Towards scalable medical image compression using hybrid model analysis
Shunlei Li, Jiajie Lu, Yingbai Hu, Leonardo S. Mattos, Zheng Li
Abstract
Abstract The exponential growth in medical image generation poses significant challenges for storage and management. Lossless compression of medical images is essential to reduce storage demands while ensuring image quality is preserved. Wavelet-based compression techniques, widely recognized in the literature, are commonly used to process and transmit medical images by isolating the Region of Interest (ROI) from other areas. Meanwhile, Convolutional Neural Networks (CNN) have shown promising results for medical image compression. In this study, we propose a hybrid model combining Discrete Wavelet Transforms (DWT) and CNN for medical image compression. DWT is applied to encode the ROI, while CNN is employed for non-ROI regions. Here, Singular Value Decomposition (SVD) is used to extract ROI features. We introduce the SDWTCNN framework, which integrates DWT and CNN to achieve scalable image compression with lower complexity compared to similar methods. The performance of SDWTCNN is evaluated using different performance metrics, demonstrating its effectiveness in maintaining image quality at various compression rates. Experimental results confirm the efficiency of our framework for storing and retrieving medical images in healthcare applications. Specifically, our SDWTCNN achieves 4.3 dB better performance on the BGPD dataset and 3.8 dB on the BraTS dataset than the existing best method in terms of the PSNR metric.