Technical noteTri-dimensional automatic segmentation of PET volumes based on measured source-to-background ratios: influence of reconstruction algorithms
Introduction
The use of multi-modality imaging for the delineation of target volumes in Head and Neck radiotherapy planning is a promising concept as it allows for instance to combine complementary anatomical and functional images.
Determination of a volume from functional imaging is a critical step, which is usually performed by thresholding, i.e. by segmenting a lesion of interest on the basis of a given level of radioactivity. Automatic thresholding method has been developed for SPECT studies and further extended and validated for fluorodeoxyglucose-positron emission tomography (FDG-PET) in patients with lung tumors [5], [6]. Studies on phantom were conducted to derive the relationship between the true volume and the threshold to be applied on the PET images. Such threshold varied according to the signal-to-background (S/B) ratios, being typically higher for less contrasted images. Such method, however, suffers from serious shortcomings, at least when functional imaging is to be used for radiotherapy treatment planning purpose. First, such automatic segmentation requires an a priori estimate of the volume of the lesion of interest from computed tomography (CT) slices or other anatomical images. In absence of an a priori knowledge of the volume, it has been shown that a fixed threshold value between 36 and 44% of the maximal activity predicts well the true volume, but only for lesions larger than 4 ml. Second, for lesions smaller than 4 ml as usually observed in head and neck tumors, or for poorly contrasted lesions (S/B below 5.5), such thresholding method could lead to substantial overestimate of the true volume, even with the knowledge of the a priori volume. As the main challenge of introducing functional imaging in radiotherapy treatment planning is to potentially resize the target volume or refine its shape relative to the anatomical information, the use of a CT-scan volume as a gold standard from which a given threshold of activity will be selected, appears as a circular reasoning.
Section snippets
Phantom object
A spherical lucite phantom (Canberra-Packard®, Zellik, Belgium) filled with six spheres ranging from 0.55 to 17.15 ml was used. Each sphere was filled with 74–111 kBq (2–3 μCi) of 18F. The phantom was filled first with ‘cold’ water, then with increasing activity of 18F to get signal-to-background ratios of 8.7, 6.5, 4.7, 3.0, 2.1, and 1.5. These values are within the range observed in head and neck tumors.
PET acquisition
PET scan acquisitions were performed in a three-dimensional mode on a Siemens Exact HR
Results
The relationship between the best threshold of activity to accurately measure the true volume of the spheres and the measured S/B ratio were determined for the four reconstruction algorithms (Fig. 1). As explained in Section 2, the smallest sphere of 0.55 ml (10.2 mm of diameter) was excluded from the plots to avoid any partial effect in the object reconstruction. The plots were fitted using an inverse function (y=a+b×1/x), which provided the best regression parameters. These relationships
Discussion
The key feature of the method described in this study was that the volumetric assessment did not require any a priori knowledge of the volume of the object of interest, as opposed to previous works in the field [5]. Our method only required the measurement of the source-to-background ratio. This is an advantage not only because volumetric assessment can be performed in absence of anatomical information of the object of interest, but also because it does not link the two imaging modalities, thus
Acknowledgements
The authors wish to thank Prof. C. Michel for assistance during the PET acquisition. Jean-François Daisne was a recipient of research grants from the Fonds de la Recherche Scientifique Medicale (Télévie # 7.4563.99 and 7.4529.01). The project was supported by a grant from the ‘Fonds Special de Recherche’ of the Université Catholique de Louvain and by the ‘Loterie Nationale’ from Belgium.
References (9)
- et al.
Positron emission tomography partial volume correction: estimation and algorithms
J Cereb Blood Flow Metab
(2002) - et al.
Performance evaluation of microPET: a high-resolution lutetium oxyorthosilicate PET scanner for animal imaging
J Nucl Med
(1999) - et al.
Impact of computed tomography (CT), magnetic resonance (MR) and positron emission tomography (PET) with fluorodeoxyglucose (FDG-PET) image coregistration on GTV delineation in head and neck squamous cell carcinoma (HNSCC)
Radiother Oncol
(2002) - et al.
Design and evaluation of an LSO PET detector for breast cancer imaging
Med Phys
(2000)