MRI vs CBCT image guidance when treating lymph nodes in patients with locally advanced (LA)-NSCLC

Abstract

Purpose or Objective CBCT is routinely used to guide curative radiotherapy (RT) for lung cancer. This is limited by poor soft tissue discrimination. Consequently anatomical surrogates for lymph nodes (LNs) are used to facilitate registration and generous PTV margins are needed. MRI can improve soft tissue discrimination, and may allow a more accurate setup. The aim of this study was to compare MRI to CBCT for the identification and localisation of LNs in patients undergoing curative RT for LA-NSCLC. Material and Methods Four experienced RTTs rigidly registered the outlined LNs on planning CT (pCT) with the CBCT and MR images of 10 patients with LA-NSCLC. MR and CBCT images were acquired 2 weeks into RT. The MR sequences used were: T2 turbo spin echo (TSE) non-fat saturated (Seq1) and T1 DIXON water (Seq2). Observers were asked to localise each LN on the study image and register this to the LN contour on the pCT (study images had been offset in advance by 5cm in the X, Y and Z planes). The resulting X, Y, Z translations were recorded. Observers then rated their confidence in performing the registration and the quality of the image in permitting visualisation of the LN, with a view to contouring (Fig 1). The process was repeated for each individual LN using all study images. For each LN, translations from all observers were reviewed for each plane and the standard deviation (σ) was calculated. The σ from all planes were used to calculate the vector value (σV). The mean σV for each image modality was compared to assess inter-observer variation. Results 24 LNs distributed between hilar (station 10-11) and mediastinal regions (1-7) were assessed. Fig 2 shows the distribution of individual LN σV values for each imaging modality. The mean σV for CBCT was not significantly different than either MR Seq1 (0.22 vs 0.27, p= 0.09) or Seq2 (0.22 vs 0.20, p= 0.47). However the mean σV for Seq 1 was significantly greater than Seq2 (0.27 vs 0.20, p= 0.04). Compared to CBCT the mean confidence scores for Seq1 (2.5 vs 1.8, p<0.001) and Seq2 (2.7 vs 1.8, p<0.001) were both significantly better. Seq2 had a significantly greater mean score than Seq1 (p=0.041). Regarding image quality, compared to CBCT, both Seq1 (2.66 vs 1.75, p<0.001) and Seq2 (2.74 vs 1.75, p<0.001) had significantly greater scores. No difference was seen between Seq1 and Seq2 (p=0.5). Conclusion Observers scored the image quality of MR higher than CBCT and had significantly greater confidence using MR to perform direct LN registrations. T1 DIXON outperformed the T2 TSE MR sequence in providing more accurate registrations. The T2 TSE shows fat and LNs as bright contrast. This might make it difficult to differentiate the edges of mediastinal LNs, whereas the T1 DIXON provides a fat suppressed image which may aid LN identification in this scenario. However no difference was seen between MR and CBCT.. This work suggests MR may be useful in increasing the accuracy for set-up correction during RT for LA NSCLC. Further evaluation of its potential role is needed.