An enhanced deep learning method for multi-class brain tumor classification using deep transfer learning

• Published in: Multimedia tools and applications Vol. 82; no. 20; pp. 31709 - 31736
• Main Authors: Asif, Sohaib, Zhao, Ming, Tang, Fengxiao, Zhu, Yusen
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.08.2023; Springer Nature B.V
• Subjects: Accuracy; Brain; Brain cancer; Classification; Computer Communication Networks; Computer Science; Computer vision; Data Structures and Information Theory; Datasets; Deep learning; Diagnostic systems; Glioma; Image classification; Magnetic resonance imaging; Medical imaging; Medical research; Multimedia Information Systems; Special Purpose and Application-Based Systems; Track 2: Medical Applications of Multimedia; Tumors; Xception; Deep learning; Multi-class brain tumor classification; Image processing; Transfer learning
• ISSN: 1380-7501; 1573-7721
• DOI: 10.1007/s11042-023-14828-w

Multi-class brain tumor classification is an important area of research in the field of medical imaging because of the different tumor characteristics. One such challenging problem is the multiclass classification of brain tumors using MR images. Since accuracy is critical in classification, computer vision researchers are introducing a number of techniques; However, achieving high accuracy remains challenging when classifying brain images. Early diagnosis of brain tumor types can activate timely treatment, thereby improving the patient’s chances of survival. In recent years, deep learning models have achieved promising results, especially in classifying brain tumors to help neurologists. This work proposes a deep transfer learning model that accelerates brain tumor detection using MR imaging. In this paper, five popular deep learning architectures are utilized to develop a system for diagnosing brain tumors. The architectures used is this paper are Xception, DenseNet201, DenseNet121, ResNet152V2, and InceptionResNetV2. The final layer of these architectures has been modified with our deep dense block and softmax layer as the output layer to improve the classification accuracy. This article presents two main experiments to assess the effectiveness of the proposed model. First, three-class results using images from patients with glioma, meningioma, and pituitary are discussed. Second, the results of four classes are discussed using images of glioma, meningioma, pituitary and healthy patients. The results show that the proposed model based on Xception architecture is the most suitable deep learning model for detecting brain tumors. It achieves a classification accuracy of 99.67% on the 3-class dataset and 95.87% on the 4-class dataset, which is better than the state-of-the-art methods. In conclusion, the proposed model can provide radiologists with an automated medical diagnostic system to make fast and accurate decisions.