HFormer-Low-Dose-CT-Denoiser

Unofficial implementation of the paper "Hformer: highly efficient vision transformer for low-dose CT denoising". Link to original paper : https://link.springer.com/article/10.1007/s41365-023-01208-0

There are several assumptions made here, such as C is assumed to be 64, K is assumed to be 2, n is taken to be 1, linear is taken to be 1x1 convolutions, etc.

Dataset used : AAPM Low Dose CT Grand Challenge

Grand Challenge Link : The 2016 NIH-AAPM-Mayo Clinic Low Dose CT Grand Challenge by Mayo Clinic https://www.aapm.org/GrandChallenge/LowDoseCT/
To download the partial dataset, click here : https://aapm.app.box.com/s/eaw4jddb53keg1bptavvvd1sf4x3pe9h
To download the full dataset (note : uploading a agreement is required), click here : https://www.cancerimagingarchive.net/collection/ldct-and-projection-data/

Results

How to use

In train/hformer_train.py, set the dataset path and model output / history path. In the test/hformer_test.ipynb file, set the correct path to the .k5 file.

References used

• full-preprocessing-tutorial
• RED-CNN