FPGA-Assisted MRI-based Brain Tumour Segmentation using VGG-16: A Transfer Learning Approach for Early-stage Cancer Detection

Brain cancer is a critical medical condition that demands early detection and accurate segmentation from Magnetic Resonance (MR) images. Many deep learning techniques have demonstrated remarkable potential in medical image analysis. However, deep learning frameworks have significant computational complexity, and the segmentation of MR images is time-consuming and demands superfast parallel processing controllers to perform segmentation tasks consecutively. Thus, this research aims to develop and implement a state-of-the-art deep learning framework using a customised Virtual Geometry Group (VGG-16) Convolutional neural network (CNN) architecture. Specifically, the BraTS 2019 and 2020 datasets, containing MRI (Tumour and non-tumour) brain images, have been employed to train a model that has been tested and deployed on the MyRIO (7 series FPGA compliant with A9 zinc Processor). Our proposed FPGA-assisted MR brain tumour segmentation accelerators outperform GPU and CPU implementations by a significant margin. This notable speed advantage ensures faster execution and processing times when segmenting brain tumours. The proposed framework not only attained significant accuracy but also reduced computational processing times. Hence, the proposed FPGA-assisted framework can be tailored for different real-time applications where time constraints are paramount, specifically in Intensive Care Units.