Which is the better option for performing radiomics analysis on CT images? Should radiomics features be extracted directly from CT images with HU values, or should the CT images be converted to ED maps first before extracting features?

The reason I am asking this question is: although there are many great pieces of research looked into the repeatability and reproducibility of radiomics features based on CTs, including test-retest, intra-, and inter- CT scans, however, I found that few of them talked about whether they calculated the features based on HU images or ED images. So I am quite curious, for instance, if we want to calculate the radiomics feature reproducibility between different CT scanners, in order to eliminate the potential difference from the scanner's physical setting (such as energy, phantom setup error, etc..), which method is more convincing and fair?