Accurate attenuation correction is a prerequisite for the determination of precise regional radioactivity concentrations in positron tomography. Attenuation correction can be performed using an external source of radiation and two measurements: a blank scan performed with no subject in the tomograph, and a transmission scan performed with the subject in the field of view. The ratio of blank to transmission counts gives the appropriate attenuation correction factor for each line of response. In theory, this provides a perfect correction for photon attenuation, but in practice the technique is limited by noise due to limited counting statistics and scattered radiation in the measured transmission data. In the present work, ¹³⁷Cs is proposed as a suitable radiation source for transmission measurements in 'singles' mode, a technique that substantially increases the statistical accuracy of the transmission data. ¹³⁷Cs can be used without any recalibration of the tomograph, and the spatial resolution is comparable to that obtained using ⁶⁸Ge. Since ¹³⁷Cs emits a monoenergetic gamma ray at 662 keV, and emission data are acquired by detecting annihilation photons of energy 511 keV, a simple extrapolation method is developed to extrapolate the attenuation coefficients measured at 662 keV to 511 keV. To eliminate scatter contamination in the transmission data, a dual-energy-window scatter correction technique is developed whereby correction can be made on-the-fly during data acquisition. Using the developed extrapolation method and dual energy scatter correction method, the linear attenuation coefficients measured in 'singles' mode using ¹³⁷Cs agree well with the expected values. To achieve further suppression of noise in the transmission data, a segmented attenuation correction technique is also developed in this work. The technique uses artificial neural networks for processing the count-limited transmission data. The technique has been validated in phantoms and verified in human studies. The results indicate that attenuation coefficients measured in the segmented transmission images are accurate and reproducible. Activity concentrations measured in the reconstructed emission image can also be recovered accurately with this technique. The accuracy of the technique is subject independent and insensitive to scatter contamination in the transmission data. It can predict accurately the value of the attenuation coefficient for any material in the range from air to water. Satisfactory results are obtained if the transmission data contains as few as 400,000 true counts per plane. Thus, accurate attenuation data can be obtained by acquiring a short transmission scan using the 'singles' method, and then processing these data using the artificial neural network technique.

Doctor of Philosophy (PhD)