- Email: [email protected]
Defining Truth in Contouring: A Study of Automatic Atlas-based Lymph Node Segmentation (LNS) for Head and Neck Cancer (HNC)
Proceedings of the 51st Annual ASTRO Meeting
2462
Can HPV-related Tonsillar Squamous Cell Carcinoma (TSCC) be Treated with Radiation Alone?
1
K. Chu , B. Wehrli2, E. Wiebe1, K. Fung2, E. Winquist1, V. Venkatesan1, J. Yoo2, J. Franklin2, A. Hammond1, N. Read1 1
London Regional Cancer Program, London, ON, Canada, 2London Health Sciences Center, London, ON, Canada
TSCC is increasing in incidence. Chemoradiation (CRT) has improved outcomes, but at the expense of increased toxicity. Tobacco and alcohol are known risk factors, but recent recognition of the role of human papillomavirus (HPV) infection is changing the understanding of TSCC epidemiology. HPV-related TSCC (HPV+) is suggested to have improved outcomes and prognosis compared to non-HPV-related TSCC (HPV-). This suggests HPV+ TSCC may be treated less aggressively than HPV- TSCC. We have collected data since 1993 on TSCC patients seen at our center. We hypothesized that HPV+ TSCC patients might be effectively treated with radiation (RT) alone, and examined this in our historical cohort of patients. Purpose/Objective(s): To compare locoregional recurrence (LRR) rates, disease-free (DFS) and overall survival (OS) in HPV+ TSCC treated with RT or CRT. Materials/Methods: A single institution cohort of consecutive TSCC patients treated 1993 – 2006 inclusive was retrospectively analyzed. Data collected included treatment modality, pt demographics, and outcomes. Diagnostic paraffin embedded tissue samples were retrieved. Tumor overexpression of p16 protein using immunohistochemical (IHC) staining for p16 and in situ hybridization (ISH) for HPV DNA were used as surrogates to identify HPV-related tumors. Results: p16 IHC results were available for 48/135 TSCC patients (36%). Of these 48 patients, 87% of patients had Stage III/IV disease. 27 were treated with RT and 21 with CRT with a median follow-up of 31.7 months. 25/48 (52%) overexpressed p16 with similar proportions in RT-treated (13/27) and CRT-treated (12/21) patients. More HPV+ tumors were poorly differentiated (74% vs. 38%). Fewer patients with HPV+ tumors had been smokers (92% vs. 100%) or had a history of alcohol consumption (68% vs. 82%). Median DFS and OS were better for HPV+ tumors (46.1 months vs. 7.4 months, p = 0.001; 86.1 months vs. 15.8 months, p \ 0.0001; respectively). LRR rates in HPV+ patients were lower with CRT (28.4% vs. 49.7%, p 0.011). Median time to LRR has not been reached in HPV+ patients. DFS and OS in HPV+ patients were better with CRT compared to RT (63.6% vs. 33.6%, p = 0.001; 81.8% vs. 26.0% p \ 0.0001; respectively). Conclusions: These data confirm a better prognosis in patients with HPV+ TSCC treated with RT compared to HPV- patients. A benefit of CRT compared to RT alone was seen in patients with HPV+ TSCC. As p16 overexpression has a high sensitivity and specificity for detection of HPV DNA by ISH, the results suggest that despite improved outcomes in HPV+ tumors, chemotherapy cannot be excluded from treatment. These preliminary data support maintaining current standards of practice for TSCC regardless of HPV status, and further study of HPV+ TSCC in controlled clinical trials. Multivariable analysis, acquisition of additional tumor samples, and HPV status assessment using ISH is currently ongoing to validate our results and will be presented. Author Disclosure: K. Chu, None; B. Wehrli, None; E. Wiebe, None; K. Fung, None; E. Winquist, None; V. Venkatesan, None; J. Yoo, None; J. Franklin, None; A. Hammond, None; N. Read, None.
