Conformity analysis of target volumes for Margin-Intense Radiotherapy in Borderline-Resectable Pancreatic Cancer Daniel LP Holyoake (1,5), Maxwell Robinson (1,5), Derek Grose (2), David McIntosh (2), David Sebag-Montefiore (3,4), Ganesh Radhakrishna (4), Neel Patel (5), Somnath Mukherjee (1,5), Maria A Hawkins (1,5)
Introduction Margin-directed neoadjuvant radiotherapy for pancreatic cancer aims to improve rates of surgical resection with clear margins. The target volume encompasses adjacent and/or infiltrated vasculature but target volume definitions and in some cases lack reproducibility. SPARC is a CRUK-funded phase 1 study of pre-operative margin-intense stereotactic radiotherapy for patients with borderline-resectable pancreatic cancer and incorporates a comprehensive radiotherapy quality assurance protocol to ensure consistency in target definition and radiotherapy delivery.
Figure 3a. Median conformity indices for investigator target structures when compared to gold-standard. High values are desirable for the first three parameters while low values are desirable for the final two. Over-contouring tends to achieve low geographical miss scores at the penalty of high discordance index scores. 0.8
GTV_T
0.7
Median
0.6 0.5
DICE coefficient
0.4
Jaccard coefficient Van’t Riet number Geographical Miss Index Discordance index
0.3 0.2 0.1
Methods
0
On a test case of borderline-resectable pancreatic cancer, ‘gold-standard’ structures were defined by a team of two clinical oncologists and one radiologist. Target structures were then produced independently by six clinical oncologist investigators using a detailed manual and atlas: CTV_M, the target structure for the margin-directed boost, and GTV_T, the extent of visible tumour. Conformity analysis metrics were generated to compare investigator structures with the goldstandard: DICE coefficient =
2 × (𝐴∩𝐵) 𝐴+𝐵
Geographical miss index =
Jaccard coefficient =
𝐵−(𝐴∩𝐵) 𝐵
(𝐴∩𝐵) 𝐴∪𝐵
Discordance index = 1 −
0.51 0.30 0.32 0.10 0.64
IQR 0.400.60 0.240.42 0.230.39 0.080.23 0.450.74
CTV_M Median 0.68 0.40 0.48 0.32 0.39
IQR 0.600.75 0.320.48 0.360.57 0.160.39 0.190.44
GTV_T CTV_M
Figure 3b. Location of centre of mass of target structures in relation to gold-standard (indicated by red square marker)
(𝐴∩𝐵) 𝐴
(𝐴∩𝐵) (𝐴∩𝐵) ′ Van t Riet number = × 𝐴
𝐵
Figure 1. Gold standard (red/orange) and investigator contours of CTV_M (images a & c) and GTV T (image b) on contrast-enhanced axial CT
Results Gold-standard and median investigator volumes for GTV_T were 2.1cm3 and 5.35cm3 (IQR 4.13-6.68cm3) respectively, and for CTV_M were 1.1cm3 and 1.3cm3 (IQR 0.93-1.53cm3). The median distance between centre of mass of gold-standard and investigator volumes was 0.32cm (0.19-0.47cm) for GTV_T and 0.24cm (0.09-0.36cm) for CTV_M. Median DICE conformity coefficients for GTV_T and CTV_M were 0.51 (0.40-0.60) and 0.68 (0.60-0.75), median discordance indices (measurement of over-inclusive contouring) for GTV_T and CTV_M were 0.64 (0.54-0.74) and 0.39 (0.19-0.44).
Volume, cm3
Figure 2. Absolute volumes of investigator and gold standard structures
8 7 6 5 4 3 2 1 0
Structure volume [cm3]
GTV_T CTV_M Gold Standard
Gold standard Clinician 1 Clinician 2 Clinician 3 Clinician 4 Clinician 5 Clinician 6
GTV_T 2.1 6.4 5 4.7 7.5 5.7 2.4
CTV_M 1.1 1.2 1.6 1.4 1.5 1 0.7
Conclusions The investigator CTV_M structure volumes were more similar to the gold standard, showed less inter-observer variance in volume and somewhat less centre-of-mass deviation from the gold-standard compared with the investigator GTV_T structures. Investigator GTV_T volumes were larger than the gold standard, while this was not the case for the CTV_M, suggesting over-contouring of primary tumour, which was supported by inversely-related differences in discordance index and geographical miss index scores. Other conformity indices were similar for GTV and CTV. The method of margin-directed radiotherapy target definition appears consistently reproducible, but accurate delineation of pancreatic malignancies remains difficult and oncologists should have expert radiology support in this task.
1 CRUK/MRC Oxford Institute for Radiation Oncology, University of Oxford, Oxford, UK; 2 The Beatson West of Scotland Cancer Centre, Glasgow, UK; 3 The University of Leeds, CRUK Leeds Centre, St James’s University Hospital, Leeds, UK; 4 The St James’s Institute of Oncology, St James’s University Hospital, Leeds, UK; 5 The Churchill Hospital, Old Road, Headington, Oxford, UK. SPARC: SBRT pre-operatively for borderline resectable pancreatic cancer; UKCRN ID: 18496; ISRCTN: 14138956; Sponsor: The University of Oxford, UK. Funder: CRUK [grant number C43735/A18787]. Trial management: Oncology Clinical Trials Office (OCTO), University of Oxford. Daniel LP Holyoake is funded by a Nuffield Clinical Research Fellowship (
[email protected]) Maria A Hawkins is funded by the MRC (Medical Research Council) (grant MC_PC_12001/2