Secure and Intelligent Blockchain-Based Federated Directed Acyclic Graph Neural Network for Privacy-Preserving Brain Tumor Detection

Abstract

The convergence of healthcare has accelerated the adoption of federated learning for the group analysis of sensitive health information. On the other hand, deep learning has significantly enhanced medical image processing, enabling more accurate diagnoses and offering better treatment options in the era of the Internet of Things (IoT). Still, maintaining both high diagnostic quality and patient data privacy is a key problem because traditional techniques usually compromise either security or performance. To address these challenges, the current research proposes a Blockchain-Based Directed Acyclic Graph Neural Network with a Meerkat Optimization Algorithm (DAGNNet-MAO) for brain tumor detection. A set of high-resolution brain MRI images from the Brain Tumor Dataset is initially preprocessed using Robust Double-Weighted Guided Image Filtering (RDWGIF) to remove noise and improve image quality. The One Single Transformer (OST) is then used to extract features and enable local and global representations. The features are classified using DAGNNet with Meerkat Optimization Algorithm (MAO)-optimized model parameters, and prediction is secured through Blockchain-based Privacy-Preserving Federated Learning (BPPFL). The proposed model brings out impressive performance with 99.8% accuracy, 99.0% F1-score, 99.1% precision, 99.3% recall, 99.1% specificity, a low error rate of 0.8%, high throughput of 59.4 Mbps, low latency of 8.6 ms, and security level of 99%, showing that it is highly effective for fast, accurate, and privacy-protecting brain tumor detection in intelligent healthcare systems.