PT  - JOURNAL ARTICLE
AU  - Willem Grootjans
AU  - Edwin Usmanij
AU  - Wim Oyen
AU  - Erik van der Heijden
AU  - Eric Visser
AU  - Dimitris Visvikis
AU  - Mathieu Hatt
AU  - Johan Bussink
AU  - Lioe-Fee De Geus-Oei
TI  - Performance of semi-automatic image segmentation algorithms for measurement of early treatment response using 18F-PET/CT in patients with locally advanced NSCLC during concomitant chemo-radiotherapy
DP  - 2016 May 01
TA  - Journal of Nuclear Medicine
PG  - 1467--1467
VI  - 57
IP  - supplement 2
4099  - http://jnm.snmjournals.org/content/57/supplement_2/1467.short
4100  - http://jnm.snmjournals.org/content/57/supplement_2/1467.full
SO  - J Nucl Med2016 May 01; 57
AB  - 1467Objectives Automatic lesion segmentation is an important step towards standardizing early response measurements using 18F-fluorodeoxyglucose (FDG) positron emission tomography (PET) in combination with X-ray computed tomography (CT). In this study, the performance of different automatic segmentation algorithms for early response monitoring in non-small cell lung cancer (NSCLC) patients during chemo-radiotherapy was evaluated.Methods Twenty-seven patients with locally advanced NSCLC treated with concomitant chemo-radiotherapy underwent combined FDG-PET/CT imaging before and in the second week of treatment. Several image segmentation methods were used including fixed threshold segmentation at(i)40% SUVmax (T40),(ii) 50% SUVmax (T50), (iii) relative threshold level (RTL), (iv) signal to background ratio (SBR), and (v) fuzzy locally adaptive Bayesian (FLAB) segmentation. The predictive value of different PET parameters, including metabolic tumour volume (MTV), mean standardized uptake value (SUVmean), and total lesion glycolysis (TLG) of the primary tumor and lymph nodes for progression free survival (PFS) and overall survival (OS) was determined using a Cox proportional hazards model.Results The MTVs for primary tumour showed a high degree of conformity regarding shape and spatial overlap on both pre- and in-treatment FDG PET images. However, due to limited size and contrast, there were considerably more segmentation failures when performing lymph node segmentation. Of all algorithms, SBR and FLAB had the fewest lymph node segmentation failures for both pre- and in-treatment FDG PET images. The importance of adequate lymph node segmentation was reflected by a stronger association of PFS and OS when lymph nodes are included in the TLG calculation, yielding a summed TLG. Pre-treatment TLG was significantly associated with PFS and OS, regardless of the segmentation algorithm used. This was also the case for relative TLG decrease (ΔTLG) between the pre- and in-treatment TLG, with the exception of the RTL and T40 method when considering OS.Conclusions Pre-treatment TLG was significantly associated with PFS and OS in NSCLC patients treated with concomitant chemo-radiotherapy, regardless of the segmentation algorithm used and despite the limited number of patients. This is also the case for ΔTLG, with the exception of the RTL and T40 method. Significance levels increased when lymph node TLG was included in the analysis emphasizing the importance of adequate lymph node segmentation in PET images. View this table:Association of summed pre-treatment and ΔTLG, obtained with different algorithms, with OS