Region of Interest Based Medical Image Compression Using DCT and Capsule Autoencoder for Telemedicine Applications

Medical imaging has a significant role in today’s healthcare system. Advancement in imaging modalities leads to a remarkable proliferation of resolution and volume of digital data. Telemedicine based on a centralized database requires effective storage, processing, and digital data transmission over the internet. Even after the tremendous advancement in electronic communication techniques, it is still challenging to transmit this huge volume of medical data over the internet efficiently. This leads to an inevitable requirement of image compression that reduces the data size and hence the transmission requirements. In this paper, we offer an ROI-based image compression technique for MR images, which tackles the frequency components present in the processed medical image. The Fuzzy C-Means clustering approach is used to separate the region of interest from the non-region of interest. Capsule autoencoder method is used for compressing the non-ROI and Discrete Cosine Transform with HuffmanRun-length encoding is used for the compression of the region of interest. A detailed analysis of the compression process in terms of various evaluation functions like peak signal to ratio and compression ratio has been carried out and the result shows superior performance over the previous methods.