Formalising the Signal Processing Specification for Photon-Counting CT Radiomics: Stability Conditions Derived from Signal Chain Analysis

Background: Photon-counting computed tomography (PCCT) is the biggest detector-level change in clinical CT in fifty years, removing electronic noise and making every scan spectral. Radiomics pipelines built for energy-integrating CT cannot be assumed to carry over, and multiple phantom studies report feature instability across PCCT acquisition and reconstruction parameters. No formal specification exists for what a stable PCCT radiomics pipeline requires. Methods: We model the PCCT acquisition-to-feature pipeline as a signal chain and apply first-order sensitivity propagation, a standard method in sensor and measurement-system engineering. We characterise the Poisson counting noise model, the energy-bin threshold vector, the detector modulation transfer function, and the reconstruction kernel frequency response, and define feature stability as a bounded sensitivity of each Image Biomarker Standardisation Initiative (IBSI) feature class to each acquisition parameter. Results: The derived conditions predict that shape features are the most stable class, that first-order features are governed by virtual monoenergetic image (VMI) energy, and that texture and wavelet features are governed by the reconstruction kernel and detector transfer function. As a quantitative test, the model derives a first-order stability crossover near 93 keV from public NIST attenuation data alone, coinciding with the empirical reproducibility collapse below 90 keV reported in a published phantom study. Conclusions: Signal chain analysis converts the empirical catalogue of PCCT radiomics instability into closed-form stability conditions. The framework specifies which acquisition parameters must be standardised, informs radiology deployment and regulatory validation, and identifies the acquisition-side parameters that lie outside current IBSI scope.