20–21 Nov 2026
Online Conference
Asia/Karachi timezone

From Filtered Back-Projection to Interpretable Self-Supervised Denoising: A Comparative Study of Image Reconstruction for Low-Dose CT in Radiation Oncology

21 Nov 2026, 13:35
5m
Online Conference

Online Conference

Online Conference

Speaker

Noor Ul Saba (Riphah International University)

Background

Low dose CT offers less radiation dose images come with increased blur making precise outlines of body areas of interest harder to spot when designing treatment plans for cancer care. Analysts often still rely on the old method of mapping data backward before boosting signal known across medicine.

Discussion (methods + interpretation)

Our implementation included Fourier slice gridding back projection and filtered back projection (FBP) utilizing five different apodization filters. We assessed these methods using PSNR and SSIM considering various projection numbers and levels of Poisson noise. Subsequently we applied Filter2Noise (F2N) a previously published self supervised bilateral filter that uses attention guidance to a single simulated low dose chest slice.

Result

Using 180 projections FBP (Shepp Logan filter) achieved a PSNR of 29.94 dB (SSIM 0.926) while unfiltered back projection resulted in 14.89 dB for 45 projections it yielded 25.44 dB. With the lowest dose (5×10³ photons/bin) the Hann window attained SSIM 0.837 compared to 0.557 for the ramp filter. F2N enhanced a simulated low dose chest slice from 34.98 dB/SSIM 0.8206 to 38.70 dB/SSIM 0.9014 (+3.72 dB) without utilizing a clean reference.

Aim & Purpose

It sought to assess and explain the performance of the self supervised Filter2Noise denoiser when applied to a simulated low dose CT slice. This work is significant because reducing radiation dose in image guided radiotherapy necessitates clear data efficient techniques that maintain image quality relevant for contouring.

Email address shameerawan862@gmail.com
Phone number 03317423327

Author

Noor Ul Saba (Riphah International University)

Presentation materials