Hybrid Ensemble Learning Framework for Multi-Class Brain Hemorrhage Classification

View at publisher Summarise with VIRAT
Authors: Gayathri Devi S, Monika J, Monishaa S, Rishibalan B
Cited by: 0
References: 15

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Early identification and subtyping of intracranial hemorrhage (ICH) subtypes from CT scans is important for timely medical treatment and better patient outcome. This paper introduces a new hybrid ensemble learning approach that inte grates deep convolutional neural network feature extraction with machine learning meta-learners for multi-class brain hemorrhage subtyping. By rigorous assessment of various CNN architectures such as custom CNN (53 %), DenseNet (47 %), and EfficientNet (63 %), we have selected best feature extractors for our ensemble system. Our system combines EfficientNetB3 and a custom SEResNeXt architecture to extract hierarchical features followed by ensemble classification employing Deep Neural Networks with Monte Carlo dropout, Support Vector Machines, and XGBoost. A logistic regression meta-learner combines predictions optimally by stacking methodology. The model includes uncertainty quantification through MC-dropout to improve clinical reliability. Tested on a large dataset of 47,014 CT images over six classes (EDH, IPH, IVH, SAH, SDH, and Healthy), our model attains 91.7 % accuracy with strong performance across all subtypes of hemorrhage. The hybrid architecture outperforms single models and single CNN methods while delivering interpretable uncertainty estimates critical for clinical decision-making.

Fields of Study & Subjects

Brain Tumor Detection and Classification Intracerebral and Subarachnoid Hemorrhage Research Advanced Neural Network Applications

Publication details

Journal / source
International Conference on Communication, Computing and Electronics Systems (ICCCES)
Pages
1036–1041
Publication date
21 January 2026
Language
EN