1
Collaborative Learning Using Internet2 and Remote Collections of Stereo Dissection Images Medical Education Article 1
1
2
Parvati Dev, Ph.D. , Sakti Srivastava, MBBS, MS , Steven Senger, PhD 1
Stanford University Medical Media and Information Technologies (SUMMIT) Stanford University School of Medicine 251 Campus Drive, MSOB Room 240 Stanford, CA 94305-5466 2Department of Computer Science 212 Wing Technology Center University of Wisconsin - La Crosse La Crosse, WI 54601 (2 tables and 6 figures) Running Head: Collaborative Learning using Internet2
Source of Support: This work was supported in part by NLM/NIH Contract 185N034 Corresponding author: Parvati Dev, Ph.D Stanford University Medical Media and Information Technologies (SUMMIT) Stanford University School of Medicine 251 Campus Drive, MSOB Room 240 Stanford, CA 94305-5466 650-723-8087 tel 650-498-4082 fax
[email protected]
2
Abstract We have investigated collaborative learning of anatomy over Internet2 using an application called the Remote Stereo Viewer (RSV). This application offers a unique method of teaching anatomy using high-resolution stereoscopic images in a client-server architecture. Rotated sequences of stereo image pairs were produced by volumetric rendering of the Visible female and by dissecting and photographing a cadaveric hand. A client-server application (RSV) was created to provide access to these image sets, using a highly interactive interface. The RSV system was used to provide a ‘virtual anatomy’ session for students in the Stanford Medical School Gross Anatomy course. The RSV application allows both independent and collaborative modes of viewing. The most appealing aspects of the RSV application were the capacity for stereoscopic viewing and the potential to access the content remotely within a flexible temporal framework. The RSV technology, used over Internet2, thus serves as an effective complement to traditional methods of teaching gross anatomy. Key Words: medical education; gross anatomy; stereoscopic images; distance learning; Internet2
3
Introduction The Visible Human Project (Spitzer et al, 1996) made available a very large collection of digital images of human anatomy (Ackerman, 1999). This rich data set has been used subsequently to create numerous learning resources (Schubert et al, 1997; Hoffman et al, 1997; Jastrow and Vollrath, 2002). Other projects have added resources such as the Visible Korean Human (Park et al, 2005) and Lucy 2 (Heinrichs et al, 2004). We have used the availability of such digital image collections to begin creation of a networked resource of digital anatomy images for teaching and learning (Dev and Senger, 2005). In particular, we are investigating the advantages and limitations of using the Internet as a complementary approach to traditional cadaver-based teaching. This article describes our experience in designing, implementing, and evaluating a unique method of teaching anatomy using high-resolution stereoscopic images in a client-server architecture. This work was done as a demonstration project for the National Library of Medicine under its Next Generation Internet initiative (http://www.ngi.gov). The NGI initiative seeks to demonstrate the need for and define the characteristics of the network technology required to support innovative biomedical applications. It envisions a secure, high-bandwidth low-latency network for the biomedical research community. Materials and Methods Creating an array of images for stereo viewing and interaction. We created anatomy image data sets with various dimensions of interactivity either by rendering two-dimensional views of a three-dimensional volume of cross-sectional data or by photographing rotated views of a dissected specimen. For rotational viewing and interaction, we acquire or create a series of rotated views of the anatomy at 5o intervals. The data volume or the dissected specimen may also be photographed at different tipping angles followed by rotation as above. The result is a twodimensional array of images with combinations of different rotation and tip angles. Additional dimensions of interactivity include changing transparency and layers of dissection. Visible Human data was processed to generate a series of sequentially rotated images created at intervals of 5o. 72 images were required to obtain 360 o of rotation (72 x 5 = 360 o). Figure 1 shows five representative frames, acquired at intervals of 15o. The frames were originally rendered at a resolution of 1024 x 1024 pixels, compressed for transmission using a JPEG compression algorithm, and displayed at the user desired resolution, typically 768 x 768. These images were rendered from the Visible Female CT data to show bone, muscle and skin, at increasing levels of transparency, using segmentation and rendering software developed by one of us (Senger, 1999). We have also generated rotation series of the Stanford Visible Female and other cross-sectional data using commercially available software (Amira, and Volume Graphics). The image data was stored on server computers and organized for rapid retrieval. Images are stored as compressed JPEG images (approximately 200KB), and are transported to the client on demand using a custom protocol written over the UDP (Unreliable Datagram Protocol) layer, to reduce transmission overhead compared to the more common TCP/IP protocol. The transport rate was configured to be equal to the JPEG decompression rate of the client, allowing the client to interleave receipt and decompression of the data. Acquisition of stereo images of a dissected specimen. A fresh cadaveric hand (amputated at mid-forearm) was obtained from the Cadaver Donor Program at Stanford University School of
4 Medicine. The hand was mounted on a PiXiTM mechanical turntable (Kaidan Corp; Feasterville, PA) using a custom-designed central mounting post, and the entire construct was inverted. Images were taken with a Kodak 520C digital camera with a resolution of 1764 x 1160 pixels. The turntable was rotated through 360o, in steps of 5o, to produce a set of 72 images. Background color, adequate lighting, camera shutter, and aperture settings were adjusted to produce optimum photographs. The hand was then removed from the turntable and dissected superficially. As the hand was remounted, Kirschner guide wires through the radius and ulna served as landmarks to standardize hand position relative to the first set of images. An additional set of 72 images at this progressively deeper layer of dissection was then photographed. This procedure of dissection and photography was repeated until seven sets of images were produced at increasing depths of dissection. The last set consisted exclusively of the deep layers of muscle of the hand. Figure 2 shows several representative “frames” from