Introduction  
Methods
Results
Conclusions
References
 

Introduction

      In April 2000 the National Library of Medicine (NLM) posted a beta test site titled “AnatLine,”  providing a variety of new resources derived from The National Library of Medicine's Visible Human Project (VHP).  Included among these resources disseminated via the WWW were digitizations of the 70mm photographic images from the VHP Male; the segmentation byte masks of thoracic structures prepared by EAI, Inc. under contract with NLM; and some rendered scenes.  These data are stored in an elaborate database structure designed and implemented by NLM, utilizing the vocabulary of  NLM's Unified Medical Language System (UMLS).

      Developers have made fantastic strides in using the VHP data in ways not initially anticipated.   However, the task of segmenting individually identified structures in the data has been a major concern, and the topic of numerous reports in the literature [1, 2, 3, 4, 5], as it represents the rate limiting factor in the development process.  Presentations at this conference addressed hybrid methods of segmentation [6] and demonstrated elegant graphic supplements to segmentation for smaller nerves and vessels [7].  The AnatLine Database (AnatLine Version Beta 1.0) provides segmentation masks for a considerable number of structures in the thorax of the VHP Male.  Ackerman  [8] recently described the Visible Human Project's history and his vision of its educational potential, and Spitzer [9] reported at this meeting progress in NIH's multi-institute project to create a functional atlas of head and neck anatomy. The potential is already being realized, as many developers have described applications of Visible Human Project data in education.   At the undergraduate level, Senger [10] reported using the Visible Human Project in an anatomy and physiology course, and inclusion of spline models in future Medical College Aptitude Tests (MCAT) has also been proposed  [11].  Applications in medical education abound, whether in conjunction with gross anatomy [12], or to correlate with cross sectional [13] or imaging [14]  anatomy.  Further advances in educational uses of the VHP data were discussed at this meeting, including suggestions that computer modeling might contribute to increased understanding by anatomists as well as students [15].  Many applications such as the University of Washington's Digital Anatomist  [16] and cross sectional atlases (e.g.,  [17])  are accessible via the WWW.  The ability of the computer to generate scenes on demand   [18] requires implementation of an internally consistent knowledge system, the topic of several papers at this meeting ( [19,  20,  21]).   An additional indication that the complexity of developments appears to be rising to another level is the addition of functional correlations such as ECG [22], simulations [23,  24] and animations [25,  26].   

      The implications of three dimensionality for enhancing anatomical understanding from viewing computer models has been pointed out [27, 28], and interactive 3D lessons are being incorporated into the curriculum at UCSD and Uniformed Services University of the Health Sciences [29]  This paper describes one way of processing the VHP data to create Virtual Reality Modeling Language (VRML) models,  which developers in such diverse locations as Australia, Germany, France, and the United States are currently utilizing in connection with the anatomical sciences.

Methods

      Data are located via NLM’s Database Query search engine  (Figure 1), which utilizes relationships and unique identifiers of the UMLS.  Guidance is provided regarding valid terms available for search in the database (Figure 2).  A search for “heart located indexing of the term in the male dataset, represented in one Volume of Interest file, 138 cross sections, one rendered image, and one segmented byte mask file (Figure 3).  Related terms for Physiological System (cardiovascular system), Body Regions (mediastinum, thorax, thoracic cavity) and Substructures (pericardium, right side of heart, left side of heart, right atrium, left atrium, right ventricle, left ventricle) are also returned (Figure 4), based on relationships in the UMLS and contents of the AnatLine Database.  Selecting one of the hotlinked terms under the headings of  Physiological System, Body Region or Substructures returns the contents of the database for that term (e.g., "right side of heart" Figure 5).   Selection for downloading of the various file types (cross sections, volume of interest, segmented byte mask or rendered image) whether singly or in batches is accomplished through the AnatLine GUI, which indicates the required disk storage for each file or group of files (Figure 6).  The General Information page of the AnatLine website provides links to details of the  new “ vhi” file format which NLM has designed for these data.  This file format is handled for viewing by the AnatLine program VHDisplay, and is parsed into component files by VHParser which can be downloaded from the AnatLine site.

      Several Volume of Interest files have been downloaded, including “neck,” “spine,” and “heart.”  VHParser was then used to open these files (Figure 7) extract a stack of color cross sectional images of the volume of interest, and byte masks for individual structures (Figure 8).  A software utility written by one of the coauthors (JES) was used to read the byte masks and color image files and to write a series of cross sectional image files (Figure 9) which were then used as input for his IsoView software system for producing and manipulating surface models.  This utility was used to read the ".inf" file parsed from the region of interest or byte mask ".vhi" file, further crop the AnatLine volume of interest cross section color images to contain only the specific sub-structure of interest, subsample the data to reduce the number of pixels, and set the value of all pixels not contained in the byte mask to black (Figure 10).  This makes it easy to generate the 3D surface model of the structure of interest using  threshold contours (Figure 11).

  Results

      IsoView calculated surface models based on the threshold contours of the cross sectional images parsed and expanded from the AnatLine dataset (Figure 12).   Colors of the rendered texture map were derived from the color values of the corresponding pixel in the color image stack.  Smoothing of the surface, normals and colors was performed within IsoView (Figure 13), followed by decimation to reduce the number of triangles in the models.  Extent of decimation was determined empirically in order to achieve an acceptable balance between anatomical detail and speed of movement of the model in space, keeping in mind that the primary goal is to illustrate anatomical relationships rather than to provide “photographic realism.” Similar to the "rendered image file" for Heart (DRS110018787.vhi) in the AnatLine database (Figure 14),  IsoView can be used to combine several component surfaces into a scene, and to render the scene using a data-derived surface texture map or to specify such aspects of the appearance as color, lighting, texture, transparency and shininess (Figure 15) .  Another utility program was used to write VRML (wrl) format files, for subsequent inclusion with text and other images in computer aided instructional packages. A scene including the heart, lungs partially transparent, and part of the bronchial tree was assembled in IsoView from individual AnatLine segmentations, and converted to a VRML model of the heart and lungs (Movie 1). Examples of screen captures of this model viewed and manipulated using Netscape with a Cosmoplayer plug-in are shown in  Figure 16.  A computer aided learning web page is being prepared presents the VRML world with rollovers and hyperlinks to identifications and descriptive text boxes associated with individual parts of the model.

  Conclusions