Task 2: Segmentation of Cerebral Microbleeds

On this page we introduce and explain task 2 in more detail. Please see the description pagedata page and the evaluation page for more information about this task.

Goal

The goal of this task is to develop an automated method to segment cerebral microbleeds in MRI scans. The training set contains segmentations of microbleeds in the full brain for every case.  The submitted automated methods will be applied on the hidden test set and should output microbleed segmentations  for the full brain. The predicted segmentation mask should be continuous and will be thresholded at 0.5 during evaluation to obtain a binary segmentation mask (0: background, 1: microbleed). 

Data

See the  data page.

Segmentations

For every case all microbleeds have been segmented. Below we show a 3D visualization of segmented microbleeds in one of the training cases:


An example of an axial T2* slice with microbleed segmentations is shown below:

Evaluation

Methods will be evaluated on segmentation, volume estimation, detection and count estimation. See the  evaluation page.

Additional information

  • An interesting public dataset that can be used to have additional training cases, is the microbleeds dataset that was released with the following paper: 
    Q. Dou*, H. Chen*, L. Yu, J. Qin, L. Shi, P. A. Heng, et al. "Automatic Detection of Cerebral Microbleeds from MR Images via 3D Convolutional Neural Networks." IEEE Transactions on Medical Imaging (TMI), 2016.
    This set contains 20 cases with dot annotations of microbleeds.