Research Article
Improving Multi-class Brain Tumor Detection Using Vision Transformer as Feature Extractor
@INPROCEEDINGS{10.1007/978-3-031-35078-8_1, author={Adeel Ahmed Abbasi and Lal Hussain and Bilal Ahmed}, title={Improving Multi-class Brain Tumor Detection Using Vision Transformer as Feature Extractor}, proceedings={Intelligent Systems and Machine Learning. First EAI International Conference, ICISML 2022, Hyderabad, India, December 16-17, 2022, Proceedings, Part I}, proceedings_a={ICISML}, year={2023}, month={7}, keywords={deep learning Brain tumor types convolution neural network vision transformer}, doi={10.1007/978-3-031-35078-8_1} }- Adeel Ahmed Abbasi
Lal Hussain
Bilal Ahmed
Year: 2023
Improving Multi-class Brain Tumor Detection Using Vision Transformer as Feature Extractor
ICISML
Springer
DOI: 10.1007/978-3-031-35078-8_1
Abstract
Accurate brain tumor subtypes classification is significant for prognosis and treatment. The aim of this research is to improve the multiclass brain tumor classification using vision transformer as feature extractor. In this study, we first optimized and employed deep learning ResNet101 for feature extraction and fed to machine learning classifiers for multi-class classification. We then optimized and employed vision transformer and fed these features to machine learning decision classifier. We measured the performance with standard performance metrics. The Artificial Intelligence vision transformer with decision tree classifier yielded highest multi-class classification performance with 99.89% accuracy and 1.00 AUC to detect pituitary followed by 97.69% accuracy and AUC of 0.96 to detect meningioma. The results are compared with ResNet101 with transfer learning. ResNet101 deep features by utilizing KNN yielded detection of pituitary tumor (98.80% accuracy, 0.99 AUC), glioma (93.47% accuracy, 0.93 AUC). The results revealed that proposed approach with vision transformer and decision tree features extractor are more robust in detecting multiclass brain tumor prediction. The proposed approach can be better utilized for betterment of treatment and prognosis to obtain improved clinical outcomes.
