Abstract
Predicting functional outcome after ischaemic stroke informs treatment intensity, rehabilitation planning and family counselling. Clinical scores such as PLAN and iScore stratify risk additively and cannot capture nonlinear interactions. We trained four machine learning models (logistic regression, random forest, SVM, gradient boosting) on the International Stroke Trial (19,435 patients, 17 features) to predict 6-month outcome. Logistic regression achieved the highest AUROC of 0.786 [0.769, 0.803], outperforming reimplementations of the PLAN score (0.717, DeLong p < 0.001) and iScore (0.756, DeLong p < 0.001). These are approximations rather than the published scores: IST lacks several of their constituent variables, most importantly baseline NIHSS, so the reported margins are upper bounds on the true advantage over the validated instruments. All four models achieved comparable discrimination (0.778–0.786) with overlapping intervals, indicating a ceiling set by the available features rather than model complexity; five feature pairs exceed |r| = 0.7, supporting that reading. Grouped permutation importance, which permutes one-hot columns as a block, identified deficit severity, age, conscious level and stroke subtype as the strongest predictors, consistent with clinical knowledge. GBM was best calibrated (Brier 0.180, ECE 0.014 [0.013, 0.035]), a gap over logistic regression that bootstrapping confirms is not sampling variation. Standard tabular models on readily available clinical variables can therefore improve outcome prediction over these score approximations, with application to decision support in resource-constrained settings.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/SWITCH_003.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to Open Review
Open Review Page: https://openreview.net/forum?id=kC4wxd5a5z
BibTex
@InProceedings{KruJai_Machine_MICCAISAT2026,
author = { Kruger, Jaime Leigh},
title = { { Machine Learning for Stroke Outcome Prediction: Comparing Tabular Models to Clinical Scores on IST } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17255},
month = {pending},
page = {pending}
}