LV statistical shape modelling challenge: myocardial infarction

Organisers: Pau Medrano-Gracia, XingYu Zhang, Avan Suinesiaputra and Alistair Young

Contact and application*: ssm2015@auckland.ac.nz

 
Statistical shape modelling has been shown to be a powerful tool for visualising geometric and functional patterns of variation not only in the heart but in all organs. Biologically the heart presents great anatomical and functional variation making the encoding of these differences an interesting challenge in itself. After a myocardial infarction, the heart remodels to maintain physiological constraints. We hypothesise that a probabilistic model of the LV can predict a patient’s disease status.
A challenge to model the statistical shape of the left ventricle (LV) is therefore proposed this year aiming to create a probabilistic model and detect myocardial infarction.
 
The training dataset will comprise one hundred (100) cases with myocardial infarction and an additional one hundred (100) asymptomatic cases from the DETERMINE and MESA datasets respectively, contributed to the Cardiac Atlas Project (www.cardiacatlas.org).
 
Shapes will be provided as corresponding Cartesian point sets in cardiac MRI magnet coordinates at end-diastole (ED) and end-systole (ES). Classification labels indicating disease (0 = normal, 1 = infarcted) will be provided for the training dataset. No images will be provided.
 
The goals are to:
  • Establish a statistical shape model from the set of 3D shapes
  • Classify cases between normal or abnormal (with myocardial infarct)
 
The participants’ methods will be tested in a different set of 200 cases, again containing 100 asymptomatic cases and 100 infarcted cases. Classification accuracy and related measures of agreement will be calculated (specificity, sensitivity, etc.).
It is expected that the probabilistic models can be easily visualised but there is no restriction on the type of decomposition used to partition the shape space (i.e. can be linear or non-linear). Both supervised and non-supervised classification methods can be submitted (if there are enough of both a comparison might be made).
A peer-reviewed paper summarising the findings of this challenge will be submitted to an appropriate journal.
 
 
Results Results

 Extract from the collation paper:

Statistical shape modeling of the left ventricle: myocardial infarct classification challenge
Pau Medrano-Gracia, Xingyu Zhang, Avan Suinesiaputra, Brett Cowan and Alistair A. Young

Classifications results from the validation dataset achieved a median 93% specificity, sensitivity and accuracy (Table 2). Accuracy ranged from 73 to 98%.

Table 2. Participants methods were tested in a separate validation set. All values are %.

 
Challenge paper title
First Author
Specificity
Sensitivity
Accuracy
A
Systo-diastolic LV shape analysis by Geometric Morphometrics and Parallel Transport highly discriminates myocardial infarction
Paolo Piras
93
97
95
B
Statistical Shape Modeling using Partial Least Squares: Application to Myocardial Infarction Assessment
Karim Lekadir
99
97
98
C
Classification of Myocardial Infarcted Patients by Combining Shape and Motion Features
Wenjia Bai
97
94
95.5
D
Detecting Myocardial Infarction using Medial Surfaces
Pierre Ablin
89
90
89.5
E
Left ventricle classification using Active Shape Model and Support Vector Machine
Nripesh Parajuli
97
93
95
F
Supervised Learning of Functional Maps for Infarction Classification
Anirban Mukhopadhyay
73
73
73
G
Joint Clustering and Component Analysis of Spatio-Temporal Shape Patterns in Myocardial Infarction
Catarina Pinto
94
86
90
H
Myocardial Infarction Detection from Left Ventricular Shapes using a Random Forest
Jack Allen
91
92
91.5
I
Combination of Polyaffine Transformations and Supervised Learning for the Automatic Diagnosis of LV Infarct
Marc-Michel Rohé
95
95
95
J
Automatic detection of cardiac remodeling using global and local clinical measures and random forest classification
Jan Ehrhardt
89
92
90.5
K
Automatic Detection of Myocardial Infarction Through a Global Shape Feature Based on Local Statistical Modeling
Mahdi Tabassian
89
97
93
 
Median
 
93
93
93

Application & Results Application & Results

To apply, you will need to agree to and sign the following data-usage agreement document:

Download agreement

 

 Please e-mail the completed form to ssm2015@auckland.ac.nz

 
*Application closes 1 week before the paper submission is due


Results template available here. Please download, fill out both sheets, replace name of the file with the first author's name (as in the data-usage agreement) and send to ssm2015@auckland.ac.nz no later than 22 July. Please note that you must be registered in order to participate.

Important Notes Important Notes

Upon application, you will receive a link to the dataset.

The TAR file contains a compressed Zip file with all the models and labels for 200 models. MESA models correspond to normal cases whereas DETERMINE models correspond to patients with myocardial infarction. The key (labels) is available in the CSV file.

 
Notes:
(1) Points are corresponding across patients but are in the original DICOM coordinates. Two phases are available, at end-systole (ES) and end-diastole (ED). Points are stored in the 'vertices' files in (x,y,z) format.
 
(2) Details on the procedure to fit these finite-element models can be found here: http://pubs.rsna.org/doi/abs/10.1148/radiology.216.2.r00au14597
 
(3) Because of the sampling of these finite-element models, some points are repeated (especially in the apical area). Depending on your processing pipeline, you might need to filter these out first.
 
(4) For more information on MESA and DETERMINE: http://www.cardiacatlas.org/web/guest/data-access
 
(5) A triangular mesh file is included specifying the connectivity of the vertices to create a surface. This is the same for both the endocardial and epicardial surfaces since the same finite-element model was used. This may be useful for visualisation.
 
(6) We have applied the bias correction to correct for image-protocol shape bias between MESA and DETERMINE. A full reference can be found here:

Medrano-Gracia, Pau, et al. "Atlas-based analysis of cardiac shape and function: correction of regional shape bias due to imaging protocol for population studies." J Cardiovasc Magn Reson 15 (2013): 80.

Therefore all models are free from shape bias due to imaging protocol and can be directly compared.

(7) The following indices correspond to points approximately in the middle of the septum, pointing toward the centre of the right ventricle (same for both surfaces):

226 227 228 229 230 231 232 233 234 498 499

500 501 502 503 504 505 762 763 764 765 766

767 768 769

Sponsors Sponsors
Silver Sponsors

Elite Sponsors

Important dates Important dates

21/06/2015  -  Abstract and registration

Please register the title/abstract in the OCS submission system
 

21/06/2015 03/07/2015 - Paper submission EXTENDED

Please submit your paper describing the methodology and preliminary results (if available). See page "Paper Submission" for details.

15/07/2015 22/07/2015 - Paper acceptance notification

The program committee (PC) will decide whether the paper is accepted.

22/07/2015 - Challenge results due (challenge participants only)

If accepted, you may refine/(re)submit the results. By this date, we expect the final list of cases and their classification (see results template).

29/07/2015 04/08/2015 - Revised paper submission after review

If accepted, there is a chance to incorporate feedback from reviewers.

09/10/2015 - STACOM 2015 workshop
 
Participants Participants

An updated list of participants can be found here: