FPGA Based MRI Brain Tumor Segmentation Using Modified FCM Method

Detecting brain tumors with Magnetic resonance imaging (MRI) is vital for early diagnosis, giving patients a better chance for effective treatment by helping doctors pinpoint and understand the tumor. MRI technology is crucial for identifying tumor growth, enabling patients to receive timely treatment whether the growth is malignant or not. This work proposes an optimized FPGA version of a modified Fuzzy C-Means (FCM) algorithm with three clusters segmenting the brain tumors based on their pixel intensity deviation. The modification involves moving the cluster centers and varying the cluster weights with the mean intensity of the cluster of all the pixels, improving the accuracy. Block RAM of the FPGA device is utilized to store the input, intermediate, and output data to improve timing performance. A pipeline architecture is adopted for all units, including membership calculation, division unit, and centroid updating unit. Pipelining enables these units to work in parallel and thus reduce delays and allow a constant stream of data, which enhances the FCM design performance. The hardware implementation and verification are carried out on the Xilinx XC3S500E-4FG320 FPGA and Zynq UltraScale + ZCU104 platforms, while the Application-Specific Integrated Circuit (ASIC) implementation is performed using 45 nm technology library. The modified FCM technique outperforms conventional methods with a 108.271 MHz operating frequency, 0.859W power consumption, and hardware utilization of 8471 LUTs, 70 DSPs, and 3,897 FFs for 256 × 256 design. ASIC chip occupied 3 × 3 mm of core area and operates at a frequency of 253 MHz with power consumption of 161 mW.