Methodology to create 3D models of COVID-19 pathologies for virtual clinical trials

Abstract

. Purpose: We describe the creation of computational models of lung pathologies indicative of COVID-19 disease. The models are intended for use in virtual clinical trials (VCT) for task-specific optimization of chest x-ray (CXR) imaging. Approach: Images of COVID-19 patients confirmed by computed tomography were used to segment areas of increased attenuation in the lungs, all compatible with ground glass opacities and consolidations. Using a modeling methodology, the segmented pathologies were converted to polygonal meshes and adapted to fit the lungs of a previously developed polygonal mesh thorax phantom. The models were then voxelized with a resolution of and used as input in a simulation framework to generate radiographic images. Primary projections were generated via ray tracing while the Monte Carlo transport code was used for the scattered radiation. Realistic sharpness and noise characteristics were also simulated, followed by clinical image processing. Example images generated at 120 kVp were used for the validation of the models in a reader study. Additionally, images were uploaded to an Artificial Intelligence (AI) software for the detection of COVID-19. Results: Nine models of COVID-19 associated pathologies were created, covering a range of disease severity. The realism of the models was confirmed by experienced radiologists and by dedicated AI software. Conclusions: A methodology has been developed for the rapid generation of realistic 3D models of a large range of COVID-19 pathologies. The modeling framework can be used as the basis for VCTs for testing detection and triaging of COVID-19 suspected cases.

Procurement, commissioning and QA of AI based solutions: An MPE’s perspective on introducing AI in clinical practice

“…The same images may be useful for testing AI based solutions. In a recent paper, an AI algorithm has been applied to detect simulated Covid-19 pathology in simulated x-ray images [34]. Simulated images have been used to train AI algorithms for detecting the COVID-19 from 2D chest x-rays [35] and it is likely that this approach will be implemented for many more applications in rare diseases or for emerging technologies, making it a topic of interest for the MPE.…”

Development and validation of a 3D anthropomorphic phantom for dental CBCT imaging research

“…Two oral radiologists, each with 10 years of experience, scored the realism level of the developed models, as presented in Table 1, from the simulated images. For scoring, the five‐point scale shown in Table 2 was used 11 . The position of the tasks was indicated, as the purpose of the study was to assess model realism, and not to conduct a search study.…”

“…Reader study A reader study 11 was set-up to assess the realism level of the normal anatomy of the model, as well as the pathosis (seven types of RFs), and the modelled restorations (five fillings,one implant,and three RCTs).In addition, for the modelled restorations, the realism level of the generated artefacts was also scored. The RFs were modelled in frontal as well as posterior teeth.…”

“…One of the crucial steps in setting‐up a VIT platform is the development of computational patient models, representative of the clinical population under study, containing clinically interesting tasks. Virtual anthropomorphic phantoms have been previously developed for various applications in medical imaging 9–19 . For dental CBCT imaging in particular, several phantoms are available for dosimetry applications 10,12–16,18 .…”

“…Virtual anthropo-morphic phantoms have been previously developed for various applications in medical imaging. [9][10][11][12][13][14][15][16][17][18][19] For dental CBCT imaging in particular, several phantoms are available for dosimetry applications. 10,[12][13][14][15][16]18 Nevertheless, for image quality studies, the available models do not suffice for various reasons: lack of detailed dental and cranio-maxillofacial structures, the absence of clinically interesting tasks, and in case of voxel phantoms, often insufficient spatial resolution of the model.…”