Bur J Nucl Med (1989) 15:712 718 EuropeanN u c l e a r Journal of Medicine © Springer-Verlag 1989 Online brain attenuation correction in PET: towards a fully automated data handling in a clinical environment* C. Michel, A. Bol, A.G. De Voider, and A.M. Goffinet Positron Tomography Laboratory, Universit6 Catholique de Louvain, Belgium Abstract. We have improved the calculation of the brain attenuation correction in positron emission tomography (PET) and set up a procedure which allows the clinician to get a fully corrected image in a single reconstruction step, without human intervention. By using a general object description scheme based on polygonal contour trees we are able to calculate the attenuation correction for brain tissue, bone and head holder. The head contour is generated from the emission sinogram. On a set of 15 adult patients, the emission values obtained using this calculated attenua- tion compare favorably with those obtained with an attenu- ation resulting from a transmission measurement. Residual discrepancies are attributed to incomplete scatter compen- sation between emission and transmission. The robustness of the algorithm has been tested on more than 100 brain fluorodeoxyglucose (18FDG) studies in adult patients, in- cluding pathological cases. Its applicability for ~8FDG stu- dies in children and for other tracers such as water (H2~50) and fluoroethylspiperone (18FESP) is also presented. Key words: Positron emission tomography Quantification - Attenuation calculation - Contour determination Positron Emission Tomography (PET), is aimed at deriving quantitative information from regional isotope concentra- tion within a section of the body. The activity data are converted into quantitative regional physiological or func- tional information using tracer kinetic models. The pre- cision of the procedure is dependent on the adequacy of the corrections applied to the emission scan prior to recon- struction: (i) random coincidence subtraction, (ii) normaliza- tion (i.e. geometrical solid angle and intrinsic detection effi- ciency correction for each line of response (LOR)), (iii) dead time correction, (iv) Compton scatter correction and (v) at- tenuation correction. Most of the commercially available tomographs provide accurate schemes for the first three corrections. Scatter cor- rection allows restoration of linearity and contrast in the activity distribution. Unfolding methods using exponential like kernels (space variant or invariant) are generally pro- posed both for emission scans (King et al. 1981; Bergstr6m * This article was presented at the 1st EEC workshop on accuracy determination in PET, January 19-20th. 1989 Pisa, Italy (COMAC- BME Concerted Project "Characterization and Standardization of PET Instrumentation") Offprint requests to: C. Michel, Positron Tomography Laboratory, 2 chemin du cyclotron, B-1348 Louvain-la-Neuve, Belgium et al. 1983) and transmission scans (Chan et al. 1986) and their implementation requires an experimental investigation in order to determine the best values for the kernel parame- ters. When the attenuation correction is calculated, a simple scatter correction uses experimentally determined effective attenuation coefficients. Two attenuation correction meth- ods are commonly used: a measurement by transmission using either an external ring or a rotating pin source (Deren- zo et al. 1981), and a calculation using either simple geomet- rical shapes or boundaries extracted from the analysis of short transmission images (Huang et al. 1981). The latter will be referred as the boundary method. For brain studies using one ring positron tomographs, the transmission measurement is too time consuming to be applied routinely in a clinical environment. Moreover, a slight motion of the patient's head can rarely be totally prevented (sometimes over a 2 h period) which hampers the use of this technique. On the other hand, the geometric model has several drawbacks: higher attenuation of the skull and attenuation of the head holder are neglected, and the elliptical model itself which is often inappropriate. Al- though the boundary method is necessary when dealing with complex or nonconvex objects, it has the drawbacks of the transmission method and requires additional process- ing for transmission image reconstruction and analysis. The purpose of this work is to describe and validate a method able to calculate accurate attenuation correction for objects described by polygonal contour trees. For brain studies, the p distribution is simple and its modelling only requires the determination of contours for head, skull, and head holder, provided their respective absorption coeffi- cients are known. Since the latter is visible in transmission, its contour can be determined using the boundary method. When the tracer uptake in the muscles surrounding the skull is high enough, the head shape may be determined by using an edge detection algorithm applied on the emis- sion scan. Assuming the skull thickness is known from other imaging modalities, the attenuation may be calculated ac- cording to the tree algorithm. This method is implemented on an ECAT III scanner (Hoffman et al. 1986) and is vali- dated on a set of 15 patients by comparing emission values obtained with the calculated attenuation to those obtained using the transmission method. Methods and material General attenuation calculation algorithm for complex objects When the finite detector width effect is neglected (Huang et al. 1979), the attenuation correction for a given LOR,