Abstract
Brain tumors are serious and potentially life-threatening conditions, making early detection important. When a tumor is suspected, accurately identifying the affected region in Magnetic Resonance Imaging (MRI) scans based on intensity, shape, and local features is essential. This work proposes a Field-Programmable Gate Array (FPGA) architecture that employs a fixed-point implementation of the pixel intensity deviation-based Fuzzy C-Means (FCM) clustering technique for tumor segmentation in MRI scans, using three successive MRI slices as input features to incorporate inter-slice contextual information while maintaining hardware efficiency. The FCM technique is modified by introducing new weights to adjust the cluster center to obtain improved segmentation performance. The primary goal of the proposed architecture is to minimize hardware resource utilization while maintaining segmentation performance. The intermediate membership storage Random Access Memory (RAM) is removed by implementing the temporary storage unit using registers, resulting in a significant reduction in BRAM utilization on FPGA devices. To further boost the system's real-time processing performance, a 7-stage pipelined membership calculation unit is added, which can improve the timing performance and processing throughput of the design. The proposed fixed-point FPGA architecture on Xilinx XC7S50-CSGA324 (Spartan-7) FPGA and high-performance Zynq UltraScale+MPSoC devices consumes reduced hardware resources and achieves a high operating frequency. The FPGA implementation realizes a maximum clock rate of 102.27 MHz and output power of 0.714 W for a 320 × 320-sized image. The Application-Specific Integrated Circuit (ASIC) implementation, synthesized using 45nm gsclib045 standard cell library, yields a compact die area of 4.3 mm × 4.3 mm including the on-chip RAM used for pixel storage, while operating at 218MHz with power consumption of just 18.96mW. To further demonstrate the scalability of the proposed architecture, synthesis was also performed using the Arizona State University Predictive (ASAP) 7 nm FinFET technology library, achieving a core area of 1.02 mm × 1.02 mm operating at 336 MHz with power consumption of 6.4 mW. The proposed modified fixed-point FCM algorithm achieves a segmentation accuracy of 97.8%, demonstrating its effectiveness for brain tumor segmentation while maintaining low hardware complexity.