Real and assumed insights statistical models and imaging biomarkers - - PowerPoint PPT Presentation

Digital Health Lab Day – Wädenswil – 03.10.2019

Sven Hirsch, Norman Juchler

Real and assumed insights – statistical models and imaging biomarkers for disease characterization of intracranial aneurysms

Primarily driven by clinical needs Primarily research-driven

Clinical data is created primarily to treat patients

2

Data acquisition Structured database Analysis / modelling Tools 80%

Working with clinical data is a challenge!

3

Challenges

§ Access procedures (ethical approval, anonymization) § Missing links between different sources § Data often lacks harmonization and documentation § Data tends to be inconsistent § Data skewed by selection bias

Solutions

§ Curation of consistent research databases § Robust pipelines for data processing § Transfer of clinical domain knowledge required § Always consider selection bias

Machine learning for disease characterization

5

Intracranial aneurysms are focal deformations of cerebral arteries

6

§ Typically located in the direct proximity of vessel bifurcations and the Circle of Willis § 3% of the population is affected 1 § Mostly free of symptoms – until rupture § Rupture risk ~1% per year 2 § Mortality 15%, invalidity 31% upon rupture

1 M. Vlak et al. "Prevalence of unruptured intracranial aneurysms”, 2011 2 A. Ahmed et al. “Aneurysms, intracranial”. Encyclopedia of the Neurological Sciences. 2014

Illustration of cerebral vasculature with an intracranial aneurysm. (Source: Sentera Healthcare)

X-ray Angiography

Source: AneuX, HUG-p163. Image size: 256x256px, Voxel size: ~0.25mm

From shape to prognosis…

8

AneuX AneurysmDatabase

§ 1350+ patient records § 900+ medical images § 350+ aneurysm geometries

aneurIST database …

A multicentric initiative to improve the treatment of intracranial aneurysms Morphology study Irregularity Rating study

Vision: Shape

as bio-marker for disease progression

Data: Medical

imaging data and clinical data

Evaluation of scoring schemes Irregularity Rating study Study on location dependency

Classical approach:

Quantitative morphology for disease status prediction

9

§ Size and shape are associated with pathologic wall conditions § Irregular shape or “ugliness” already used in clinics as subjective risk indicator § How to quantify irregularity/morphology? Rupture: Yes / No?

Size § Volume § Surface area § Aneurysm size § Height § Max. diameter § Neck diameter § …

Quantitative morphology for disease status prediction

10

Shape § Aspect ratio § Non-sphericity § Ellipticity § Undulation § Bottleneck factor § … Curvature Various curvature “energies” Writhe-number based descriptors Captures surface asymmetries and “geometric deformation” energies Zernike Moment Invariants Generic shape descriptor § ZMI cumulants § Designed for shape queries

Result: Aneurysm morphology carries significant information about the disease status

11

Prediction performance for exemplary descriptors

NSI Size Shape Curvature All

Prediction based

• n single indices

Prediction based on multiple indices Feature configurations: 1. Volume (V) 2. Surface (S) 3. Neck diameter (Dn) 4. Dome height (H) 5. Aneurysm size (ASZ) 6. Non-sphericity index (NSI) 7. Ellipticity index (EI) 8. Undulation index (UI) 9. Aspect ratio (AR) 10. Normalized mean curvature 11. Normalized Gaussian curvature

Size Shape

11. Normalized Gauss 12. Size: 13. Size + AR: 14. Size + AR: 15. Shape: 16. Shape: 17. Curvature: 18. Shape + size: 19. Size: 20. All:

Classifier: SVM, trained with 6-fold nested cross- validation and 50 repetitions, using 450 datasets

…but the initial expectations were not met

§ Moderate overall prediction accuracy (0.75 in the best case) § Morphometric predictors possibly not specific enough? § Labelling problem? (Rupture status vs. stability status) § Selection bias? § Missing clinical data for stratified analysis § How to integrate tools into clinical routine?

Result: Aneurysm morphology carries some information about the disease status

12

Improvements: § Curation of a complete, validated database § Realignment with clinical needs § Follow previously defined clinical hypotheses § Move away from rupture status prediction

We had to revise our engineering approach…

13 Morphology study Irregularity Rating study

Vision: Shape

as bio-marker for disease progression

Data: Medical

imaging data and clinical data

Evaluation of scoring schemes Irregularity Rating study Study on location dependency

Integration of clinical knowledge

14

Irregularity: A “crowd-sourced” metric for shape

15

§ What constitutes an “irregular” shape? § Which morphometrics represent perceived irregularity the best? § What is the relationship between irregularity and clinical factors?

