HO-ViT: An efficient and robust brain tumor MRI classification model with hybrid optimization
- 1Department of Computer Science and Engineering, Saveetha School of Engineering, Saveetha Institute of Medical and Technical Sciences, Saveetha University 1 , Chennai, Tamil Nadu,
- 2Department of Electrical and Communications Systems Engineering, Botswana International University of Science and Technology 2 , Palapye, Botswana
Abstract
Accurate classification of brain tumors using magnetic resonance imaging (MRI) is important in order to plan treatment and offer the patient a better chance of survival. However, the heterogeneity of the MRI data, suboptimal feature selection, and ineffective hyperparameter optimization tend to be constraints of conventional diagnostic methods and currently available deep learning models. To address these challenges, this paper proposes Hybrid Optimization-Vision Transformer (HO-ViT) as an effective and powerful brain tumor classification model that integrates HO methodology with the ViT framework, namely, using DeiT-Base (Data-efficient Image Transformer), which is a successful implementation of the Vision Transformer family, as a feature extractor. The HO-ViT model proposed is an effective synergistic integration of the powerful global exploration efficiency of the Grey Wolf Optimizer and the fast convergence speed of particle swarm optimization in finding the optimal feature subsets and tuning ViT hyperparameters. Before classification, high level preprocessing algorithms, such as image enhancement, image denoising, and precise tumor segmentation, are done to enhance boundary delineation of tumors and suppress the noise artifacts. The DeiT-Base is the main component of the feature extractor, which uses self-attention mechanisms to capture the long-range dependencies and the intricate spatial representations of brain MRI scans. Large-scale experiments with the benchmark dataset, Figshare Brain MRI, indicate that the proposed HO-ViT framework is better than traditional deep learning models and single-optimizer ones in terms of accuracy, sensitivity, specificity, and F1-score. Moreover, the hybrid optimization approach demonstrates faster convergence and also increases the classification strength. These findings affirm the HO-ViT to be an effective and computationally efficient tool in clinical decision support in real-time applications in neuro-oncology.
Related
- HayabusaNet: Hybrid attention-based multiscale fusion CNN for accurate and efficient brain tumor classification in MRI scans
- Efficient Alzheimer’s disease classification with robust rank pooling
- A hybrid deep feature extraction and ensemble learning framework for robust multi-class ECG image classification
- Optimization-enhanced machine failure classification using critical sensor features and hybrid learning models with advanced optimization techniques
- Multi-class Alzheimer’s disease classification using VGG19 transfer learning with hybrid Harris Hawks Optimization and random forest
BibTeX
@article{sethuraman2026hovit,
title = {{HO-ViT: An efficient and robust brain tumor MRI classification model with hybrid optimization}},
author = {Radhika Sethuraman and Gururama Senthilvel P and Padmakala S and Ravi Samikannu},
journal = {AIP Advances},
year = {2026},
month = jun,
doi = {10.1063/5.0334187},
url = {https://doi.org/10.1063/5.0334187},
note = {CC BY},
}
RIS
TY - JOUR
TI - HO-ViT: An efficient and robust brain tumor MRI classification model with hybrid optimization
AU - Radhika Sethuraman
AU - Gururama Senthilvel P
AU - Padmakala S
AU - Ravi Samikannu
PY - 2026
DA - 2026/06/01
JO - AIP Advances
DO - 10.1063/5.0334187
UR - https://doi.org/10.1063/5.0334187
AB - Accurate classification of brain tumors using magnetic resonance imaging (MRI) is important in order to plan treatment and offer the patient a better chance of survival. However, the heterogeneity of the MRI data, suboptimal feature selection, and ineffective hyperparameter optimization tend to be constraints of conventional diagnostic methods and currently available deep learning models. To address these challenges, this paper proposes Hybrid Optimization-Vision Transformer (HO-ViT) as an effective and powerful brain tumor classification model that integrates HO methodology with the ViT framework, namely, using DeiT-Base (Data-efficient Image Transformer), which is a successful implementation of the Vision Transformer family, as a feature extractor. The HO-ViT model proposed is an effective synergistic integration of the powerful global exploration efficiency of the Grey Wolf Optimizer and the fast convergence speed of particle swarm optimization in finding the optimal feature subsets and tuning ViT hyperparameters. Before classification, high level preprocessing algorithms, such as image enhancement, image denoising, and precise tumor segmentation, are done to enhance boundary delineation of tumors and suppress the noise artifacts. The DeiT-Base is the main component of the feature extractor, which uses self-attention mechanisms to capture the long-range dependencies and the intricate spatial representations of brain MRI scans. Large-scale experiments with the benchmark dataset, Figshare Brain MRI, indicate that the proposed HO-ViT framework is better than traditional deep learning models and single-optimizer ones in terms of accuracy, sensitivity, specificity, and F1-score. Moreover, the hybrid optimization approach demonstrates faster convergence and also increases the classification strength. These findings affirm the HO-ViT to be an effective and computationally efficient tool in clinical decision support in real-time applications in neuro-oncology.
SN - 2158-3226
ER -