No-reference based automatic parameter optimization for iterative reconstruction using a novel search space aware crow search algorithm

A research team from institutions including the University of Cambridge and MedUni Vienna has published a novel AI framework that automates the complex parameter tuning required for iterative reconstruction in Cone-beam computed tomography (CBCT). The method, based on a modified 'crow search algorithm' (CSA), introduces a search-space-aware global strategy and a superior local search mechanism to find optimal settings without needing a reference image for comparison. This solves a major bottleneck: manually tuning hyperparameters for these algorithms is notoriously time-consuming and operator-dependent, yet drastically affects final image quality and the ability to reduce patient radiation dose.

The team's 'no-reference' approach was rigorously tested on three different imaging machines and four real-world datasets using three challenging iterative reconstruction methods. Results showed the automated system consistently outperformed expert manual parameter settings. It achieved a 4.19% improvement in average fitness and specific gains of 4.89% and 3.82% on established no-reference quality metrics (CHILL@UK and RPI_AXIS). Qualitatively, the AI-optimized scans better preserved fine anatomical details. The proposed chaotic initialization scheme also helped accelerate the algorithm's convergence, making the optimization process more efficient.