Please enable JavaScript.
Coggle requires JavaScript to display documents.
Ischemic Stroke, Note: Aside from image, some features from other tests…
Ischemic Stroke
Highly complex disease
Use of MRI and CT Scan
Thresholding is used to detect damages
Based on fixed volumes
Suffers from several challenges
Thresholding is static
Does not account new available data
Not expressive enough to encompass details from PWI scans
Stroke is dynamic
Images are different from person to person
Cannot use thresholding as biomarker
Goal is to incorporate spatial features into a statistical model
Use of CNN architectures
Materials and Methods
Imaging protocols were in place in extracting images
T2-FLAIR
DWI
PWI MRI
Additional maps were acquired
Cerebral blood volume
Cerebral blood flow
Mean capillary transit time
Cerebral metabolism of oxygen
Relative transit Heterogeneity
Networks
CNN(Tmax)
Use of thresholding as biomarker
CNN(shallow)
CNN with fewer layers but with Spatial information
CNN(Deep)
More layers
Analysis
Use of statistical methods
Area under the curver
Independent of thresholding
Probability assessment of having higher risks
Contrast metrics
One minus the risks predicted from AUC
Used an dependent treatment variable
Compare use of rtPA treatment
Results and Discussion
CNN models achieved better heat maps than traditional models
CNN(Deep) most accurate
Can learn features without the need to hand-craft
More layers in a network resulted to better performance
Able to retain complex information
Note: Aside from image, some features from other tests may improve predictions
Nielsen, A., Hansen, M., Tietze, A., and
Mouridsen, K. Prediction of tissue outcome and
assessment of treatment effect in acute ischemic stroke
using deep learning. Stroke 49 (05 2018),
STROKEAHA.117.019740.