Robust Feature Selection for Classifying Early Ischemic Changes in Posterior Stroke

Rist L, Ditt H, Sühling M, Honus V, Schramme P, Maier A, Taubmann O (2025)


Publication Type: Conference contribution

Publication year: 2025

Journal

Publisher: Springer

Series: Lecture Notes in Computer Science

City/Town: Cham

Pages Range: 61-70

Conference Proceedings Title: Image Analysis in Stroke Diagnosis and Interventions

Event location: Marrakesh MA

ISBN: 9783031811005

DOI: 10.1007/978-3-031-81101-2_7

Abstract

Fast assessment in stroke diagnosis is essential to improve the treatment outcome. Scoring systems such as the ASPECT score facilitate the triage of patients according to stroke severity. For occlusions in the posterior circulation, pcASPECTS evaluates whether certain posterior cerebral regions show early signs of stroke and can be applied using the admission NCCT scan. This work investigates the automatic classification of early stroke changes in the two largest posterior regions based on NCCT images. The main focus lies on the implementation of robust measures that can counteract noise and scanner variances since they are harmful to established Radiomics pipelines. For 170 respectively derived regions from 85 patients, the described pipeline can reach up to 83.84% AUC with 79.40% sensitivity for the cerebellum and 73.14% AUC with 57.97% sensitivity for the occipital regions. A simple in-patient normalization scheme proved to be the most effective measure by improving the AUC by +8.17% and the sensitivity by +16.80%. Additional robustness techniques such as noise augmentation or discarding unstable and correlated features using the post-treatment scan resulted in only slight deviations from the best result, making them valuable tools for improving robustness when using Radiomics for posterior stroke classification.

Authors with CRIS profile

Involved external institutions

How to cite

APA:

Rist, L., Ditt, H., Sühling, M., Honus, V., Schramme, P., Maier, A., & Taubmann, O. (2025). Robust Feature Selection for Classifying Early Ischemic Changes in Posterior Stroke. In Image Analysis in Stroke Diagnosis and Interventions (pp. 61-70). Marrakesh, MA: Cham: Springer.

MLA:

Rist, Leonhard, et al. "Robust Feature Selection for Classifying Early Ischemic Changes in Posterior Stroke." Proceedings of the 4th International Workshop, SWITCH 2024, and 6th International Challenge, ISLES 2024, Marrakesh Cham: Springer, 2025. 61-70.

BibTeX: Download