EXPLORING LUNG MICROSTRUCTURE WITH MONTE CARLO Annie Malkus

EXPLORING LUNG MICROSTRUCTURE WITH MONTE CARLO Annie Malkus from the Fain Lab at University of Wisconsin-Madison HTCondor Week, May 2018, Madison, WI 1

2 LUNG STRUCTURE Eric, P. W. , Hershel Raff, and Kevin T. Strang. "Vander’s Human Physiology, The Mechanism of Body Function. " (2014).

3 LUNG STRUCTURE Eric, P. W. , Hershel Raff, and Kevin T. Strang. "Vander’s Human Physiology, The Mechanism of Body Function. " (2014).

4 CT Proton MRI Structure HP 3 He MRI Function

LUNG MICROSTRUCTURE Normal ~100 μm Emphysema Slide content courtesy of Sean Fain 5

LUNG MICROSTRUCTURE Normal Asthma ? ~100 μm Emphysema Slide content courtesy of Sean Fain 6

7 LUNG MICROSTRUCTURE 100 μm Slide content courtesy of Sean Fain

8 LUNG MICROSTRUCTURE MC can help us probe these scales 100 μm Slide content courtesy of Sean Fain

9 HYPERPOLARIZED GAS Smoker (44 pk*yrs) Smoker (40 pk*yrs) Slide content courtesy of Sean Fain et al. Radiology 2006 Ventilation Normal 0 0. 6 cm 2/s 0 0. 7 cm 2/s

10 HYPERPOLARIZED GAS Smoker (44 pk*yrs) Smoker (40 pk*yrs) Fain et al. Radiology 2006 Ventilation Normal Still not probing small airway scales… Slide content courtesy of Sean Fain 0 0. 6 cm 2/s 0 0. 7 cm 2/s

11 MATERIAL AND METHODS Breathing apparatus (in action) Slide content courtesy of David Mummy Helium Hyperpolarizer

12 CT Proton MRI Structure HP 3 He MRI Function

13 MAGNETIZATION ACCUMULATES PHASE BASED ON STRENGTH OF MAGNETIC FIELD

14 MAGNETIZATION ACCUMULATES PHASE BASED ON STRENGTH OF MAGNETIC FIELD This is how we probe small airway scales…

15 DIFFUSION

16 DIFFUSION B(b) G t

17 DIFFUSION B(b) G t

18 DIFFUSION

19 DIFFUSION B(b)

20 DIFFUSION B(b)

21 DIFFUSION

22 HYPERPOLARIZED GAS Smoker (44 pk*yrs) Smoker (40 pk*yrs) Slide content courtesy of Sean Fain et al. Radiology 2006 Ventilation Normal 0 0. 6 cm 2/s 0 0. 7 cm 2/s

23 HYPERPOLARIZED GAS Smoker (44 pk*yrs) Smoker (40 pk*yrs) ADC Map Fain et al. Radiology 2006 Ventilation Normal Slide content courtesy of Sean Fain 0 0. 6 cm 2/s 0 0. 7 cm 2/s

24 HOW WE USE MC TO MAKE MEASUREMENTS IN MR • Run simulations over scales in regime of interest

25 HOW WE USE MC TO MAKE MEASUREMENTS IN MR • Run simulations over scales in regime of interest • Compute signal with the experiment’s actual scan parameters S R

26 HOW WE USE MC TO MAKE MEASUREMENTS IN MR • Run simulations over scales in regime of interest • Compute signal with the experiment’s actual scan parameters • Fit MC to model S R

27 HOW WE USE MC TO MAKE MEASUREMENTS IN MR • Run simulations over scales in regime of interest • Compute signal with the experiment’s actual scan parameters • Fit MC to model • Compare data to model S R

28 DIFFUSION—MONTE CARLO (MC)

29 DIFFUSION MC • Matlab • Build geometry • Populate with particles • C++ • Random walks • 106 -107 particles • ~103 time steps • 1000 jobs at a time

30 DIFFUSION MC • Matlab • Build geometry • Populate with particles • C++ • Random walks • 106 -107 particles • ~103 time steps • 1000 jobs at a time Two days of organizing and running.

31 DIFFUSION MC • Matlab • Build geometry • Populate with particles • C++ • Random walks • 106 -107 particles • ~103 time steps • 1000 jobs at a time • Limitation: • Space • Organization • (DAG)

32 LUNG MICROSTRUCTURE

LUNG MICROSTRUCTURE Normal Asthma ? ~100 μm Emphysema Slide content courtesy of Sean Fain 33

LUNG MICROSTRUCTURE ? Slide content courtesy of Sean Fain 34

35 GEOMETRY

36 GEOMETRY

37 GEOMETRY

38 SOME MODEL APPLICATION

39 SUMMARY • Hyperpolarized Gas MR can show structure of lungs • Diffusion imaging with hyperpolarized gas can reveal scales • MC can build quantitative measures of scales • MC can inform the design of future scan sequences targeting regimes of interest.

40 WISH LIST AND FUTURE WORK • Markhov Chain Monte Carlo Fitting • DAG • (1000 steps)x(100 chains) per pixel • (128 x 16) pixels • Large file transfers • (128 x 16)x 10 doubles per subject. • K-Space Fitting • 1 hour per subject • Fits before image reconstruction with raw data • Fourier transforms (matrix multiplication, other linear operations) • Matlab • Gradient Descent family of algorithms • 64 x 64 complex double images per subject • 500, 000 data points

41 ACKNOWLEDGEMENTS Lung Group: • Sean Fain, Ph. D • Scott Nagle, MD, Ph. D • Bob Cadman, Ph. D • Wei Zha, Ph. D • Andrew Hahn, Ph. D • Greg Barton, Ph. D • David Mummy, MS • Jeffrey Kammerman, MS • Luis Torres, MS • Katie Carrey, MS UW Collaborators: • Kai Ludwig, MS • Erin Adamson, MS UW Staff: • Gina Crisafi • Jan Yakey, RN, CCRC • Kelli Hellenbrand • Jenelle Fuller • Sara John • Molly Ellertson

43

44 B 0

45 M B 0

46 M B 0

47 M B 0 B 1

48 B 0 M

49 B 0 M

50 B 0 M Signal!

51 HYPERPOLARIZED GAS Smoker (44 pk*yrs) Signal! Slide content courtesy of Sean Fain 0 0. 6 cm 2/s 0 0. 7 cm 2/s Fain et al. Radiology 2006 Smoker (40 pk*yrs) Ventilation Normal

52 B 0 M

53 B 0 M

54 B 0