What question did this study set out to answer?

March 26, 2026Procedia Computer Science0 citationsOpen Access

A Dual Machine Learning Framework for Automated ASPECTS Scoring and Treatment Eligibility in Acute Ischemic Stroke

RARayhane Ben Amor IAInes Ben AlayaTunis El Manar University WBWanis BarrehTunis El Manar University

Key Points

The aim is to develop a dual machine learning framework for automated ASPECTS scoring and treatment eligibility in acute ischemic stroke.
Developed two machine learning models for patient classification and ASPECTS region identification.
Trained and evaluated models on a dataset of 40 patients with annotated CT slices from the AISD dataset.
Measured model performance using accuracy in classification and detection of affected regions.
Achieved 93.33% accuracy in classifying patients for revascularization treatments.
Attained 93% accuracy in detecting affected and unaffected regions of interest.
Identified limitations due to small sample size and lack of external validation.

Abstract

The Alberta Stroke Program Early CT Score (ASPECTS) is a standardized method to assess early ischemic changes in the middle cerebral artery (MCA) territory on Non-contrast computed tomography (NCCT) scans. It plays a critical role in acute stroke management by guiding clinicians identify candidates for intravenous thrombolysis or mechanical thrombectomy based on the extent of ischemic damage. However, manual ASPECTS evaluation suffers from inter-observer variability and depends heavily on clinician experience. This study proposes a machine learning-based approach using two complementary models : the first classifies patients according to their eligibility for revascularization treatments, while the second identifies affected and unaffected ASPECTS regions and automatically estimates the ASPECTS score. Both models were trained and evaluated on a dataset of 40 patients, with annotated CT slices from the publicly available AISD dataset (Acute Ischemic Stroke Dataset, N=397). The approach achieved promising performance, with 93,33% accuracy in patient classification, and 93% accuracy for detecting affected and unaffected regions of interest (ROIs). However, the small sample size (40 patients) and lack of external validation limit the generalizability of the findings. Future work will focus on automating region of interest segmentation using deep learning and improving the models’ generalizability across diverse datasets and clinical environments.

AI에게 질문

Bookmark

View Full Paper