Deep learning based medical image compression using cross attention learning and wavelet transform

Abstract

Abstract Efficient compression of medical images is vital for telemedicine and cloud-based healthcare, where bandwidth and storage constraints pose significant challenges. Conventional lossless approaches provide limited compression, whereas lossy techniques risk compromising diagnostic accuracy. To address these limitations, we introduce a novel hybrid compression framework that combines Discrete Wavelet Transform (DWT) with a deep Cross-Attention Learning (CAL) module to preserve clinically relevant details while reducing redundant information. The proposed pipeline first decomposes input images into multi-resolution sub-bands via DWT, followed by a CAL-driven encoder that emphasizes high-information regions through dynamic feature weighting. A lightweight Variational Autoencoder (VAE) refines feature representation prior to entropy coding for final compression. Extensive experiments on benchmark datasets, including LIDC-IDRI, LUNA16, and MosMed, demonstrate that our approach achieves superior performance in terms of PSNR, SSIM, and MSE compared to state-of-the-art codecs such as JPEG2000 and BPG. These results highlight the method’s potential for real-time medical image transmission and long-term storage without sacrificing diagnostic integrity.