one dissection level of this data array of 504 images. The hand dissection data has two dimensions of interaction: rotation around the long axis of the hand, and dissection depth to expose successive layers of tissue. At each dissection layer, there are 72 images separated by 5 degrees of rotation allowing one complete rotation of the hand. Seven such dissection layers were photographed. Conceptually, this array of images is organized in the computer as a stack of seven rings, or a colosseum, so that the learner traverses around the hand by moving around a ring, and changes depth by moving between rings (Fig. 3). Technology to support viewing of remote stereo images. Stereo images were viewed on an appropriately configured computer and monitor. Teacher and students used similarly configured workstations. Each workstation (cost about $1,200) consisted of a conventional Windows computer (700 MHz Pentium3, 256 MB RAM, Windows NT) (Dell, Inc, Round Rock, Texas) augmented with a specialized graphics card (Oxygen GVX1, 3dlabs, Milpitas, California). Multiple people could view the screen, each wearing specialized glasses that were automatically controlled to alternate between left and right eye vision. A NuVision stereo enabler unit (MacNaughton, Inc., Beaverton, Oregon) was connected to the computer to control the glasses. Video conferencing software (NetMeeting, Microsoft Corporation, Redmond, Washington) was used for communication between workstations. Standard telephone lines were used as audio channels to facilitate communication between the students and the instructor. In our more recent configurations, we use computers with a 2GHz Pentium4 processor, 512 MB RAM, Windows 2000 or Windows XP operating system, and graphics cards ranging from the Oxygen GVX1 to the Wildcat4 (all graphics cards from Nvidia, Santa Clara, California or 3dlabs, Milpitas, California). The image database resided on SGI 3200 servers (Silicon Graphics Inc., Mountain View, California), on the IRIX unix operating system, locally at the SUMMIT laboratory at Stanford, and remotely at the Visualization Laboratory at University of Wisconsin, La Crosse. The image database was accessed by students and teacher through a password-protected website. Each workstation accessed the Internet through a 100 Mbps network interface card A Windows NT-based client software application, the Remote Stereo Viewer (RSV), developed by us for this purpose, was created to provide access to the hand and other similar data sets. Simple interaction using the computer mouse controlled the rotated view. Left/right mouse motion rotated the image by shifting through consecutive images within each rotational set. Vertical mouse motion shifted through the successive layers of dissection. The client software, on the learner’s computer, translates mouse motion into the appropriate image requests to the
5 server. The server transmits each requested JPEG image to the client for display. Adjacent images, separated by 5 degrees, are viewed simultaneously, producing the stereo effect. A 3D pointer that approximates the appropriate depth of the anatomy at that point, was developed for pointing out anatomical structures. Technology to support video conferencing. For our initial teaching session, with the complete first year anatomy class, in 2001, we used a widely available video conference program, NetMeeting, with an inexpensive web camera, microphone and speaker at each client station. The video image allowed the teacher to recognize the students in each group but was too small to support interaction based on “body language”. The audio communication was acceptable. In subsequent teaching sessions, we have utilized additional systems. We currently use the open source, multisite, Windows-based, video conference system, Access Grid, or its commercial instance, supplied by Insors Corp (inSORS Integrated Communications, Inc., Chicago, Illinois). Successful performance of the Access Grid for multiple simultaneous users required the high bandwidth available on Internet2. Technology to support collaborative learning. Three technologies were provided to support collaborative learning. Video and audio conferencing allowed students (and the teacher) to discuss and ask questions. Secondly, any learner could become the “leader” in selecting the image to be viewed and have all other workstations “follow” automatically. Thirdly, the leader could control the 3D pointer that was visible on all follower workstations. An important aspect of the collaboration technology was that it allowed both independent and collaborative modes of viewing. In the collaborative viewing mode, clients had simultaneous access to the server and hence the same set of images. One client controls the image and dissection layer viewed, while others observed. Control of the image orientation and pointer position can be easily shared among clients, thereby creating a "virtual classroom", in which anyone can lead the others through a set of images. Assuming or relinquishing control is a simple two click process. Any learner could break off collaboration by choosing to leave the “class” and to have independent, personal control of the image viewed. A subtle aspect of the collaborative capability was that, if the leader chose to move slowly through an image sequence, a follower could change their view locally and yet be brought back to the group view whenever the leader moved to the next image. System configuration for virtual classroom. A typical configuration is summarized in Figure 4. The teacher used a workstation configured for stereo viewing (fig. 4a). The image of anatomy is presented in stereo. The emitter device on top of the monitor communicates with the specialized glasses, synchronizing the viewing eye with the display of the left or right eye image. In figure 4b, a small group of students is at a similar workstation in a different building. They also view the image in stereo. Both the teacher and the students can control the image rotation independently. Figure 4c shows the teacher’s view of the student groups. This is displayed to the teacher on a second monitor. In this study, the students heard the teacher but did not see him. In subsequent configurations, the students were able to see and hear both the teacher and the other student groups. The three client workstations were used by student groups in the Stanford University School of Medicine Anatomy computer facility and a fourth client was available at SUMMIT in a nearby building for use by the teacher. A similar configuration, with only two workstations, was used when one teacher was instructing another teacher, at a remote campus, in the use of this new teaching tool. We tested various configurations with multiple clients and different servers.
