Efficient Medical Image Compression Using a Hybrid Deep Learning Framework: Integrating Transform Coding and Neural Architectures for Enhanced Diagnostic IntegrityOpen Access

Abstract

Medical image compression is essential for maintaining diagnostic accuracy while reducing the storage and transmission requirements of medical image data. This study presents a hybrid deep learning model that integrates conventional transform coding methods, including wavelet and discrete cosine transform for frequency domain coding, autoencoders for dimensionality reduction, and convolutional neural networks or spatial feature extraction. The approach significantly reduces the dimensions of medical images and achieves high compression ratios while preserving essential diagnostic information. Performance analysis shows that the proposed model outperforms previous methods, achieving a compression rate of 68%, a peak signal-to-noise ratio (PSNR) of 46, and a structural similarity index (SSIM) of 0.91. The results indicate that the model achieves exceptional compression efficiency and image quality, making it suitable for practical medical imaging applications that require significant compression and image quality preservation.