2463
Defining Truth in Contouring: A Study of Automatic Atlas-based Lymph Node Segmentation (LNS) for Head and Neck Cancer (HNC)
L. J. Stapleford1, J. D. Lawson2, C. Perkins1, S. Edelman1, L. W. Davis1, M. W. McDonald1, A. Waller1, E. Schreibmann1, T. Fox1 Emory University, Atlanta, GA, 2University of California, San Diego, San Diego, CA
1
Purpose/Objective(s): LNS for HNC intensity modulated radiation therapy (IMRT) is time-consuming with significant interobserver contour variability. The aim of this study was to evaluate if automatic atlas-based LNS could improve efficiency and decrease inter-observer variability while maintaining accuracy. Materials/Methods: Five physicians (MDs) with HNC IMRT expertise used CT data from 5 HNC patients with non-bulky nodes to create bilateral neck clinical target volumes (CTVs) covering specified LN levels, using RTOG consensus guidelines for reference. A 2nd contour set was generated using the Velocity AI HNC atlas (Velocity Medical Solutions). MDs modified the atlas-generated contours to make them suitable for treatment planning. As all 5 MDs created manual and atlas-modified (AM) contours, the result for each CTV was a set of 1 atlas, 5 manual, and 5 AM contours. To assess contour variability, the Simultaneous Truth and Performance Level Estimation (STAPLE) algorithm was used to take the collections of contours and calculate a probabilistic estimate of the ‘‘true’’ segmentation. For each CTV, the two sets of MD contours (5 manual, 5 AM) were each used to create a STAPLE volume, STAPLE-manual and STAPLE-AM, respectively. Differences between the manual, atlas, and AM contours were analyzed using the following metrics: volume, sensitivity, false positivity (%FP), Dice similarity coefficient (DSC), and mean/max surface disagreement. Accuracy was assessed by comparing manual and atlas contours to the STAPLE-manual to determine which set most consistently approximated the ‘‘true’’ manual contour. Variability was assessed by measuring the homogeneity within manual and AM contour sets in regards to volume and degree of overlap with their respective STAPLE ‘‘truths.’’ Results: Compared to the ‘‘true’’ manual contours (STAPLE-manual), the atlas contours had a high degree of accuracy, with sensitivity, DSC, and mean/max surface disagreement values comparable to the average manual contour (86%, 76%, 3.3/ 17.4 mm atlas vs. 73%, 79%, 2.8/17 mm manual). The AM group was more consistent than the manual group for multiple metrics, with the most notable gains being a reduction in the range of contour volume (106–430cc manual vs. 176–347cc AM) and %FP (1–37% manual vs.1–7% AM). Average contouring timesavings with the atlas was 11.5 minutes per patient (32% reduction). Conclusions: Using the STAPLE algorithm to generate ‘‘true’’ contours from multiple expert manual and AM contours, we have shown that compared to manual segmentation, atlas-based automatic LNS for HNC is accurate, efficient and reduces inter-observer variability. Further analysis of IMRT contour variability may improve consistency across the field and augment the education process for MDs learning IMRT. Author Disclosure: L.J. Stapleford, None; J.D. Lawson, None; C. Perkins, None; S. Edelman, None; L.W. Davis, None; M.W. McDonald, None; A. Waller, None; E. Schreibmann, None; T. Fox, Velocity Medical Solutions, E. Ownership Interest; Velocity Medical Solutions, F. Consultant/Advisory Board.
S389
2462
Can HPV-related Tonsillar Squamous Cell Carcinoma (TSCC) be Treated with Radiation Alone?
1
K. Chu , B. Wehrli2, E. Wiebe1, K. Fung2, E. Winquist1, V. Venkatesan1, J. Yoo2, J. Franklin2, A. Hammond1, N. Read1 1
London Regional Cancer Program, London, ON, Canada, 2London Health Sciences Center, London, ON, Canada
TSCC is increasing in incidence. Chemoradiation (CRT) has improved outcomes, but at the expense of increased toxicity. Tobacco and alcohol are known risk factors, but recent recognition of the role of human papillomavirus (HPV) infection is changing the understanding of TSCC epidemiology. HPV-related TSCC (HPV+) is suggested to have improved outcomes and prognosis compared to non-HPV-related TSCC (HPV-). This suggests HPV+ TSCC may be treated less aggressively than HPV- TSCC. We have collected data since 1993 on TSCC patients seen at our center. We hypothesized that HPV+ TSCC patients might be effectively treated with radiation (RT) alone, and examined this in our historical cohort of patients. Purpose/Objective(s): To compare locoregional recurrence (LRR) rates, disease-free (DFS) and overall survival (OS) in HPV+ TSCC treated with RT or CRT. Materials/Methods: A single institution cohort of consecutive TSCC patients treated 1993 – 2006 inclusive was retrospectively analyzed. Data collected included treatment modality, pt demographics, and outcomes. Diagnostic paraffin embedded tissue samples were retrieved. Tumor overexpression of p16 protein using immunohistochemical (IHC) staining for p16 and in situ hybridization (ISH) for HPV DNA were used as surrogates to identify HPV-related tumors. Results: p16 IHC results were available for 48/135 TSCC patients (36%). Of these 48 patients, 87% of patients had Stage III/IV disease. 27 were treated with RT and 21 with CRT with a median follow-up of 31.7 months. 25/48 (52%) overexpressed p16 with similar proportions in RT-treated (13/27) and CRT-treated (12/21) patients. More HPV+ tumors were poorly differentiated (74% vs. 38%). Fewer patients with HPV+ tumors had been smokers (92% vs. 100%) or had a history of alcohol consumption (68% vs. 82%). Median DFS and OS were better for HPV+ tumors (46.1 months vs. 7.4 months, p = 0.001; 86.1 months vs. 15.8 months, p \ 0.0001; respectively). LRR rates in HPV+ patients were lower with CRT (28.4% vs. 49.7%, p 0.011). Median time to LRR has not been reached in HPV+ patients. DFS and OS in HPV+ patients were better with CRT compared to RT (63.6% vs. 33.6%, p = 0.001; 81.8% vs. 26.0% p \ 0.0001; respectively). Conclusions: These data confirm a better prognosis in patients with HPV+ TSCC treated with RT compared to HPV- patients. A benefit of CRT compared to RT alone was seen in patients with HPV+ TSCC. As p16 overexpression has a high sensitivity and specificity for detection of HPV DNA by ISH, the results suggest that despite improved outcomes in HPV+ tumors, chemotherapy cannot be excluded from treatment. These preliminary data support maintaining current standards of practice for TSCC regardless of HPV status, and further study of HPV+ TSCC in controlled clinical trials. Multivariable analysis, acquisition of additional tumor samples, and HPV status assessment using ISH is currently ongoing to validate our results and will be presented. Author Disclosure: K. Chu, None; B. Wehrli, None; E. Wiebe, None; K. Fung, None; E. Winquist, None; V. Venkatesan, None; J. Yoo, None; J. Franklin, None; A. Hammond, None; N. Read, None.
