VLSI-Accelerated Chest X-Ray Image Segmentation using Adaptive Hybrid Clustering: MATLAB–FPGA Co-Design and Performance Evaluation

Chest X-ray segmentation is a fundamental preprocessing stage for computer-aided analysis of pulmonary disease, but software-only segmentation can become a bottleneck when low latency and embedded execution are required. This paper presents a MATLAB–VLSI co-design for chest X-ray image segmentation using the hybrid clustering procedure described in the supplied implementation. The processing chain reads a chest X-ray image in MATLAB, converts it to grayscale, forms a text-based numerical representation, and sends the pixel data to a hardware-oriented hybrid clustering core. The hybrid method performs automated cluster-count selection, cluster-to-cluster similarity analysis, internal pixel-difference analysis, cluster merging/reduction to two principal regions, maximum-pixel-based region finalization, and generation of an output text file that is reconstructed as a segmented image in MATLAB. The design is implemented and evaluated using Xilinx Vivado with a Zynq-7000 target. Three sample chest X-ray images are reported in the source evaluation. The baseline K-means implementation produces average accuracy, Dice coefficient, and Jaccard index values of 70.494%, 50.8043%, and 38.8676%, respectively, whereas the proposed method reports 99.9959%, 99.9921%, and 99.9842%. FPGA results show a reduction in LUT utilization from 36 to 29 and I/O utilization from 96 to 94. The proposed implementation reports 21.216 W dynamic power and 0.485 W static power, compared with 30.326 W and 4.528 W for the baseline. These results demonstrate the potential of hybrid clustering combined with hardware acceleration for low-latency medical-image segmentation, while also highlighting the need for broader dataset validation and standardized ground-truth evaluation.