Overview

CT scans are often the preferred method among neuroradiologists for quickly diagnosing ischemic stroke. Despite their lower image resolution, CT scans have significant advantages over MRI scans that cannot be overlooked. For instance, a CT scan is much faster than an MRI scan, a quality that is essential in the field of stroke detection, where time is a critical factor [1]. In addition, CT scans are less claustrophobic, do not require the patient to remain still for an extended period of time, and can be performed on patients with implanted medical devices like pacemakers [1]. Yet, despite the CT scan’s prevalence, it can lead to highly variable diagnoses among doctors due to its inherent subjectivity [2]. A neural network algorithm capable of detecting ischemic stroke in CT scans would provide added confidence to a radiologist making a diagnosis of ischemic stroke on a low-resolution CT. 

Objectives for algorithm implementation:

  1. Binary classification of healthy vs. unhealthy tissue
  2. Automated image processing and feature extraction via windowing of CT scans
  3. Produce a visual representation of classification results

Design Overview:

Overview of algorithm design process.

 

References:

[1] Hess, Christopher. “Exploring the Brain: Is CT or MRI Better for Brain Imaging?” UCSF Department of Radiology and Biomedical Imaging, 16 Nov. 2015. Web.  <https://radiology.ucsf.edu/blog/neuroradiology/exploring-the-brain-is-ct-or-mri-better-for-brain-imaging>.

[2] Hampton-Till, James, Michael Harrison, Anna Luisa Kuhn, et. al. “Automated Quantification of Stroke Damage on Brain CT Scans: e-ASPECTS”. European Medical Journal Neurology 3.1 (2015): 68-74. 6 Aug. 2015. Web. <http://emjreviews.com/wp-content/uploads/Automated-Quantification-of-Stroke-Damage-on-Brain-Computed-Tomography-S….pdf>.