<![CDATA[There are still many challenges in the field of radiomics, especially sensitivity which is influenced by many factors, one of which is noise. The large amount of bias noise that occurs reduces the performance of the diagnosis, perhaps because it does not involve wavelet derived features. The wavelet derived feature is believed to be more resistant to noise interference in the image. So, the proposed use of wavelet derived features accompanied by a comparative review of its effectiveness with the use of traditional radiomic features is implemented in this study. In this study, two liver tumor datasets were used, namely LiTS17 and 3D-IRCADb-01 data to test the reliability of features in different data. The two data are given the same treatment, namely by adding interference to the CT image by setting the SNR indicator. The label used is limited to the tumor part, the liver parenchyma is not included, so for LiTS17 a thresholding label is done first. CT images and tumor labels extracted data from traditional radiomic features and wavelet-derived radiomic features (Daubechies level 1). The statistical approach uses the ICC method in assessing agreement between observers. As a result, of the six feature groups (First-order, GLCM, GLSZM, GLDM, NGTDM, GLRLM) only NGTDM features are more effective on wavelet derived features. Whereas in the main LiTS17 data, the wavelet derived features do not have a major effect, traditional radiomic features sufficiently show ICC values in the good to excellent category.]]>
Copyrights © 2023
