Abstract
Decisions regarding acute stroke treatment rely heavily on imaging, but interpretation can be difficult for physicians. Machine learning methods can assist clinicians by providing tissue outcome predictions for different treatment approaches based on acute multi-parametric imaging. To produce such clinically viable machine learning models, factors such as classifier choice, data normalization, and data balancing must be considered. This study gives comprehensive consideration to these factors by comparing the agreement of voxel-based tissue outcome predictions using acute imaging and clinical parameters with manual lesion segmentations derived from follow-up imaging. This study considers random decision forest, generalized linear model, and k-nearest-neighbor machine learning classifiers in conjunction with three data normalization approaches (non-normalized, relative to contralateral hemisphere, and relative to contralateral VOI), and two data balancing strategies (full dataset and stratified subsampling). These classifier settings were evaluated based on 90 MRI datasets from acute ischemic stroke patients. Distinction was made between patients recanalized using intraarterial and intravenous methods, as well as those without successful recanalization. For primary quantitative comparison, the Dice metric was computed for each voxel-based tissue outcome prediction and its corresponding follow-up lesion segmentation. It was found that the random forest classifier outperformed the generalized linear model and the k-nearest-neighbor classifier, that normalization did not improve the Dice score of the lesion outcome predictions, and that the models generated lesion outcome predictions with higher Dice scores when trained with balanced datasets. No significant difference was found between the treatment groups (intraarterial vs intravenous) regarding the Dice score of the tissue outcome predictions.
Highlights
Due to the overwhelming success of stent retrieval devices in recent randomized controlled trials[1,2,3,4,5], which was further confirmed in a pooled meta-analysis[6], guidelines for the treatment of stroke patients with proximal arterial occlusions recommend administration of intraarterial (IA) mechanical thrombectomy in conjunction with standard intravenous (IV) recombinant tissue plasminogen activator[7,8]
Inclusion criteria were as follows: (1) occlusion of either distal intracranial carotid artery or M1 segment of medial cerebral artery; (2) treatment consisting of exclusively intravenous thrombolysis therapy or in combination with intra-arterial treatment; (3) initial multi-parametric magnetic resonance imaging (MRI) including diffusion-weighted (DWI) and perfusion-weighted MRI (PWI) sequences acquired within 300 minutes of witnessed stroke symptom onset; (4) availability of follow-up MRI or computer tomographic imaging acquired 5–7 days after stroke onset; (5) absence of a preexisting proximal stenosis of the intracranial carotid artery assessed by ultrasonography, MRA, or angiographic imaging; (6) absence of an expansive and/or symptomatic intracranial hemorrhage
The results of the current study agree well with the ISLES stroke penumbra estimation (SPES) results, it should be noted that this ISLES benchmark study only evaluated machine learning classifiers for segmenting the penumbra based on multi-parametric MRI datasets, while no prediction of final infarction based on acute imaging was performed, which limits the comparison of the results
Summary
Due to the overwhelming success of stent retrieval devices in recent randomized controlled trials[1,2,3,4,5], which was further confirmed in a pooled meta-analysis[6], guidelines for the treatment of stroke patients with proximal arterial occlusions recommend administration of intraarterial (IA) mechanical thrombectomy in conjunction with standard intravenous (IV) recombinant tissue plasminogen activator (rtPA)[7,8]. Treatment-specific modelling of ischemic lesion evolution for the prediction of voxel-wise tissue fate based on acute imaging is likely to become increasingly relevant to both research and clinical practice in the upcoming years as new treatment options are adopted. This tissue outcome prediction problem was the subject of the ISLES 2017 challenge Isles-challenge.org/ISLES2017/), in which the top-performing models, convolutional and residual neural networks, employed deep learning techniques While these novel machine learning techniques constitute a potential improvement over traditional classifiers, there is not yet consensus regarding the optimal classifier and training setup for a true comparison of novel deep learning methods with conventional machine learning techniques[29,30]. These parameters need to be optimized for a true and fair comparison of novel deep learning approaches as well as new conventional machine learning techniques
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
