG-PET (glucose metabolism)

blood perfusion and permeability

Post-1 cycle

Post-all NAC

Xia Li, Nkiruka Atuegwu, Lori Arlinghaus

16

•  Dramatic increases in quality of data available from non-invasive imaging

Imaging is moving from qualitative anatomical measurements to quantitative functional measurements at physiological, cellular, & molecular levels

•  We talked about:

MRI—anatomy, blood vessels, blood flow, cellularity

PET—metabolism

Imaging Summary

•  Other imaging measurements we did not talk about:

cell proliferation, pH, pO2 (MRI)

Receptor expression, apoptosis (PET & SPECT)

What is the best way to use such data to assist oncology?

17

Outline





18

•  Pre-operatively treated locally advanced cancers chemo-, radiation, endocrine, and targeted therapies

•  Five theoretical advantages to NAT:

1)  Earlier treatment of occult metastatic despite “curative” resection

2)  Determine sensitivity to treatment while tumor is in situ

3)  Complete a full course of treatment (more likely pre-operatively than post-operatively)

4)  Identify patients who develop metastases on NAT as they are unlikely to benefit from resection

5)  Decrease primary tumor volume

Sener. JSO 2010;101:282

19

•  Key point: several disease sites where if you can get the patient to respond in the neoadjuvant setting improved outcome

•  How to identify non-responders from responders early in therapy?

Develop predictive models that use data obtained from individuals

•  But more than that—would enable patient specific “clinical trialing”

  Can perform individualized, in silico clinical trials to optimize the drug regiment, order, timing, dose, etc.

  Models must make specific predictions

20

Outline





21

•  In order to have meaningful weather predictions, needed: 1) Rudimentary understanding of atmospheric dynamics 2) Regular radiosonde measurements (...and then satellite data) 3) Stable numerical methods 4) Electronic computers

•  100 years ago, none of this existed

Lynch. J Computational Physics 2008;227:3431-44.

“A century ago, weather forecasting was a haphazard process, very imprecise and unreliable. Observations were sparse and irregular, especially for the upper air and over the oceans. The principals of theoretical physics played little or no role in practical forecasting: the forecaster used crude techniques of extrapolation, knowledge of local climatology and guesswork on intuition; forecasting was more an art than a science.”

22

•  So how did meteorology make such dramatic advances? Advances in atmospheric physics, numerical methods, computing

Better data!

Lynch. J Computational Physics 2008;227:3431-44.

National Climate Data Center NEXRAD (NEX generation RADar) Data Sites

~160 sites

23

Outline





24 Preziosi. Cancer Modeling and Simulation.

•  Let the tumor cells proliferate up to a certain “carrying capacity” = "

•  Solution is given by:

•  The last equation states that the tumor cells will continue to grow (exponentially) up to the carrying capacity of the system determined by "

25 Atuegwu et al. Transl Oncol. 2013;6:256-64

Experimental Simulated r = 0.90

Experimental

Sim

ulat

ed

Pearson’s correlation = 0.83 (p = 0.01) Concordance correlation = 0.80

Nsimulated(t3) x106

Nex

peri

men

tal (

t 3) x

106

Some summary stats (n = 27):

•  k (proliferation rate) separated pCR and non- pCR after 1 cycle of NAC (p = 0.019)

sensitivity = 82%, specificity 73%

26

•  Going forward, need to make greater use of available data

•  An example mathematical model :

Random dispersal of tumor cells (diffusion)

Proliferation of tumor cells

Rate of chance of # of

tumor cells

  ADC values from DW-MRI to assign NTC(r,t) and extract k(r)

  Everything on the right hand side is known

27

•  Going forward, need to make greater use of quantitative, multi-modality data





tumor cells

Rate of change of # of

endothelial cells

Diffusion of EC Chemotaxis of EC

Assume chemotaxis is in direction of areas of proliferating cells of higher density; can

also estimate this from DW-MRI data

28




endothelial cells


Rate of change of O2




tumor cells

Perfusion data from DCE-‐MRI

ADC & PET data



Rate of change of O2

Rate of change of glucose


endothelial cells





tumor cells

Experimental system -- C6 glioma

Anatomical (Registration)

DW-MRI Cell Number

DCE-MRI Ktrans,ve, and vp

18F-FDG PET

MRI on days 9, 10, 11, 13, 15, and 17 PET on days 9, 15, and 17

Hormuth et al. 2013 Oak Ridge National Lab BSEC conference.

Predicted tumor cell number

Observed tumor cell number

Hormuth et al. 2013 Oak Ridge National Lab BSEC conference.

32

•  Having a model, driven by patient specific data would enable personalized, in silico therapy modeling theoretical/predictive oncology

•  Could “give” the patient therapy in silico, then see how they “respond”

Could systematically adjust therapies, order of combination therapy, dosing scheduling, etc.

Since the quantitative imaging data can be acquired in 3D, at multiple time points and noninvasively, it is the only game in town

Could enable (more) rational clinical trials design/execution

Eminently testable in pre-‐clinical setting… and is translatable

33

Squad •  Lori Arlinghaus, PhD •  Nkiruka Atuegwu, PhD •  Richard Baheza, MS •  Stephanie Barnes, PhD •  Jacob Fluckiger, PhD •  David Hormuth, BS •  Xia Li, PhD •  Mary Loveless, PhD •  David Smith, PhD •  Jared Weis, PhD •  Jennifer Whisenant, MS •  Jason Williams, PhD

Collaborators •  Vandana Abramson, MD •  Bapsi Chak, MD •  Ingrid Mayer, MD •  Mark Kelley, MD •  Brian Welch, PhD •  Rick Abramson, MD •  Mike Miga, PhD •  Vito Quaranta, MD

Funding •  NCI U01 CA142565 •  NCI R01 CA138599 •  NCI R01 CA129661

•  NCI R25 CA092043 •  NCI R25 CA136440 •  NCI P30 CA68485 & VICC

Very sincere thank you to the women who participate in our studies.

Technology

Dr. Thomas Yankeelov: Integrating Advanced Imaging and Biophysical Models to Predict Tumor Growth