§ Focus on subjective assessment

• f geometries

§ Aggregation using rank-based analysis

We collected qualitative rating data using our interactive rating tool

16

Regular Irregular 1 9

§ Critical characteristics:

§ Presence of blebs/lobules § Asymmetric aspect § Elongation / non-sphericity § Curvature and total writhe

§ No association was found with other clinical factors… § …except for aneurysm location § Metrics for aneurysm morphology shows strong dependency on location

Perceived irregularity is associated with rupture

17

Rupture

78 unruptured 41 ruptured 119 total

Sex

36 male 98 female 134 total

Smoking

40 non-smokers 74 smokers 114 total

Hypertension

70 non-hypertens. 49 hypertensive 119 total

IA size (aSz)

134 aneurysms Mean: 6.9mm Std: 3.4mm

Patient age

104 aneurysms Mean: 53.0y Std: 12.1y

Characteristic AUC p-val AUC p-val AUC p-val AUC p-val

Sp

p-val

Sp

p-val Rough surface 0.59 0.12 0.55 0.41 0.51 0.90 0.50 0.97 0.68 *** 0.03 0.76 Blebs 0.71 ** 0.54 0.53 0.56 0.32 0.52 0.70 0.50 *** 0.10 0.33 Lobules 0.79 *** 0.64 0.01 0.52 0.68 0.52 0.69 0.40 ** 0.02 0.83 Asymmetry 0.81 *** 0.59 0.10 0.51 0.85 0.53 0.62 0.49 *** 0.09 0.39 Complex vasc. 0.51 0.85 0.56 0.28 0.52 0.76 0.57 0.17 0.13 0.28 0.03 0.76 Irregularity 0.81 *** 0.58 0.14 0.50 0.90 0.51 0.79 0.71 *** 0.08 0.42 Aneurysm size 0.71 ** 0.62 0.03 0.57 0.18 0.52 0.24

• 0.14

0.16 Non-sphericity 0.83 *** 0.60 0.07 0.55 0.38 0.50 0.97 0.61 *** 0.01 0.93 Curvature 0.78 *** 0.58 0.15 0.55 0.34 0.52 0.78 0.81 *** 0.09 0.38 Patient age 0.58 0.21 0.59 0.18 0.55 0.48 0.74 * 0.14 0.16

• 0.0

0.2 0.4 0.6 0.8 1.0 FPR (false positive rate) 0.0 0.2 0.4 0.6 0.8 1.0 TPR (true positive rate) irregularity (AUC=0.81) asymmetry (AUC=0.81) lobules (AUC=0.79) blebs (AUC=0.71) rough surface (AUC=0.59) complex vasc. (AUC=0.51) NSI (AUC=0.83) GLN (AUC=0.78) aSz (AUC=0.71) irregularity + loc. (AUC=0.87) unruStured ruStured ruSture6tatus 0.0 0.2 0.4 0.6 0.8 1.0 irregularity ICA oSh 0CA 01 PCoPA ACoPA

Irregularity is an independent risk factor, along with location (and aneurysm size)

The power of bigger data

20

Data infrastructure and harmonization

21 aneurIST 2006 - 2010 ISGC 2014 - … AneuX 2006 - …

An ongoing effort towards a multi- centric, research-

• riented data

collection

Tool for case matching and statistical analysis

22

§ Select cases according to predefined criteria § Useful to compare subcohorts of patients and to “skim” the data for statistical relationships § Intuitive representation of simplifies the interpretation of results

Example: Mosaic plots to compare multiple and solitary aneurysms

23 § AComA most common location in solitary IA § MCA most common location in multiple IAs § Female have more multiple IAs § Smokers more likely having multi IA § Multiple IAs rupture less

Independent variable (predictor) Dependent variable (response)

§ Useful to compare two categorical variables § Width of cols and rows indicates relative proportions § Color encodes the outcome of statistical tests:

§ Blue: significantly over-represented § Red: significantly under-represented

AComA MCA Pcom

Solitary IA Multiple IAs V-B

Location of ruptured or larger IA

Application: Assess clinical tools (Example: PHASES score)

24

§ Always consider biases! § Visualization, statistical tests and perseverance are your best friends § The AneuX AneurysmDataBase is one of the largest databases of its kind § Beware of clinical scores based on univariate analysis § Better tools are required to accommodate the multifactorial nature of the disease (e.g. Bayesian network reasoning)

Conclusions / Experiences

25

Acknowledgments

26

Hirslanden Klinik, Zurich § Daniel Rüfenacht § Isabel Wanke § Stephan Wetzel Institute of Applied Simulation, ZHAW § Norman Juchler § Sabine Schilling § Erich Zbinden University Hospitals Geneva § Philippe Bijlenga § Sandrine Morel § Nicolas Roduit § Nicolas Dupuy § Rafik Ouared

Vital-IT

Vital-IT, Lausanne § Jérôme Dauvillier § Robin Liechti § Olivier Martin