TY - JOUR

T1 - Precision Denoising in Medical Imaging via Generative Adversarial Network-Aided Low-Noise Discriminator Technique

AU - Alanazi, Turki M.

AU - Mercorelli, Paolo

N1 - Publisher Copyright: © 2024 by the authors.

PY - 2024/12

Y1 - 2024/12

N2 - Medical imaging is significant for accurate diagnosis, and here, noise often degrades image quality, thus making it challenging to identify important information. Denoising is a component of traditional image pre-processing that helps prevent incorrect disease diagnosis. Mitigating the noise becomes difficult if there are differences in the low-level segment features. Therefore, a Generative Adversarial Network (GAN)-aided Low-Noise Discriminator (LND) is introduced to improve the denoising effectiveness in medical images with a balanced image resolution with noise mitigation. The LND function is a key that distinguishes between high- and low-noise areas based on segmented features, which are also achieved by tuning the peak signal-to-noise ratio (PSNR). Considering the training sequences, the LND-identified intervals lessen the sequences to improve the changes in pixel reconstruction. The generator function in this method is responsible for increasing the PSNR improvements over the different pixels cumulatively. The proposed method successfully improves the pixel reconstruction by 11.05% and PSNR by 9.75%, with 9.75% less reconstruction time and 13.11% less extraction error for the higher pixel distribution ratios than other contemporary methods.

AB - Medical imaging is significant for accurate diagnosis, and here, noise often degrades image quality, thus making it challenging to identify important information. Denoising is a component of traditional image pre-processing that helps prevent incorrect disease diagnosis. Mitigating the noise becomes difficult if there are differences in the low-level segment features. Therefore, a Generative Adversarial Network (GAN)-aided Low-Noise Discriminator (LND) is introduced to improve the denoising effectiveness in medical images with a balanced image resolution with noise mitigation. The LND function is a key that distinguishes between high- and low-noise areas based on segmented features, which are also achieved by tuning the peak signal-to-noise ratio (PSNR). Considering the training sequences, the LND-identified intervals lessen the sequences to improve the changes in pixel reconstruction. The generator function in this method is responsible for increasing the PSNR improvements over the different pixels cumulatively. The proposed method successfully improves the pixel reconstruction by 11.05% and PSNR by 9.75%, with 9.75% less reconstruction time and 13.11% less extraction error for the higher pixel distribution ratios than other contemporary methods.

KW - generative adversarial network

KW - image denoising

KW - machine learning

KW - medical diagnosis

KW - neural networks

KW - Engineering

KW - Mathematics

UR - http://www.scopus.com/inward/record.url?scp=85211932290&partnerID=8YFLogxK

U2 - 10.3390/math12233705

DO - 10.3390/math12233705

M3 - Journal articles

AN - SCOPUS:85211932290

VL - 12

JO - Mathematics

JF - Mathematics

SN - 2227-7390

IS - 23

M1 - 3705

ER -