This retrospective observational study from a University hospital in western India included all patients with posterior circulation AIS from October 2022 to May 2025. Each data set consisted of input labels: clinical features; stroke volume (3-D Slicer on MRI DWI images) and outcomes (output labels) as modified Rankin Score (mRS) at 3 months.

A total of 170 patients were studied; 52 males (30.6%) with mean age of 59.54+/-14.7 years. Median NIHSS was 4 (IQR 2-5); Median POST-NIHSS was 7 (IQR 4-12.25) and Mean stroke volume was 15.36 cc +/-28.8 cc. MRS was found to correlate with the NIHSS (r=0.66, p value <0.0001). MRS was not found to correlate with Stroke volume (Pearson’s correlation 0.248, p value 0.0011).   

Among the four ensemble boosting models evaluated (Random Forest, XG Boost, Light GBM and Cat Boost) for predicting 90-day functional outcomes in stroke patients; Random Forest achieved the highest accuracy (89%) and precision (88%), while Light GBM demonstrated the best balance with the highest F1-score (80%) and recall (78%). The choice of model for clinical implementation should be guided by the specific clinical priority: Random Forest when maximizing precision is critical to minimizing false positives, and Light GBM when a balanced measure of precision and recall is desired.

AI based stroke prognostication models can help physicians in counseling patients and relatives about outcomes in posterior circulation stroke. We can further test the accuracy of these AI models against traditional stroke prognostication scales like SPAN-100.