Boundary artifact reduction in projection-based material decomposition using cluster analysis in Timepix3 detector

We explore the utility of photon counting spectral detectors in projection-based material decomposition for applications such as in mammographic imaging. These detectors, with their ability to count individual x-ray photons and measure their energy and time of arrival offer significant advantages. Material decomposition using photon counting detectors (PCDs) could result in artifacts, particularly at the transitions between different materials, especially of high attenuation like metals or bone when embedded in soft tissue. This paper presents a novel approach to eliminate these artifacts at material boundaries in projection domain. We perform pixel classification or clustering based on the estimated mass attenuation for two energy windows to reduce the number of material bases in decomposition. Furthermore, we introduce an iterative method that allows significant reduction of artifacts, especially at material edges. Our findings reveal the potential of Timepix3 in biomedical imaging and offer a robust technique for improved material decomposition for improved diagnostic interpretations in pre-clinical and clinical spectral CT.