Volumetric Reconstruction and Web‐Based Visualization of Giga‐resolution Anatomical Datasets Using 3D Point Clouds: From Full‐color Image Sequences to Spatial Anatomy

Abstract

Introduction The Visible Human Project (VHP) data sets are comprehensive visual archives that provide a reliable submillimeter registration of human anatomy. The evolution of computer graphics and the use of standardized pipelines have offered a general improvement in the three-dimensional (3D) visualization of data primarily obtained from radiological and histological sections. Due to the massive amount of high-resolution image sequences found in the data sets, there is currently a lack of complete and accessible 3D versions of the VHP. Recently, digital representations based on a set of 3D points following a given coordinate system, called “point clouds” (3D-PC), have become an essential tool for the study of topographical models in the geospatial field, with a wide variety of applications. This study describes a 3D-PC-based workflow developed by our team with the aim of online volumetric visualization, interaction, and democratization of the VHP data sets. Materials & Methods Three VHP data sets (i.e., VHP male, VHP female, and VHP 2.0) containing full-color cryo-sections were acquired from the online repository of the National Library of Medicine. The workflow for each one of the data sets was divided into four steps: 1) Batch-processing of image sequences: the image files were converted into TIFF files, and the pixels contained in the files were processed into arrays with fixed scale size, 3D position, and color value using in-house software (tif2laz v1.0). 2) Images to 3D-PC: the previously assigned values were used for the creation of volumetric models formatted as 3D-PC, and the resulting file was saved as a compressed geospatial file (i.e., .laz) with a converter (tif2laz v1.0) to facilitate its online visualization and interaction. 3) 3D Filtering: the 3D-PC model was previewed, and the existing noise was removed using 3D software (CloudCompare v.2.11.2). 4) Deployment: the model was subsequently deployed on an octree-based 3D-PC renderer platform (Potree v.1.7.2), and an ordinary web server was created for the online use of the final models. Results Total and segmented visualization of the models, including interactions with 3D clipping boxes, and measurement tools, were possible using the online platform. The process data characteristics were as follows: 1) VHP male dataset (Fig. 1A & 2A): 4.3 billion 3D points obtained from 1871 images (4096 x 2700 resolution) at 1 mm interval. Data set size: 64 GB. 2) VHP female (Fig. 1B & 2B): 12 billion 3D points from 5189 images (4096 x 3061 resolution) at 0.33 mm interval. Data set size: 189 GB. 3) VHP 2.0 (Fig. 1C & 2C): 1.5 billion 3D points from 1481 images (1056 x 1528 resolution) at 0.5 mm interval. Data set size: 17 GB. Conclusion The use of 3D-PC demonstrated to be a cost-efficient method for the spatial representation of the anatomical geometry generated from the VHP data sets. The advent of novel computer graphic and geospatial technologies along with the optimization of online 3D rendering techniques, would likely offer new opportunities for the education and research of giga-resolution anatomical data sets. Further research needs to be done to elucidate the potential applications to the use of 3D-PC in anatomy and medical imaging.