DeepLumen Fast and Accurate Segmentation of Coronary Arteries for - - PowerPoint PPT Presentation

▶
deeplumen fast and accurate segmentation of coronary
SMART_READER_LITE
LIVE PREVIEW

DeepLumen Fast and Accurate Segmentation of Coronary Arteries for - - PowerPoint PPT Presentation

Feb 02, 2024 145 likes •718 views

DeepLumen Fast and Accurate Segmentation of Coronary Arteries for Improved Cardiovascular Care Kersten Petersen, Michiel Schaap, David Lesage, Matthew Lee, and Leo Grady GTC 2017 How do we find the right treatment for patients with symptoms of

slide-1
SLIDE 1

DeepLumen Fast and Accurate Segmentation of Coronary Arteries for Improved Cardiovascular Care

Kersten Petersen, Michiel Schaap, David Lesage, Matthew Lee, and Leo Grady GTC 2017

slide-2
SLIDE 2

2

How do we find the right treatment for patients with symptoms of coronary artery disease (CAD)?

Image from Cardiac Health

slide-3
SLIDE 3

Agenda

DeepLumen HeartFlow Analysis Coronary Artery Disease

slide-4
SLIDE 4
  • 1/3 of all global deaths are from CAD
  • 500k people in U.S. die of a fatal heart

attack each year2

  • >$200B spent each year in the US alone
  • n the diagnosis and treatment of patients

with CAD2

4

Coronary Artery Disease (CAD)

1. Atlas of Heart Disease and Stroke, WHO, 2004 2. Heart Disease and Stroke Statistics--2011 Update : A Report From the American Heart Association, Circulation 2011, 123:e18-e209

slide-5
SLIDE 5

Coronary Artery Disease (CAD)

5

Plaque in the artery walls can obstruct the blood flow to the heart.

Images from Texas Heart and Lakeland Health

Lumen

(inner part of artery)

Plaque

(calcium, fat, cholesterol, fibrin, cellular waste)

slide-6
SLIDE 6

6

Treatment Options

Obstructive Disease

Stent Bypass

Non-obstructive Disease

Medication Lifestyle Changes

Images from news.com.au and Wikipedia

slide-7
SLIDE 7

7

Usual Clinical Pathway

Non-invasive Tests Invasive Cath Lab

Lifestyle Changes or Medication Stent or Bypass Patient with non-obstructive disease Patient with

  • bstructive

disease

slide-8
SLIDE 8

8

THE PROBLEM Currently, many patients are unnecessarily sent to the Cath Lab, where they face an expensive, time-consuming, and invasive test.

Image from MedStar Franklin Square Medical

slide-9
SLIDE 9

Agenda

DeepLumen HeartFlow Analysis Coronary Artery Disease

slide-10
SLIDE 10

Coronary Artery Disease

Agenda

HeartFlow Analysis DeepLumen

slide-11
SLIDE 11

11

HEARTFLOW ANALYSIS A non-invasive CAD test that is more accurate than existing non-invasive CAD tests.

slide-12
SLIDE 12

12

HeartFlow’s Clinical Pathway

HeartFlow Analysis Invasive Cath Lab

Lifestyle Changes or Medication Stent or Bypass Patient with

  • bstructive

disease far fewer Patient with non-obstructive disease

slide-13
SLIDE 13

13

IDEA Calculate fractional flow reserve.

slide-14
SLIDE 14
  • Most precise CAD test (gold standard) for making treatment decision.
  • However, FFR is invasive, expensive, and time-consuming.

14

Fractional Flow Reserve (FFR)

1. De Bruyne et al., NEJM 201 2. Pills et al., JACC 2007 3. Tonino et al., NEJM 2009

Proximal Pressure (Pa) Distal Pressure (Pd)

FFR = Pd Pa

slide-15
SLIDE 15

15

HeartFlow Analysis

CT data submitted Anatomic model Physiologic model Functional assessment with Computational Fluid Dynamics HeartFlow Analysis delivered

slide-16
SLIDE 16

16

Comparison to FFR (Gold Standard)

Specificity 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Sensitivity 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

HeartFlow Analysis All other existing CAD Tests

1. Koo et al., JACC 2011 2. Min et al., JAMA 2012 3. Norgaard et al., JACC 2014 4. Norgaard et., Eur Radiol 2015

slide-17
SLIDE 17

17

PLATFORM Trial

No clinically adverse events

HeartFlow Analysis

  • Cost savings of 26% to the health system after

accounting for a $1,500 cost of the HeartFlow Analysis

  • Prospective multi-center clinical trial

1. Douglas et al., EHJ 2015 2. Hlatky et al., JACC 2015 3. Douglas et al., JACC 2016

slide-18
SLIDE 18
  • CE Mark (07/2011)
  • De novo 510(k) FDA

clearance (11/2014)

  • Regulatory

approval in Japan (11/2016)

  • NICE Guidance

recommends HeartFlow Analysis (02/2017)

18

The HeartFlow Analysis

  • >150 peer-reviewed

publications

  • >150 issued and

allowed patents worldwide

  • HeartFlow announces

collaboration with Siemens Healthineers to offer an integrated solution for noninvasive assessment of CAD (announced 03/2017)

slide-19
SLIDE 19

19

Commercial Use

Over 10,000 patients have received the HeartFlow Analysis worldwide

slide-20
SLIDE 20

Coronary Artery Disease

Agenda

HeartFlow Analysis DeepLumen

slide-21
SLIDE 21

Coronary Artery Disease

Agenda

DeepLumen HeartFlow Analysis

slide-22
SLIDE 22

22

The Analysis Process

CT data submitted Anatomic model Physiologic model Functional assessment with Computational Fluid Dynamics HeartFlow Analysis delivered

slide-23
SLIDE 23

23

CT Data to Anatomic Model

CT data submitted Anatomic Model Heart & Large Structures Vessel Paths Lumen

slide-24
SLIDE 24

24

Quality Control

Anatomic Model Heart & Large Structures Vessel Paths Lumen

Manual correction by
 certified experts

slide-25
SLIDE 25

25

Challenges

Top: Lumen CT data, Bottom: (Pixel) annotations of lumen

slide-26
SLIDE 26

26

Calculating Lumen: Input

CT Data Vessel Paths

slide-27
SLIDE 27

27

Calculating Lumen: Curved Planar Representation (CPR)

