Compression and Extraction of the Region of Interest from Medical Images through Deep Learning Methods

The objective of any medical image compression strategy include reducing the bit rate and increasing compression efficiency while trying to maintain the quality of diagnostic images. A partial region-based compression method is applied in order to achieve high compression rate and at the same time maintaining the image quality. Most of the medical images contain regions of interest which are critical for making the diagnoses. These regions, however, need to be removed and recreated in high quality. The present paper aims at describing and exploring deep learning techniques which can be used to extract regions of interest in MR brain imaging. The restored and the recreated high quality regions are then compressed through both lossless and lossy techniques.