Hybrid Deep Learning and Fuzzy Vector Quantization for ROI-Preserving Medical Image Compression
Abstract
Effective medical image compression is essential for cloud-based healthcare storage systems and telemedicine. When reaching large compression ratios, traditional compression methods can jeopardize diagnostic accuracy. In this study, a unique hybrid compression framework with a Region of Interest (ROI)-preserving mechanism that combines Deep Learning, Fuzzy Vector Quantization (FVQ), and Discrete Wavelet Transform (DWT) is proposed. Critical diagnostic regions (ROI) are automatically separated from background areas using a U-Net design. While non-ROI regions are compressed using Fuzzy C-Means Vector Quantization to achieve great overall compression efficiency, the ROI is compressed using a lossless DWT-SPIHT technique to retain sensitive clinical features. According to experimental results, the suggested approach outperforms state-of-the-art techniques in terms of Reconstruction Quality (PSNR > 45 dB in ROI) and considerable Compression Ratios, making it ideal for real-time medical data transfer.