Vessel Paths Single Vessel (CPR) Single Vessel (CT volume)

slide-28
SLIDE 28

28

Calculating Lumen: Curved Planar Representation (CPR)

Vessel Paths Single Vessel (CPR) Single Vessel (CT volume)

Frame Optimization

slide-29
SLIDE 29

29

Calculating Lumen: Curved Planar Representation (CPR)

Image (CPR) Landmarks (CPR)

GPU accelerated 
 PREDICTION

slide-30
SLIDE 30

30

Calculating Lumen: Mesh Output

Landmarks (CPR) Landmarks (CT volume) Mesh (CT volume)

GPU accelerated 
 SURFACE RECONSTRUCTION

slide-31
SLIDE 31

31

The Focus of this Talk

Image (CPR) Landmarks (CPR)

slide-32
SLIDE 32

32

Pixel Classification? No!

PROBLEMS

  • 1. Can produce spurious components.
  • 2. Can produce holes in segmentation.
  • 3. No sub-voxel accuracy
slide-33
SLIDE 33

33

Regression

Regress distances from vessel path point (red) to vessel boundary at fixed angles.

slide-34
SLIDE 34

34

Rotational Symmetry

Frame

slide-35
SLIDE 35

35

Ring Representation

slide-36
SLIDE 36

36

Ring Representation

slide-37
SLIDE 37

37

Ring Representation

slide-38
SLIDE 38

38

Unfold the Ring

1) Concatenate frames. 2) Predict distance from vessel path (red) to upper vessel boundary.

slide-39
SLIDE 39

39

Unfold the Ring

Apply the same model rotationally to predict one distance at a time. , …, , …

slide-40
SLIDE 40

40

Cyclic Padding

Cyclic Padding r 2r

slide-41
SLIDE 41

41

Cyclic Padding

Cyclic Padding r 2r

slide-42
SLIDE 42

42

Extension to 3D

Frames of CPR Longitudinal Slice of CPR

slide-43
SLIDE 43

43

3D Ring Representation

slide-44
SLIDE 44

44

3D Ring Representation

Concatenate frame to padded unfolded 3D ring w 2r h

slide-45
SLIDE 45

45

DeepLumen Prediction

r distance predictions

3D regression CNN 3D feature map: w * h * 2r (padded unfolded 3D ring)

slide-46
SLIDE 46
  • Given the vessel line and the image (5123 voxels), Deep

Lumen takes 8 seconds to segment all coronary arteries using a GeForce GTX Titan X.

46

Speed

slide-47
SLIDE 47

47

Comparison to State-of-the-Art 3D CNN

3D U-Net

Ronneberger, et al. MICCAI 2016

DeepLumen

slide-48
SLIDE 48

48

Comparison to Ground Truth

DeepLumen Ground Truth

slide-49
SLIDE 49
  • 1500 training images and 1500 testing images
  • We evaluated the median shortest distance between the

mesh from automated DeepLumen and the ground truth mesh.

  • Healthy regions: 0.08 mm


Diseased regions: 0.10 mm

  • Resolution of CT: ≈ 0.40 mm

49

Geometric Error

slide-50
SLIDE 50
  • IVUS: Intravascular Ultrasound (Resolution: ≈ 0.10 mm, invasive)
  • OCT: Optical Coherence Tomography (Resolution: ≈ 0.02 mm, invasive)

50

Validating Minimal Lumen Area

slide-51
SLIDE 51

51

Comparison of Minimal Lumen Area (MLA)

Study Comparison Expert / Auto r Leber et al. 2005 CT vs IVUS Expert 0.54 Caussin et al. 2006 CT vs IVUS Expert 0.88 Voros et al. 2011 CT vs IVUS Expert 0.65 Boogers et al. 2012 CT vs IVUS Expert 0.75 De Graaf et al. 2013 CT vs IVUS Automated 0.84 Doh et al. 2014 CT vs IVUS Expert 0.53 Park et al. 2015 CT vs IVUS Expert 0.89 Non-expert radiologist 0.82 Automated 0.80 DeepLumen (Ours) CT vs OCT Automated 0.91 (95% CI: 0.89-0.94)

slide-52
SLIDE 52
  • DeepLumen is a core component of the HeartFlow
  • Analysis. It segments vessels faster and more accurate

than state-of-the-art 3D CNN architectures.

  • It is still important that human experts provide quality

control by ensuring that the segmentations are accurate.

  • Serious clinical problems can be addressed by combining

image analysis, deep learning and simulation/modelling.

  • All of these factors are essential to find the best treatment

for a patient with symptoms of coronary artery disease.

52

Conclusion

slide-53
SLIDE 53

53

Thank you

slide-54
SLIDE 54
  • Dilated convolution
  • No pooling
  • ReLu activations
  • Batch normalization

54

Design Choices

slide-55
SLIDE 55

55

Rotational Symmetry

Predict distance at angle a a

  • a

=

Predict distance at 0º on image rotated by angle -a.