Fig. 4

We created an easy-to-modify script that attempts to predict NIH Stroke Scale (NIHSS) scores based on participant age and lesion load to each brain region described in the vascular territory brain atlas created by Faria and colleagues³³. Our script deep_learn.py, which is contained in our open-source GitHub repository: https://github.com/neurolabusc/StrokeOutcomeOptimizationProjectDemo), can be run in a Python environment or using Jupyter Notebooks, to predict NIHSS scores using two different algorithms: support vector regression (SVR - red) and neural network (NN - green). This GitHub page contains more detailed instructions on dependencies and how to run this script. Comparison of the performance of these algorithms shows that NN outperforms SVR for this classification task. Other researchers can easily modify this script to run it on subsets of our data (e.g. males vs. females, large vs. small lesions determined by a median split, etc) or compare the performance of other types of machine learning or AI models. *Each circle represents a unique participant. Lesion sizes were converted to z-scores and are represented by the size of each dot. Data points with predicted NIHSS Values > = 30 (N = 2) or < = 0 (N = 8) were excluded from the graph for visualization.