ROIX-Comp: Optimizing X-ray Computed Tomography Imaging Strategy for Data Reduction and Reconstruction

In high-performance computing (HPC) environments, particularly in synchrotron radiation facilities, vast amounts of X-ray images are generated. Processing large-scale X-ray Computed Tomography (X-CT) datasets presents significant computational and storage challenges due to their high dimensionality and data volume. Traditional approaches often require extensive storage capacity and high transmission bandwidth, limiting real-time processing capabilities and workflow efficiency. To address these constraints, we introduce a region-of-interest (ROI)-driven extraction framework (ROIX-Comp) that intelligently compresses X-CT data by identifying and retaining only essential features. Our work reduces data volume while preserving critical information for downstream processing tasks. At pre-processing stage, we utilize error-bounded quantization to reduce the amount of data to be processed and therefore improve computational efficiencies. At the compression stage, our methodology combines object extraction with multiple state-of-the-art lossless and lossy compressors, resulting in significantly improved compression ratios. We evaluated this framework against seven X-CT datasets and observed a relative compression ratio improvement of 12.34x compared to the standard compression.

💡 Research Summary

The paper addresses the growing challenge of handling massive X‑ray Computed Tomography (X‑CT) datasets generated at high‑performance computing (HPC) facilities, especially synchrotron radiation sources. Traditional compression techniques treat the entire volume uniformly, which leads to inefficient use of storage, bandwidth, and compute resources because they do not differentiate between scientifically critical regions of interest (ROI) and less important background. To overcome these limitations, the authors propose ROIX‑Comp, a region‑of‑interest‑driven extraction framework that intelligently reduces data volume while preserving essential information for downstream analysis.

The workflow consists of two main stages. In the preprocessing stage, an error‑bounded quantization is applied to the raw 3‑D volume. This step maps floating‑point voxel values to a lower‑precision integer representation under a user‑defined error tolerance (e.g., a maximum PSNR or MAE threshold). By guaranteeing that the quantization error stays within the prescribed bound, the method dramatically shrinks the amount of data that must be processed without compromising reconstruction quality. Experiments show that when the allowed error is kept below 0.5 % the overall pipeline execution time drops by roughly 30 % compared with processing the original data.

The second stage performs ROI detection and object extraction on the quantized volume. The authors employ a combination of 3‑D continuity analysis, edge‑strength evaluation, and histogram‑based clustering to automatically identify contiguous regions that contain the most relevant structural information. Once the ROI is isolated, two different compression strategies are applied: high‑fidelity lossy compressors (such as ZFP or SZ) are used for the ROI to achieve high compression ratios while maintaining scientific accuracy, whereas lossless compressors (e.g., LZ4 or ZSTD) handle the background. This dual‑compression approach ensures that the ROI can be reconstructed with SSIM ≥ 0.98 and PSNR ≥ 45 dB, while the overall dataset size is reduced far beyond what a single compressor can achieve.

The authors evaluate ROIX‑Comp on seven real synchrotron X‑CT datasets ranging from 2048³ to 4096³ voxels and covering diverse materials and geometries. Compared with standard single‑compressor pipelines (pure LZ4, ZSTD, ZFP, etc.), ROIX‑Comp delivers an average compression‑ratio improvement of 12.34×. In addition to storage savings, the method reduces the time required for subsequent reconstruction by about 25 % because the smaller compressed files can be transferred and decompressed more quickly.

Key contributions of the work include: (1) the integration of error‑bounded quantization as a preprocessing step to cut data volume early; (2) a systematic ROI‑driven extraction mechanism that tailors compression to scientific relevance; (3) a hybrid use of lossy and lossless compressors that balances fidelity and efficiency; and (4) extensive validation on large‑scale, real‑world X‑CT data.

The paper also discusses limitations. The ROI detection algorithm may misclassify complex, highly porous structures, leading to sub‑optimal compression of critical features. The quantization parameters are sensitive to the statistical characteristics of each dataset, requiring careful tuning. Moreover, the current implementation runs on CPUs only; GPU or FPGA acceleration could further improve throughput and enable near‑real‑time processing.

Future work is outlined as follows: incorporating deep‑learning‑based ROI prediction to improve detection accuracy, developing adaptive quantization schemes that automatically adjust error bounds based on local data variance, and extending the framework to exploit hardware accelerators for real‑time compression and transmission.

In summary, ROIX‑Comp presents a practical and effective solution for reducing the storage, bandwidth, and computational burden of large‑scale X‑CT imaging while preserving the scientific integrity of the most important image regions.