A secure hybrid deep learning framework for brain tumor detection and classification

Abstract

Abstract Objective The primary objective of this study is to propose a novel Brain-tumor Detection Network (BTDN) for MRI-based brain tumor diagnosis. The method aims to enhance image quality, ensure secure data transmission, and achieve highly accurate classification of brain tumors while addressing challenges related to manual interpretation and data security. Dataset The study utilizes three publicly available MRI datasets to evaluate the proposed method. The first dataset, D-I (Br35Hc), comprises brain MRI images specifically curated for tumor classification tasks. The second dataset, D-II (BraTS), is the widely recognized Brain Tumor Segmentation Dataset, frequently used for brain tumor detection and segmentation. Lastly, D-III (Kaggle Data Repository) is a collection of brain tumor MRI images obtained from the Kaggle platform, further diversifying the data sources for performance evaluation. Method The proposed methodology involves several key components to enhance the accuracy and security of brain tumor classification. During preprocessing, Contrast Limited Adaptive Histogram Equalization is employed to improve image quality by enhancing contrast, while data augmentation ensures the durability and robustness of the dataset during training. The core of the approach is the BTDN, specifically developed for precise classification of brain tumors. To address data security concerns, the study introduces Secure-Net (SN), which prevents data modification and enables safe retrieval using specific identifiers for trusted information exchange. The performance of BTDN is rigorously evaluated by comparing it against six widely used deep learning models: ResNet101, DenseNet169, VGG19, MobileNetV3, InceptionV3, AlexNet, and ConvNeXt. Results The BTDN demonstrated outstanding classification accuracies across the three datasets, achieving 99.68% on D-I (Br35Hc), 98.81% on D-II (BraTS), and 95.33% on D-III (Kaggle). These results surpass the performance of the compared deep learning models, underscoring the effectiveness of BTDN for accurate and reliable brain tumor detection and classification. Conclusion The study demonstrates that the proposed BTDN model ensures high accuracy in MRI-based brain tumor classification while incorporating SN to enhance data security and safe transmission. The remarkable performance across multiple datasets underlines BTDN’s potential as a reliable tool for precise and secure brain tumor diagnosis in clinical settings.

Keywords

• Brain tumor
• Magnetic resonance imaging
• BTDN
• CLAHE and DL