WRONG, WRONG, WRONG. The proper use of brain images is to objectively use them to select the proper stroke rehab protocols for recovery. Anyone who does prediction crap like this should be fired. Going down this route you are giving up on finding better rehab solutions because the prediction tool says you will only recover so much. The fucking tyranny of low expectations and even lower responsibility hits again.
Predicting stroke outcomes using brain images
by Dr Tina Kaffenberger
Predicting outcomes early after stroke is important for guiding therapeutic decisions. (And therapeutic decisions right now lead to only 10% chance of full recovery. That is a massive failure rate.) Multiple predictive models have been proposed but none of them can be broadly adopted into clinical practice. Computed tomography (CT) is the most common imaging used in acute stroke. However, no comprehensive predictive model based on CT scans exists – a gap, which we aim to close.
Stroke location has been shown to be one essential biomarker to predict outcome but research tools to define stroke location in CT scans precisely and automatically are missing. By creating the first neuroanatomical whole brain atlas based on a stroke population specific CT template we aim to provide a tool to create such a predictive model.
Using normal-for-age CT scans from adults we derived a standard-resolution CT template, which is age and sex matched to a general stroke population. Specifically, this accounts for age-related structural changes which can interfere with matching to atlases derived from young healthy individuals.
We then manually defined anatomical brain regions on this template to generate a whole brain atlas, which includes regions known to be associated with functional outcome. Further, we developed an algorithm to match individual images to the template space, which is a prerequisite to performing analysis on a group level.
We validated this algorithm using CT scans of 100 individuals with small/medium/large ischemic stroke lesions. The alignment between manually segmented and automatically co-matched segmentations of five cortical and subcortical brain regions was reliable. This newly created whole brain atlas has the potential to standardise emerging predictive research. Together with the automated algorithm it allows analysis of (existing) large datasets to improve prediction tools for stroke patients.
Based on this atlas we will analyse clinically acquired acute CT (and magnetic resonance imaging) data from 800 patients with ischemic stroke from “A Very Early Rehabilitation Trial”. A model will be used to find the brain areas which are most important for stroke outcome.
The comprehensive predictive tool will be:
Our aim is for it to serve as practical guidance for clinicians and therapists.
Predicting outcomes early after stroke is important for guiding therapeutic decisions. (And therapeutic decisions right now lead to only 10% chance of full recovery. That is a massive failure rate.) Multiple predictive models have been proposed but none of them can be broadly adopted into clinical practice. Computed tomography (CT) is the most common imaging used in acute stroke. However, no comprehensive predictive model based on CT scans exists – a gap, which we aim to close.
Stroke location has been shown to be one essential biomarker to predict outcome but research tools to define stroke location in CT scans precisely and automatically are missing. By creating the first neuroanatomical whole brain atlas based on a stroke population specific CT template we aim to provide a tool to create such a predictive model.
Using normal-for-age CT scans from adults we derived a standard-resolution CT template, which is age and sex matched to a general stroke population. Specifically, this accounts for age-related structural changes which can interfere with matching to atlases derived from young healthy individuals.
We then manually defined anatomical brain regions on this template to generate a whole brain atlas, which includes regions known to be associated with functional outcome. Further, we developed an algorithm to match individual images to the template space, which is a prerequisite to performing analysis on a group level.
We validated this algorithm using CT scans of 100 individuals with small/medium/large ischemic stroke lesions. The alignment between manually segmented and automatically co-matched segmentations of five cortical and subcortical brain regions was reliable. This newly created whole brain atlas has the potential to standardise emerging predictive research. Together with the automated algorithm it allows analysis of (existing) large datasets to improve prediction tools for stroke patients.
Based on this atlas we will analyse clinically acquired acute CT (and magnetic resonance imaging) data from 800 patients with ischemic stroke from “A Very Early Rehabilitation Trial”. A model will be used to find the brain areas which are most important for stroke outcome.
Ischemic lesion – masked in red
Ischemic lesion (segmented into different anatomical areas according to newly developed atlas)
- applicable early after stroke
- relevant for most stroke patients and
- valid for multiple outcomes
Our aim is for it to serve as practical guidance for clinicians and therapists.
No comments:
Post a Comment