Pen/Tablet Human-Computer Interface Reduces Radiation Oncologist Head and Neck Cancer Case Contouring Time: Preliminary Results From A Prospective Multi-Institutional Target Volume Delineation Study
 
Authors:
Clifton D. Fuller, MD, The University of Texas Health Science Center at San Antonio; Joop Duppen; Coen R.N. Rasch, MD, PhD; Martin Fuss, MD, PhD; Samuel J. Wang, MD, PhD; Niko Papanikolaou, PhD
 
Hypothesis:
Modification of the contouring interface by use of a pen/tablet system would increase efficiency for target delineation in complex head and neck cases.
 
Introduction:
Target volume delineation (wherein tumor volumes and organs at risk are defined as regions of interest (ROIs) on axial CT-slices) though arguably the most critical step in dose prescription for conformal radiotherapy, is both operator dependent and time intensive. We sought to determine whether ergonomic modification of user interfaces might improve the contouring task in terms of usability and throughput.
 
Methods:
A standardized head and neck cancer case report, and accompanying images from the patients CT/PET DICOM dataset, were distributed to radiation oncologists at 5 academic institutions. Users contoured the same case twice; once each using a standard mouse interface and a pen-tablet interface (Figure 1; Cintiq 2100, Wacom, Vancouver, WA) in a custom target volume delineation evaluation software environment (Big Brother, Netherlands Cancer Institute, Amsterdam, NL; a representative software screenshot is shown in Figure 2). Users contoured 6-7 ROIs per case. Active task time (total session time disregarding pauses > 10 sec.) and cumulative number of ROI alterations (erased points, replaced points, deleted slices, deleted contours) were tabulated. Pairwise, the Wilcoxon rank-sum test was used to compare mouse/tablet data. An electronic survey was used to asses which device users would prefer to use for future cases. A preliminary estimation of cost-effectiveness was made utilizing published academic physician salary (Wilson et al., 2005), assuming a 55-hour physician workweek.

Figure 1

Figure 2
 
Results:
A total of 94 ROIs from 14 paired cases (7 users) were technically evaluable. Median number of ROI alterations per case using the mouse was 111 (range 49-2226), compared to 451 (range 205-2345) for pen/tablet (p=0.32, n.s.). Median active contouring time was 58.4 minutes (range 31.7-140.1) for mouse compared to 41.6 minutes (range 22.5-135.4) for pen/tablet (p=0.02). All 7 users contoured the case faster with the pen/tablet, resulting in a mean time savings of 9.8 minutes (95% CI 2.1-17.5). The majority of users (5/7) would prefer to use the pen/tablet in the future; 1 user preferred the mouse, and one user indicated no preference. Estimated device cost-effectiveness could be achieved after 120 contouring sessions (estimated CI 71-529).
 
Discussion:
Target volumes and organs-at-risk for conformal radiotherapy treatment planning are necessarily defined by human users, introducing possible geometric/dosimetric variation. As the initial step in the planning process, the criticality of target delineation becomes evident, for even the most conformal radiotherapy plan, delivered with impeccable positional verification, is of little consequence if the initial prescription volumes within the treatment planning software do not accurately depict actual structures in 3D-space. Despite this, comparatively little data has to date been presented regarding strategic optimization of target delineation task itself.

The vast majority of target delineation is performed on computer software developed by radiotherapy manufacturers, and performed using standard personal computer hardware and peripheral devices. While these commercial units may have platform specific features with regard to imaging display, key functions or mouse strokes, almost invariably, treatment planning systems utilize a mouse-based user interface and screen-based display system. The typical computer keyboard-mouse-screen arose as an outgrowth of previous information systems (i.e., typewriting/word data entry), and was not initially designed for manipulation of complex visual datasets. Other visual-information data manipulation systems, such as flight simulators, videogames, and robotic surgery systems, have custom designed joysticks, manual controllers or other human interface systems, often with customized visual display systems (such as high-resolution or 3-D visualization).

Consequently, we evaluated the utility of a novel integrated pen-screen data input device, so as to ascertain the effect of the interface on target delineation efficiency. The standard mouse-keyboard-screen interface is optimized for word data entry, object selection (e.g., point and click), and linear motion recognition. However, it is reliant on comparatively gross hand and arm motor movements for motion detection. In comparison, pen-based human-computer interfaces, while poorly designed for word or data entry, are able rely upon the small motor motions of fingers and wrist, in addition to arm and hand, and can incorporate non-linear motion information tasks (e.g., drawing) more readily. Since, in our estimation, target delineation is more akin to drawing than data entry, we have chosen to compare a pen-based user interface: the Cintiq 2100 pen-based interactive display (Wacom Technology Corporation, Vancouver, WA). The Cintiq system consists of a pen based data entry device with a high-resolution tablet display (Figure 1). The system allows users to “draw” ROIs directly on DICOM images, unlike a mouse interface, which must indirectly translate and scale motion information.

Our current preliminary analysis, presented here, suggests that for complex cases (such as the head and neck case implemented in the current study), utilization of a pen-tablet interface is preferred by users, and increases efficiency as measured by ROI contouring time. Secondary analysis is currently underway to determine whether the same efficiency gains are observed for less complex anatomical contouring tasks (such as prostate or brain). Finally, the collected contours are being utilized a comparator set for a SIIM grant-supported software development to design a contour analysis software, which would then afford quantitative analysis of contours as part of a larger effort to develop training/accreditation software for target volume delineation.
 
Conclusion:
Early results suggest demonstrable time savings with pen/tablet interface when compared to mouse-based contouring for a complex head and neck cancer case, without substantially changing the number of contour alterations/modifications performed.
 
References:
Wilson LD, Flynn DF, Haffty BG. Radiation oncology career decision variables for graduating trainees seeking positions in 2003-2004. Int J Radiat Oncol Biol Phys. June 2005;62(2):519-25.