Compression of medical image is a challenging task as the compression has to be achieved without losing diagnostic quality of the images. Reducing the size of image finds its application in saving storage space and increasing transmission speed, as in case of teleradiology. On the decomposed details of an image, prediction is done. The high SNR (Signal to Noise ratio) and localization provided by wavelet coefficients makes it suitable for images. With few prediction coefficients, as small as 15, the whole image is reconstructed. The four variants of the method are studied to find its suitability for different types of images. The method is evaluated using various objective fidelity criteria.