Texture features plays an important role in radiomics. To make the texture features rotation-invariant, pyradiomics computes the texture features along all directions and use their mean values. In this study, we demonstrated that maximum and minimum values of these features along different directions, which is also rotation-invariant, may provide added value to radiomics studies. We trained models using mean, maximum and minimum values of texture features along different directions to classify clinically significant (CS) prostate cancer (PCa) and non-CS PCa on PROSTATEx dataset. We found that using extremum instead of mean texture features improved the performance of model.