Keywords:
Artificial Intelligence, Neuroradiology brain, Computer applications, CT, Computer Applications-General, Computer Applications-Detection, diagnosis, Image registration
Authors:
S. Tanamala1, S. Chilamkurthy2, R. Ghosh1, P. Rao1; 1Mumbai/IN, 2Mumbai, Ma/IN
DOI:
10.26044/ecr2019/C-2938
Methods and materials
The hemorrhage segmentation algorithm produces pixel mask for localization of intracranial hemorrhage.
To determine the corresponding affected cerebral/cerebellar regions,
the algorithm needs to be aware of anatomy of brain.
To this end,
we created anatomical atlases for five anonymized head CT scans.
Each of theese atlases were marked with marked with the following anatomical regions:
- Left/right frontal region
- Left/right temporal region
- Left/right parietal region
- Left/right occipital region
- Left/right cerebellum
- Brain stem
- Ventricles
We used multi-atlas segmentation to predict anatomy of the brain in a new scan as follows: we registered each of these five scans to the given scan (fixed image).
The registration process involved automatically searching for the transformation required so that intensities of transformed moving images matches approximately with that of fixed image.
The five atlases were then propagated back to the fixed image space and majority vote was used to predict the final anatomy of the given scan.
Finally,
We used kNearest Neighbour algorithm (kNN) to find the anatomical location of each pixel in hemorrhage pixel mask.
Percentage of lesion in each anatomical region is quantified.
If this percentage for an anatomical region is greater than 10%,
hemorrhage is considered to be affecting that region.
We have put together predicted anatomical locations and volumetric predictions to generate a line in the format of 'Intracranial hemorrhage of 10.5ml in left temporal region'.