Iterative Model Reconstruction: Simultaneously Lowered Computed Tomography Radiation Dose and Improved Image Quality

Computed Tomography (CT) scanner technology has progressed rapidly throughout the past decade, with major advances in CT x-ray detection, system speed, and image reconstruction that have resulted in a stillincreasing number of novel CT clinical applications. Concomitant to this increase in clinical applications, CT technology has also experienced innovation to support the realization of high CT image quality while adhering to the As Low As Reasonably Achievable (ALARA) principal of radiation dose management. Notable among these innovations are the introduction of modular detector designs that are optimized for low-energy, low-noise data acquisition (e.g., NanoPanel Elite, Philips Healthcare) and the introduction of innovative reconstruction techniques (e.g., iDose, Philips Healthcare) that improve image quality at low-dose, and exhibit reconstruction times that fit within traditional CT workflow. The evolution to knowledge-based iterative reconstruction algorithms that utilize additional system information to enable significant CT radiation dose reduction and image quality improvement is the next step in CT technology innovation. Although these more advanced algorithms have been used in single-photon-emission computed tomography and positron-emission tomography for some time, their use in CT was historically limited by long, clinically unacceptable, reconstruction times. Recently, IMR (Iterative Model Reconstruction, Philips Healthcare), combined with new computational hardware, has demonstrated simultaneous significant improvements in image-quality and significantly lower dose with reconstruction times of less than 5 minutes for a majority of reference protocols. Phantom tests demonstrate that IMR may simultaneously enable 60% – 80% lower radiation dose, 43% – 80% low-contrast detectability improvement, and 70% – 83% less image noise, relative to filtered back projection. Alternatively, IMR may enable 1.2x – 1.7x highcontrast spatial-resolution improvement; or 2.5x – 3.6x lowcontrast detectability improvement; or 73 90% image noise reduction, relative to filtered back projection. This article provides a review of the algorithm and its performance characteristics based on phantom studies. Keywords— Iterative Reconstruction, Knowledge-based, Model-based, IMR