SID-MAGS: SPUTUM IMAGE DENOISING VIA MULTI-SCALE DEEP ANALYSIS MODEL AND GAUSSIAN STAR FILTER
Main Article Content
Abstract
Tuberculosis (TB) is a contagious bacterial infection caused by Mycobacterium tuberculosis, primarily affecting the lungs but can spread to other organs. TB remains one of the top 10 leading causes of death globally. Microscopic TB images of Ziehl Neelsen-stained sputum smears are frequently degraded by impulse, or salt-and-pepper, noise, obscuring the visual structure of mycobacterium. Effective denoising of these images is therefore crucial for accurate diagnosis. This study introduces a novel Sputum image Denoising using Multi-scale Deep Analysis model and Gaussian star filter (SID-MAGS) denoising method, which effectively suppresses up to 99% of salt-and-pepper noise. The proposed SID-MAGS method integrates two filters: the Gaussian star filter (GSF) and the novel Multi-Scale Deep Analysis (MSDA) Model. The MSDA model enhances denoising by decomposing images into multiple scales and stages, enabling deep searching for noise-free pixels that correlate with the current noisy pixel. This decomposition into five different scales allows comprehensive noise removal, outperforming traditional techniques. Experimental evaluations using online and real-time clinical databases demonstrate that the proposed method reduces time consumption by 27.99% for 60% noise levels and achieves a Peak Signal-to-Noise Ratio (PSNR) of 29.61 dB for 90% noise, compared to 27.273 dB for the next best method. These results underscore the method's efficacy in preserving mycobacterial structures and contrast, making it a valuable preprocessing tool for automated TB diagnosis applications.