Fuzzy inference systems for segmented attenuation correction in positron emission tomography

Segmented attenuation maps can be either calculated from transmission data or with a limited accuracy from emission data. The authors developed two new methods based on adaptive fuzzy inference systems for the transmission segmentation (TSF, Transmission Segmentation by Fuzzy Inference) and for the segmentation of FDG emission images (ESF, Emission Segmentation by Fuzzy Inference). TSF: The main problem, which is the adaptation of the membership functions to the measured data, was solved by adjusting the system with the histogram fitting approach used in the authors´ routine segmentation. The fuzzy system allows for a scan time reduction down to 40 s-50 s per bed position and is limited by a local bias in the center of the attenuation map. This system was additionally combined with local thresholding operators based on a fuzzy edge detection leading to accurate results even at 30 s. ESF: In uncorrected FDG emission images the lungs appear with a positive intensity compared to the background, which is almost zero. Based on this effect, ESF allows for an automatic segmentation of 18-FDG emission scans with an error of 10-15% in image quantification. Both methods were evaluated with the same phantom and simulated data, and compared with the authors´ standard (Histogram Fitting Segmentation)