Question about LDCT simulation method - discrepancy between paper citation and implementation

[Haonan-Wang4968]

I noticed a potential discrepancy regarding the LDCT simulation method:

In the paper (Section III.A):

• You cite [44] (Zeng et al., "A Simple Low-Dose X-Ray CT Simulation") for LDCT simulation

• Equation (11) describes projection-domain noise addition:
  p_ld = ln(I_0 / (Poisson(I_0*exp(-p_hd)) + Gaussian(0, σ_e^2)))

• This method requires raw sinogram data according to [44]

In the code:

• ldct_simulation.m does implement projection-domain simulation (using MATLAB IRT for forward projection)
• But prep_mayo2016.py only reads DICOM images (already reconstructed, no sinogram)
• The public Mayo 2016 dataset doesn't provide sinogram data

My question:
Did you strictly follow the projection-domain method from [44], or use an image-domain approximation in practice?

I also checked the LD-CT-simulation repo you acknowledged - it also doesn't use raw sinogram data from scanners, but simulates via forward projection from DICOM images.

Could you clarify:

1. Was ldct_simulation.m actually used to generate training data?
2. Or did you use the real 25% dose scans from Mayo's quarter_1mm/ folder?

This would help for reproducibility. Thanks!