Lesion size quantification in SPECT using learning vector quantizer

An artificial neural network has been developed to determine the size of cold lesions detected in single photon emission computed tomography (SPECT) images. The neural network is Kohonen´s learning vector quantizer (LVQI) and is trained to discriminate cold lesions of eight different sizes ranging from one to eight pixels in radius. The images generated for the study are simulated 64×64 SPECT images of a slice of an active cylinder with a nonactive cylindrical lesion. The authors present the proposed network and experimental results for two noise levels (50000 and 100000 counts/slice), showing the capability of such a network to determine quite accurately the size of detected abnormalities in SPECT images