| CT Slice | Label | Prediction |
|---|---|---|
 |
 |
 |
 |
 |
 |
The goal of this challenge is to segment three areas of the heart. The left ventricle (yellow), myocardium (green) and right ventricle (blue) for the both end diastolic (ED) and end systolic phase (ES). The reason this is import is it allows the calculation of myocardial mass, myocardial thickness, volumes of the left and right ventricles and stroke volume at different phases. All this information is helpful in understanding the health of the heart and in diagnosing cardiomyopathy.
A full list of the metrics used can be found here. In brief they involve DICE, end diastolic and end systolic volumes for the left and right ventricles, As well as measurements involving volume and mass of the myocardium at both end diastolic and end systolic phases.
The data set includes 100 patients in the training, set which labels were provided for and 50 in the hold out test set with no labels, that needed to be submitted online to get results. The training and test consisted of 5 categories with each category making up 20% of each set.
- healthy patients
- patients with previous myocardial infarction
- patients with dilated cardiomyopathy
- patients with an hypertrophic cardiomyopathy
- patients with abnormal right ventricle
The data format was NIfITI format which is a format developed by Neuroimaging Informatics Technology Initiative and used for medical imaging. Each file contains image data, a series of slices of the MRI (the images at the top of the page are two of these slices from different NIfITI files), some associated metadata and a reference space.
Selected model 2d unet Although the 3d 2d unet ensemble showed the best results it would be computationally heavy thats why the 2d unet was decided on.