Multi-Derivative Feature Extraction and Stack Ensemble Learning for Automated Intracranial Hemorrhage Detection in Head CT Scans Images
Procedia Computer Science · 2026 · 283 · 5336-5346
DOI: 10.1016/j.procs.2026.06.584 ↗Kurz gefasst
Fachartikel in der Zeitschrift Procedia Computer Science, 2026. Verfasst von Chinmayee Pati und 2 weiteren Autor:innen. Die Arbeit nennt 16 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC-ND 4.0).
Abstract (Originalsprache)
Intracranial hemorrhage is a life-threatening medical condition that necessitates prompt and accurate diagnosis. Computed tomography (CT) is the imaging standard for hemorrhage detection, but manual interpretation is subject to variability and time constraints. Therefore, in this study, a comprehensive and data-driven machine learning framework was developed to enhance hemorrhage detection in head CT images. The proposed methodology integrates first-order derivatives (Sobel gradients in the x and y directions and Gradient Magnitude) and second-order derivatives (Laplacian) to capture both textural and structural cues essential for identifying hemorrhagic patterns. These advanced feature engineering methods proved pivotal in constructing a robust and interpretable feature set. The evaluation metrics, including accuracy, precision, recall, F1-score, SHAP-based interpretability, and feature importance, highlighted the substantial role of these derivatives in driving model performance, emphasizing their utility in capturing critical edge and texture information. Moreover, extensive cross-validation was performed to validate the robustness and generalizability of the proposed framework. The derivative-based feature set, combined with other engineered attributes, was used to train individual classification models such as Random Forest, Gradient Boosting, and LightGBM, along with a stacking-based ensemble model. The stacking classifier, comprising Random Forest and Gradient Boosting as base learners and LightGBM as the meta-learner, achieved the best performance with an accuracy of 90%, precision of 89%, recall of 88%, F1-score of 88%, and an AUC-ROC of 88%. These findings demonstrate the novelty and strength of combining multi-derivative feature extraction with stacking ensemble learning, offering an interpretable, efficient, and clinically valuable approach for automated intracranial hemorrhage detection.
Zitieren
APA
Pati, C., Nayak, S. R., & Jagadev, A. K. (2026). Multi-Derivative Feature Extraction and Stack Ensemble Learning for Automated Intracranial Hemorrhage Detection in Head CT Scans Images. Procedia Computer Science, 283, 5336-5346. https://doi.org/10.1016/j.procs.2026.06.584
Vancouver
Pati C, Nayak SR, Jagadev AK. Multi-Derivative Feature Extraction and Stack Ensemble Learning for Automated Intracranial Hemorrhage Detection in Head CT Scans Images. Procedia Computer Science. 2026;283:5336-5346. doi:10.1016/j.procs.2026.06.584
BibTeX
@article{Pati2026,
author = {Pati, Chinmayee and Nayak, Soumya Ranjan and Jagadev, Alok Kumar},
title = {Multi-Derivative Feature Extraction and Stack Ensemble Learning for Automated Intracranial Hemorrhage Detection in Head CT Scans Images},
journal = {Procedia Computer Science},
year = {2026},
volume = {283},
pages = {5336--5346},
doi = {10.1016/j.procs.2026.06.584}
}
EUVIMED