6 Use of Internet2. Interactive retrieval of a sequence of images results in the transmission of bursts of large quantities of data. (Typical burst bandwidth for our application was 70 Mbps). The commercial Internet does not support individual use of such a high bandwidth, and results in jerky displays and dropped images. We used Internet2 (www.internet2.edu), a research and education network that is available to member universities (figure 5). Internet2 typically supports “gigabit connectivity” (1,000 Mbps bandwidth) to universities. It has additional features such as extended addressing and security that we did not use. However, we did use the “multicast” feature that allowed the 3D pointer movement at one workstation to be displayed simultaneously at all workstations, thus supporting collaborative discussion between users at different workstations at different locations. Evaluation instruments. We were interested in the performance of this application over Internet2, in the students’ reaction to this learning tool, and in user perception of its utility as the performance of the network was degraded. We used survey questionnaires, network traffic measurement tools, and a 5 point perceptual scale to assess user response. Student recruitment. The study was embedded in the required curriculum for Human Gross Anatomy for first year medical students. Of 86 first year medical students, 74 students participated in the study, and completed the questionnaire. A 20-minute supplemental lesson on hand anatomy was created and deployed via RSV. This ‘virtual hand anatomy’ session was held within one week of the students’ traditional lecture and laboratory dissection session. Multiple 20-minute sessions with the same faculty member were conducted with six or seven students each time over a two-day period. Sessions were typically about 20 min in length and were divided into three parts. The first 2 minutes were devoted to technological orientation, followed by about ten minutes of a didactic, structured presentation by Dr. Srivastava, one of the regular anatomy faculty. During the last half of the session, control of hand and pointer positions was transferred among students in a more collaborative, open-ended manner. Qualitative feedback in the form of questionnaires was collected from all students immediately after their RSV session. Results Student response to collaborative learning using stereo images. Of all respondents, 93% professed a computer experience level of ‘moderate’ or above (table 1). Over 86% of the students found the sessions “helpful’ or ‘very helpful’. Students stated that they would use the simulation for help in reviewing anatomy (89%), in a self-study mode (76%), and as part of a collaborative session (39%). Demographics: Number of respondents Number with strong/moderate computer skills Attitude: Liked the stereo images Liked the 3D pointer Usefulness in learning: Rated the stereo session as helpful/very helpful
Number of students 74 69
% of students 100 % 93 %
66 37
89 % 53 %
64
86 %
7 helpful/very helpful Added to their knowledge Helped in comprehending structure relationships Cleared doubts or confusion Wanted to use it for review Wanted to use it for self study Wanted to use it for self testing Valued collaborative style of learning
50 59
68 % 80 %
21 66 56 53 29
28 % 89 % 76 % 72 % 39 %
Table 1. The most appealing aspects of the RSV were the capacity for stereoscopic viewing (89%) and the potential to access the content remotely for review within a flexible temporal framework (89%). Numerous students requested access to other parts of the body in a similar stereo format. Additional features requested included image labels, an image-based quiz, and availability of the program on the Macintosh computer. A deterrent for home use was the need for special equipment such as the advanced graphics card and the stereo glasses. Some found the stereo viewing experience uncomfortable. At the end of the quarter, the students completed an evaluation of the course, which included some questions about their stereo viewing lesson. Students ranked dissection as the most useful learning tool. Stereo images ranked at the bottom, with the stated reason that very few stereo images were provided (only hand dissection images were made available). On the other hand, they ranked stereo images the highest among resources that they wished to see increased. Network performance of the application. The performance of the RSV application was also measured during these interactive sessions. Transport bursts of between 30-40Mbps between the servers (either at Stanford or Wisconsin) and the classrooms were typical. When the Stanford server was being used, the traffic flowed through the campus network. The backbone of this network provides bandwidths in the order of gigabits and latencies of less than 1ms. When the server at Wisconsin was being used, however, the network capabilities were quite different: Stanford’s connection to the Internet2 had a bandwidth limited to 622 Mbps in the early experiments, and the delay between Stanford and Wisconsin-LaCrosse is about 60ms round trip. The time between the learner’s query (mouse click) and image retrieval and display was usually < 7 ms for images stored on the local Stanford server, and
