Low-Power VLSI Implementation of Gradient Boosting Model for Tumar Detection of MRI Brain Image

Abstract

This work proposes a low-power VLSI implementation of a Gradient Boosting (GB) ensemble model for brain tumor detection from MRI images. The approach maps a trained GB classifier into an energy-efficient hardware accelerator targeting ASIC/FPGA platforms, combining model compression, fixed-point quantization, pruning of weak trees/branches, and memory-aware scheduling to minimize dynamic power and memory footprint. Preprocessing and feature-extraction are performed using lightweight image processing modules implemented on-chip; the resulting features feed the GB inference engine realized as a tree traversal pipeline with SIMD-like parallelism and branch-prediction minimization. Design choices—such as using ternary/fixed-point arithmetic, on-chip SRAM buffering, and early-exit in tree ensembles—reduce computation and communications, improving throughput and energy per inference. In this paper, tumor detection using support vector machine (SVM), decision tree (DT) and gradient boosting (GB) machine learning (ML) technique are presented. The GB ML technique is providing good accuracy compared to other DT and SVM technique. In this model is simulated python language and calculated simulation parameter i.e. precision, recall, accuracy and F1-score.