KryogenBlue
commented
1 month ago Dear LDTS-lab,
Thank you for your interest in my work and for successfully setting up the project. I appreciate your inquiry regarding the test data. Unfortunately, due to the nature of the dataset used in this project, I am not authorized to share the specific data. It is governed by an ethics review and cooperation protocol that restricts its access to our research lab. Consequently, the validation and test datasets used in the repository are private.
However, I would like to emphasize that this repository is essentially a GPU-accelerated implementation of gjadick’s "multi-mat-decomp" project, which closely replicates the algorithm from the paper titled "A Flexible Method for Multi-Material Decomposition of Dual-Energy CT Images" proposed by General Electric in 2013. For a more detailed tutorial on the original algorithm and its validation, you may refer to the scratch.ipynb file in gjadick’s repository. Although that implementation runs on a CPU, and thus may require longer processing times, it should provide valuable insights into the algorithm's principles and performance, which I believe aligns with your current objective of understanding the underlying method.
My repository specifically focuses on accelerating the decomposition of large batches of DECT datasets using GPUs. This means that to fully utilize the code, you would need access to substantial amounts of DECT data, which, as you might know, is relatively scarce due to the high costs associated with clinical DECT scans and the restrictive nature of cooperation agreements between institutions.
While I am unable to provide access to the specific dataset used in my work, there are alternative paths you can explore to obtain DECT datasets. You may consider using publicly available datasets or requesting access to datasets shared by researchers in relevant publications. For instance, the 2022 AAPM Grand Challenge on Deep-Learning Spectral Computed Tomography (DL-spectral CT) provides a DECT dataset, though it is limited to three materials. Additionally, generating simulated CT datasets might serve as a viable option for your testing purposes.
For any further questions, feel free to reach out!
Best regards, Xiaoheng Li
Dear Author,
I am very interested in your work. I have successfully cloned and set up the project, but I noticed that test data is not included. To better understand and validate the performance of the project, could you kindly provide the test data or point me in the direction of where I could obtain it?
Thank you very much for your assistance!
Best regards,