Validating the Benefit of Combining Imaging and Clinical Data for Ischemic Stroke Outcome Prediction

Zeyad Abouyoussef, Johanna Ospel, Roberto Souza

Keywords: ischemic stroke, deep learning, outcome prediction, multimodal learning.

Abstract: Endovascular treatment (EVT) has been proven to be a successful treatment for some cases of acute ischemic stroke. However, neuro-radiologists rely on a small set of clinical features to select patients for treatment. This leads to the exclusion of patients who would have benefited from treatment and the inclusion of others who would not have benefited from it. Deep learning has been used to predict stroke outcome from baseline imaging and clinical data, with most studies reporting that combining imaging and clinical data slightly outperforms classical methods (e.g., logistic regression) trained on clinical data only. However, it is not clear how much of this improvement is attributed to the imaging data and whether it is robust to larger and more diverse test sets. We use one of the largest multi-center acute ischemic stroke datasets ($n = 1,105$) to determine whether combining imaging and clinical data outperforms classical methods. We show that combining imaging and clinical data matches the performance of logistic regression ($0.72$ Area Under Receiver Operating Characteristic Curve (AUROC)) when evaluated on a multi-center test set of over $600$ samples. We examine the models' predictions and weights and find that 1) both methods match each other's prediction for $78\\%$ of the samples, and 2) the weights associated with the imaging features are small compared to the clinical ones. This suggests that imaging features extracted from the deep learning model do not contribute to the prediction as much as the clinical ones.

Primary Subject Area: Integration of Imaging and Clinical Data

Secondary Subject Area: Interpretability and Explainable AI

Registration Requirement: Yes

Reproducibility: https://github.com/zeyad-kay/multimodal_mrs_prediction

Visa & Travel: Yes

Read CFP & Author Instructions: Yes

Originality Policy: Yes

Single-blind & Not Under Review Elsewhere: Yes

LLM Policy: Yes

Midl Latex Submission Checklist: Ensure no LaTeX errors during compilation., Replace NNN with your OpenReview submission ID., Includes \documentclass{midl}, \jmlryear{2026}, \jmlrworkshop, \jmlrvolume, \editors, and correct \bibliography command., Did not override options of the hyperref package., Did not use the times package., Use the correct spelling and format, avoid Unicode characters, and use LaTeX equivalents instead., Any math in the title and abstract must be enclosed within $...$., Did not override the bibliography style defined in midl.cls and did not use \begin{thebibliography} directly to insert references., Avoid using \scalebox; use \resizebox when needed., Included all necessary figures and removed *unused* files in the zip archive., Removed special formatting, visual annotations, and highlights used during rebuttal., All special characters in the paper and .bib file use LaTeX commands (e.g., \'e for é)., No separate supplementary PDF uploads., Acknowledgements, references, and appendix must start after the main content.

Submission Number: 340