FPGA-Accelerated Hybrid Clustering Architecture for Real-Time Chest X-Ray Segmentation

Abstract

Chest X-ray segmentation is an important preprocessing step for computer-aided diagnosis, but software implementations of iterative clustering may introduce latency that limits real-time use. This paper presents an FPGAaccelerated hybrid clustering architecture that combines image enhancement, lung Region of Interest (ROI) extraction, KMeans initialization, Fuzzy C-Means refinement, and pipelined streaming. MATLAB is used to prepare normalized 8-bit pixel streams and to visualize the segmentation output, while synthesizable Verilog HDL implements the real-time processing path in Xilinx Vivado. ROI masking restricts clustering to lung fields, reducing unnecessary processing of background and nonlung anatomy. K-Means rapidly establishes initial cluster centers, after which FCM assigns fuzzy memberships to refine uncertain tissue boundaries. The FPGA pipeline integrates input buffering, preprocessing, ROI selection, clustering, postprocessing, measurement, and output stages. The project implementation uses 1,159 slice LUTs and 137 registers, reports positive setup and hold slack, and estimates 0.086 W total onchip power. Behavioral simulation confirms synchronized pixel flow, infection flag generation, and quantitative measurement outputs. MATLAB and FPGA result images demonstrate lungmask extraction, clustered segmentation, and final infection localization. The architecture therefore provides a compact and reconfigurable foundation for portable diagnostic devices, telemedicine, and real-time computer-aided chest radiography.