To date, we have prospectively collected >450 distinct target volume structures (ROIs) in 5 distinct anatomic sites (head and neck, lung, brain, and prostate), representing several hundred cumulative man-hours on the part of >30 radiation oncologist participants from a variety of institutions. With help from our collaborative research team (comprising efforts from The University of Texas Health Science Center at San Antonio, Netherlands Cancer Institute, and Oregon Health & Science University), we have completed a preliminary analysis on a pilot dataset investigating rectal cancer contouring in a cooperative group setting (Figure 1, showing contours of 15 users overlaid on the standardized rectal cancer case DICOM file). While at present our analytic technique utilizes a comparatively mathematical overlap modeling, we plan to include more eloquent quantitative statistical measures in future iterations.

In order to maximize ease of utilization, we have opted to build the TaCTICS GUI as a plug-in module for an extant image viewing/manipulation software platform (Osirix, Rosset et al., v3.3.2). Since Osirix is a widely used software which already has ROI manipulation and DICOM RTSTRUCT>ROI conversion (Figure 2, showing collected prostate data from collected users), we hope that current ROI analysis capability may be modified to allow volumetric comparison in such a manner as to readily provide statistically and clinically meaningful feedback to users regarding target volume delineation, while simultaneously allowing creation of composite volumes (e.g., standard deviation isosurface, median isosurface, and encompassing volume). Finally, we hope that with our pilot datasets we may afford report generation of volumetric probabilistic scoring of target volume ROIs in reference to expert users, allowing novice radiation oncologists to identify specific anatomic regions where contour variation from an expert-derived standard is most pronounced (Figure 3, pilot data showing regional surface deviation of rectal cancer contours from a reference user).

Several series have also established that user variability in target volume delineation may result in potentially significant dosimetric differentials between prescribing radiation oncologists. Additionally, collected data suggest that clinical trial data may be obfuscated by user-dependent differentials in prescription volume determination. Despite this great interest, comparatively little data to date has been presented regarding strategic optimization of target delineation itself, either as a function of standardized practices, or as a function of deliberate educational curriculum. However, such steps are hampered, at present, as there remains no efficient widely available software mechanism to evaluate target volumes quantitatively.

Using previously collected pilot data, we have determined that there is substantial inter-observer variability in terms of target volume reproducibility (Figure 4). We hope that by constructing a GUI that allows users to analyze target volume ROIs and gain meaningful “scores” regarding their performance over time, exposure to training software might become a viable tool for longitudinal examination and improvement of trainee performance.

This work is supported by the SIIM Small Grant for Product Support Development.