20–21 Nov 2026
Online Conference
Asia/Karachi timezone

Artificial Intelligence in Medical Physics: Deep Learning-Based Denoising for Low-Dose CT Image Enhancement and Anatomical Detail Preservation

21 Nov 2026, 13:30
5m
Online Conference

Online Conference

Online Conference

Speaker

Kokab Fatima (Ned university of Engineering and Technology)

Discussion (methods + interpretation)

A preliminary computational experiment was conducted using a mathematical Shepp-Logan phantom as a reference image. Controlled Gaussian noise was introduced at three levels (0.03, 0.08 and 0.15), followed by Gaussian, median and bilateral filtering. Image quality was evaluated using Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index Measure (SSIM). The results were used to examine the effectiveness of conventional denoising methods under different noise conditions. As a further stage, a lightweight Convolutional Neural Network (CNN) will be developed to investigate the potential of deep learning for CT image enhancement.

Background

Computed Tomography (CT) is widely used to diagnose diseases by producing detailed images of the body. However, reducing radiation exposure can increase image noise and make important structures harder to see. Artificial Intelligence (AI) offers new ways to improve CT image quality while helping preserve important anatomical details.

Aim & Purpose

To investigate the effectiveness of a lightweight Convolutional Neural Network (CNN) in reducing noise in CT images while preserving anatomical details, and to compare its performance with conventional image-denoising techniques.

Result

Preliminary results showed that median filtering achieved the highest SSIM values of 0.9192, 0.7126 and 0.5045 at noise levels of 0.03, 0.08 and 0.15, respectively. It also produced the highest PSNR at the two higher noise levels, while bilateral filtering achieved the highest PSNR of 37.2793 dB at the lowest noise level. These findings indicate that denoising performance varies with noise intensity and the image-quality parameter used.

Email address kaukabfatima23@gmail.com
Phone number 03128663933

Author

Kokab Fatima (Ned university of Engineering and Technology)

Presentation materials