What does this research mean for the field?

An automated design space exploration framework using multi-objective evolutionary optimization for approximate Gaussian filtering reduces the power-area product of ultrasound image denoising by up to 90.1% while maintaining high reconstruction quality. Novelty: ClaimNovelty.METHODOLOGICAL. Consensus alignment: ConsensusAlignment.NEUTRAL.

What question did this study set out to answer?

The main goal is to create a framework for low-power and high-quality ultrasound image denoising.

June 3, 2026Scientific Reports0 citationsOpen Access

Multiobjective framework for hardware and quality aware approximate Gaussian filtering toward energy efficient ultrasound image denoising

SSSawaira Sana ASArslan Shaukat SKSajid Gul Khawaja

Key Points

The main goal is to create a framework for low-power and high-quality ultrasound image denoising.
Developed a design space exploration (DSE) framework integrating approximate computing with multi-objective optimization.
Evaluated four DSE schemes using various optimization algorithms including NSGA-II and NSGA-III.
Introduced the Quality-Area-Power Product (QUAP) metric for assessing hardware-efficient solutions.
NSGA-II achieved a 90.1% reduction in power-area product while maintaining PSNR > 35 dB and SSIM > 0.93.
In many-objective optimization, NSGA-III showed superior diversity and robustness under timing constraints.
The framework enables efficient synthesis of approximate Gaussian filtering accelerators for embedded systems.

Abstract

Real-time denoising of ultrasound images is essential for clinical and healthcare workflows, yet remains constrained by the computational and energy cost and latency of convolution-based filtering on resource- and energy-constrained embedded platforms. This paper presents a unified, automated design space exploration (DSE) framework for hardware- and energy-efficient ultrasound image denoising using approximate Gaussian filtering. The framework integrates approximate computing with multi-objective and many-objective evolutionary optimization to jointly balance image quality, power, area, and critical-path delay. A pipelined two-dimensional Gaussian filter architecture is employed, in which approximate adders and multipliers are jointly selected from the Evolutionary Approximation 8-bit (EvoApprox8b) arithmetic library, enabling circuit-level exploration of accuracy–efficiency trade-offs. Four DSE schemes are evaluated with increasing decision-space complexity and objective dimensionality. Nondominated Sorting Genetic Algorithm II (NSGA-II), Nondominated Sorting Genetic Algorithm III (NSGA-III), and Multi-Objective Particle Swarm Optimization (MOPSO) are systematically compared against random search. A task-driven composite metric, the Quality–Area–Power Product (QUAP), is introduced to support quality-aware ranking of hardware-efficient solutions. Experiments on representative ultrasound images show that NSGA-II achieves the most favorable trade-offs in three-objective optimization, with up to 90.1% reduction in power–area product while maintaining high reconstruction quality (Peak Signal-to-Noise Ratio (PSNR) > 35 dB, Structural Similarity Index Measure (SSIM) > 0.93). In many-objective settings with timing constraints, NSGA-III provides superior diversity and robustness. Overall, the proposed framework enables scalable synthesis of approximate Gaussian filtering accelerators for energy-efficient embedded imaging systems, supporting sustainable and accessible healthcare technologies.

Ask AI

Helpful

Bookmark

View Full Paper