2463
Defining Truth in Contouring: A Study of Automatic Atlas-based Lymph Node Segmentation (LNS) for Head and Neck Cancer (HNC)
L. J. Stapleford1, J. D. Lawson2, C. Perkins1, S. Edelman1, L. W. Davis1, M. W. McDonald1, A. Waller1, E. Schreibmann1, T. Fox1 Emory University, Atlanta, GA, 2University of California, San Diego, San Diego, CA
1
Purpose/Objective(s): LNS for HNC intensity modulated radiation therapy (IMRT) is time-consuming with significant interobserver contour variability. The aim of this study was to evaluate if automatic atlas-based LNS could improve efficiency and decrease inter-observer variability while maintaining accuracy. Materials/Methods: Five physicians (MDs) with HNC IMRT expertise used CT data from 5 HNC patients with non-bulky nodes to create bilateral neck clinical target volumes (CTVs) covering specified LN levels, using RTOG consensus guidelines for reference. A 2nd contour set was generated using the Velocity AI HNC atlas (Velocity Medical Solutions). MDs modified the atlas-generated contours to make them suitable for treatment planning. As all 5 MDs created manual and atlas-modified (AM) contours, the result for each CTV was a set of 1 atlas, 5 manual, and 5 AM contours. To assess contour variability, the Simultaneous Truth and Performance Level Estimation (STAPLE) algorithm was used to take the collections of contours and calculate a probabilistic estimate of the ‘‘true’’ segmentation. For each CTV, the two sets of MD contours (5 manual, 5 AM) were each used to create a STAPLE volume, STAPLE-manual and STAPLE-AM, respectively. Differences between the manual, atlas, and AM contours were analyzed using the following metrics: volume, sensitivity, false positivity (%FP), Dice similarity coefficient (DSC), and mean/max surface disagreement. Accuracy was assessed by comparing manual and atlas contours to the STAPLE-manual to determine which set most consistently approximated the ‘‘true’’ manual contour. Variability was assessed by measuring the homogeneity within manual and AM contour sets in regards to volume and degree of overlap with their respective STAPLE ‘‘truths.’’ Results: Compared to the ‘‘true’’ manual contours (STAPLE-manual), the atlas contours had a high degree of accuracy, with sensitivity, DSC, and mean/max surface disagreement values comparable to the average manual contour (86%, 76%, 3.3/ 17.4 mm atlas vs. 73%, 79%, 2.8/17 mm manual). The AM group was more consistent than the manual group for multiple metrics, with the most notable gains being a reduction in the range of contour volume (106–430cc manual vs. 176–347cc AM) and %FP (1–37% manual vs.1–7% AM). Average contouring timesavings with the atlas was 11.5 minutes per patient (32% reduction). Conclusions: Using the STAPLE algorithm to generate ‘‘true’’ contours from multiple expert manual and AM contours, we have shown that compared to manual segmentation, atlas-based automatic LNS for HNC is accurate, efficient and reduces inter-observer variability. Further analysis of IMRT contour variability may improve consistency across the field and augment the education process for MDs learning IMRT. Author Disclosure: L.J. Stapleford, None; J.D. Lawson, None; C. Perkins, None; S. Edelman, None; L.W. Davis, None; M.W. McDonald, None; A. Waller, None; E. Schreibmann, None; T. Fox, Velocity Medical Solutions, E. Ownership Interest; Velocity Medical Solutions, F. Consultant/Advisory Board.
S389