XSEDE-enabled high-throughput lesion activity assessment

Abstract

Caries lesion activity assessment has been a routine diagnostic procedure in dental caries management, traditionally employing subjective measurements incorporating visual and tactile inspections. Recently, advances in 2D/3D image processing and analysis methods and microfocus x-ray computerized tomography (μ-CT) hardware, together with increased power of high performance computing, have created a synergic effect that is revolutionizing many fields in dental computing. In this paper, we report such an XSEDE-enabled high-throughput lesion activity assessment workflow that exploits 2D/3D image processing, visual analytics, and high performance computing technologies. Our paper starts with a brief introduction of the image dataset in our dental studies. We then proceed to a family of 2D image analysis, ROI segmentation, and 3D geometric construction methods. By combining dental imaging technology and 2D/3D image processing algorithms, we transform the task of lesion activity assessment into a 3D-time series analysis of computer generated lesion models. Building on the computational algorithms and implementation models, we develop a high-throughput dental computing workflow exploiting MapReduce tasks to parallelize the image analysis of dental CT scans, the segmentation of region-of-interest (ROI), and the 3D construction of lesion volumes. We showcase the employment of 3D-time series analysis and several other information representations that are applied to our lesion activity assessment scenario focusing on large scale dental image data.