DermCompressNet: integrated CD-ConvNet and discrete cosine transform for dermoscopic images compression

Abstract Telemedicine has a critical role in healthcare by supporting the information exchange between the patients and the physicians as well as between the physicians for consultation. Such technology has urgent requirements for storage reduction and efficient use of the transmission channel bandwidth. Such requirements can be achieved efficiently via developing accurate medical image compression techniques. For accurate diagnosis, lossless compression methods are recommended. However, the tradeoff between the compression ratio (CR) and the preservation of the image quality is still challenging. On the other hand, the advantages of the convolutional neural networks inspired this work to design a novel proposed system for dermoscopic image compression based on the integration of the compression direction-ConvNet (CD-ConvNet) and decompression direction-ConvNet (DD-ConvNet) with discrete cosine transform (DCT) and Huffman coding, called DermCompressNet. To reconstruct a high-quality image at the receiver, the inverse processes using the DD-ConvNet network were followed. The proposed system was evaluated by measuring several image quality metrics, namely the mean square error (MSE), peak signal-to-noise ratio (PSNR), and structural similarity index measure (SSIM), along with compression quality metrics, namely CR, and computational time (CT). The experimental results achieved 34.6 dB, 2.5, 0.85, and 56% of PSNR, MSE, SSIM, and CR, respectively. A comparison studies with the JPEG and state-of-the-art methods proved the superiority of the proposed system, showing23%, 16%, 4.3%, and 1.8% improvements in the PSNR, MSE, SSIM, and CR, respectively, compared to the JPEG.