The major distinguishing features used to assess a data hiding technique are distortion and embedding capacity. This paper presents a comparison between classical Prediction error expansion based reversible watermarking and proposed prediction error expansion scheme considering region of interest for grayscale medical images. In classical prediction error expansion the augmentation of the predicted error values is used for data embedding. In the proposed scheme, prediction error expansion by preserving the Region of Interestis used. Both the schemes focuses on Reversible data hiding where the original primary image can be remodeled losslessly after extracting the payload. A performance evaluation based on Peak Signal to Noise ratio (PSNR), total payload capacity is carried out. Additional capacity and less distortion of the primary image in comparison to the basic method is obtained through the results.