A Survey on Medical Image Compression: From Traditional to Learning-Based Approaches

TL;DR

This survey addresses the challenge of medical image compression amid growing data volumes and strict diagnostic fidelity requirements. It introduces a two-axis taxonomy (2D vs 3D/4D data structure; traditional vs learning-based methods) and systematically reviews 2D and 3D/4D codecs, detailing transform-, predictive-, VQ-, and RLE-based traditional approaches alongside CNN/RNN/Transformer, generative, and INR-based learning paradigms. It highlights the slower adoption of learning-based codecs in clinical settings due to standardization, interpretability, and workflow integration concerns, while noting compelling gains in RD performance and the ability to model complex anatomical structure with ROI-aware and structure-guided strategies. The paper also discusses remaining challenges—such as DICOM/PACS interoperability, encoding efficiency for large volumetric data, and clinically relevant evaluation metrics—and outlines future directions like large-model–assisted universal compression, dataset distillation, and exploratory quantum approaches. Overall, a hybrid design that merges robust traditional pipelines with targeted learned components appears most practical for advancing clinically viable medical image compression.

Abstract

The exponential growth of medical imaging has created significant challenges in data storage, transmission, and management for healthcare systems. In this vein, efficient compression becomes increasingly important. Unlike natural image compression, medical image compression prioritizes preserving diagnostic details and structural integrity, imposing stricter quality requirements and demanding fast, memory-efficient algorithms that balance computational complexity with clinically acceptable reconstruction quality. Meanwhile, the medical imaging family includes a plethora of modalities, each possessing different requirements. For example, 2D medical image (e.g., X-rays, histopathological images) compression focuses on exploiting intra-slice spatial redundancy, while volumetric medical image faces require handling intra-slice and inter-slice spatial correlations, and 4D dynamic imaging (e.g., time-series CT/MRI, 4D ultrasound) additionally demands processing temporal correlations between consecutive time frames. Traditional compression methods, grounded in mathematical transforms and information theory principles, provide solid theoretical foundations, predictable performance, and high standardization levels, with extensive validation in clinical environments. In contrast, deep learning-based approaches demonstrate remarkable adaptive learning capabilities and can capture complex statistical characteristics and semantic information within medical images. This comprehensive survey establishes a two-facet taxonomy based on data structure (2D vs 3D/4D) and technical approaches (traditional vs learning-based), thereby systematically presenting the complete technological evolution, analyzing the unique technical challenges, and prospecting future directions in medical image compression.

Figures (9)

• Figure 1: Evolution of Learning-Based Compression: Natural vs. Medical. Note that a dual y-axis is used: the left axis corresponds to natural-image papers (0–600), while the right axis corresponds to medical-image papers (0–160). Therefore, curve heights are not directly comparable across domains.
• Figure 2: Taxonomy used for this medical image compression review.
• Figure 3: Timeline of representative medical image compression methods. Each dot corresponds to a representative cited work, rendered with color gradients to distinguish the methodological taxonomy. The timeline is visually divided into three phases (1970--2000, 2000--2015, and 2015--2025) to emphasize the transition from transform-based standards to volumetric deep learning and implicit neural representations.
• Figure 4: General framework of transform-based medical image compression methods.
• Figure 5: Architectural Comparison of Vector Quantization and Predictive Coding Methods for 2D Medical Image.