Description
Introduction
Hybrid-pixel detectors (HPDs) and in particular the subclass of photon-counting detectors (PCDs) have recently enabled substantial advances in clinical CT through improved dose efficiency, electronic noise suppression, and spectral imaging capabilities. However, these advantages have not yet translated to planar X-ray imaging, despite its seemingly simpler acquisition geometry.
CT benefits from substantial information redundancy as individual voxels source information across hundreds of detector pixels, making CT comparatively robust against detector imperfections such as dead regions, sensor inhomogeneities, ASIC border effects, calibration errors, and limited spatial sampling. In planar radiography, however, image quality depends directly on the local detector response, making these limitations substantially more critical. Consequently, many of the practical constraints of PCDs remain more pronounced in radiography than in CT, limiting their clinical exploitation.
Employing mechanically super-sampled radiographic acquisition with PCDs (PC-SSI) addresses these challenges by combining defect-aware detector motion and model-based image reconstruction to recover information lost through detector architecture limitations. Beyond mitigating detector defects and spatial sampling constraints, the approach partially decouples reconstructed image resolution from detector pixel pitch. This enables optimization of detector architectures for spectroscopic performance, count-rate capability, and per-pixel functionality without a proportional loss in image resolution.
Material & Methods
A custom-built µCT system equipped with a high-Z GaAs PCD (SANTIS HR 0804 prototype, DECTRIS AG; 500 µm sensor thickness; 75 µm pixel pitch), a micro-focus X-ray tube (MXR Microbox 100, Micro X-ray Inc.), and a beam-hardening calibration unit was used. Detector trajectories were designed to span multiple pixels while avoiding known detector defects and ASIC border regions, thereby averaging out local field non-uniformities within the sensor. The imaging pipeline combined signal-to-equivalent-thickness calibration (STC), enhanced correlation coefficient (ECC) registration for sub-pixel motion estimation, and iterative reconstruction using distance-driven regularized maximum likelihood expectation maximization (MLEM) with total variation regularization.
As a clinically relevant demonstration, the method was evaluated for mammographic imaging, where simultaneous demands on spatial resolution, spectral information, and radiation dose represent a particularly challenging application scenario.
Results
Phantom experiments demonstrated a more than three-fold improvement in effective spatial resolution compared with native detector imaging, resolving structures down to ~20 µm using a detector with a native pixel pitch of 75 µm. In a breast phantom (Mam AI-Sim, PhantomX GmbH), PC-SSI provided improved visualization of calcification-like structures compared with a clinical mammography system (MAMMOMAT B.brilliant, Siemens Healthineers) at equivalent dose and maintained superior visibility at half of the reference dose. Post-mortem mouse radiography demonstrated a substantial improvement in spatial resolution and overall image quality.
By combining information from multiple detector elements for each reconstructed image pixel, the method simultaneously reduced the impact of detector defects and local sensor non-uniformities while improving the robustness of spectral measurements.
Conclusion
PC-SSI addresses several practical limitations that currently restrict the broader adoption of PCDs in clinical radiography. Beyond improving spatial resolution and mitigating detector imperfections, the approach challenges the traditional trade-off between detector pixel pitch and image resolution. This may enable future PCD architectures optimized primarily for spectroscopic performance, count-rate capability, and advanced pixel functionality, opening new opportunities for super-resolved and spectrally enhanced clinical X-ray imaging.