SPECT scatter correction in non-homogeneous media

Abstract

Single photon emission computed tomography (SPECT) has the potential for quantitation of absolute activity concentration in vivo. The accuracy of activity estimates depends to a large extent on the accuracy of the attenuation and scatter corrections performed. This is particularly so in the thorax, where assumptions about regular object shape and constant density are inappropriate. We have developed a method of scatter correction based on convolution subtraction (CS) which takes account of variable tissue density. Rather than assuming the scatter fraction to be constant, the scatter fraction is determined at each point in the image based on measured photon transmission through the object. For comparison, a modified lower window subtraction (LWS) technique has also been developed, involving convolution of lower window data with a theoretically derived kernel, which more accurately relates lower window scatter to photopeak scatter compared with conventional LWS.

Both methods have been assessed in a phantom study using a non-uniform medium and distributed activity. The methods described were compared with the two conventional methods on which they are based. The accuracy for determining activity concentration was calculated for a 5cm diameter "hot" cylinder, the "warm" background and a 10cm diameter "cold" air cylinder. Quantitative accuracy of >95% was achieved in the "hot" cylinder and in the "warm" background for both the TDCS and MLWS methods, whilst all 4 methods yielded no significant reconstructed counts in the region of the "cold" air cylinder, indicating very good cold contrast.