A Fine-Tuned EfficientNet B1 Based Deep Transfer Learning Framework for Multiple Types of Brain Disorder Classification

Automated brain disorder classification for convenient treatment is one of the most complicated and widely spread problems. With the help of cutting-edge hardware, deep learning approaches are outperforming conventional brain disorder classification techniques in the medical image field. To solve this problem researchers have developed various transfer learning-based techniques. Pre-trained deep learning architectures are used here for feature extraction. This paper proposes a deep learning framework that includes a pre-trained fine-tuned EfficientNet B1 model to classify three different types of brain disorder and a normal category with 93%\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$93\%$$\end{document} of test accuracy. In order to evaluate the proposed framework, the dataset was trained and validated using additional deep learning models Inception V3 and ResNet50 V2 for feature extraction using softmax and support vector machine (SVM) classifiers and employing three primary optimizers: stochastic gradient descent (SGD), root mean squared propagation (RMSProp), and Adam. The EfficientNet B1 with softmax classifier and Adam optimizer outperformed the other two state-of-the-art models and achieved the best results.