Improving PET/CT imaging by Fourier-wavelet restoration

Knešaurek K., Machac J.

Division of Nuclear Medicine, Box 1141, The Mount Sinai Medical Center, New York, NY 10029, USA

Recently, positron emission tomography (PET) has become a very well recognized diagnostic imaging method in oncology, cardiac imaging and neuroimaging. Putting together computed tomography (CT) with PET in combined PET/CT systems enables us, in one procedure, to obtain both functional PET images and anatomical CT images in almost perfect coregistration. However, PET images, although extremely useful in diagnosis, treatment planning and in defining prognosis, still suffer from a high level of noise and relatively limited resolution due to the random coincidences, scatter and attenuation of 511 keV photons. To address these limitations, the restoration approach has been successfully used in nuclear medicine imaging for various applications. However, the restoration process is an ill-conditioned process, which has a tendency to significantly increase the amount of noise in the restored images. To address this problem a, new hybrid Fourier-wavelet deconvolution restoration technique has been developed. Here, the Fourier-wavelet deconvolution restoration technique using the Butterworth filter in the Fourier domain and Daubechies wavelet functions, order 2 was implemented. Wavelet noise suppression was applied by a hard threshold.

We have applied our hybrid Fourier-wavelet restoration technique in the whole-body lung 2D PET oncology imaging and in the 3D brain PET imaging, mostly for Alzheimer’s disease cases. The approach, with slight modification of the filters, can also be applied in PET imaging of other organs such as cardiac perfusion imaging. The results of our studies show an average 25% increase in contrast in lung lesions, with a 2% increase in noise in the restored images. In brain imaging the results are very similar. The typical increase in contrast in brain PET imaging was 19%, while the restored images had practically the same amount of noise (± 3%) as the original images. Our results show that the quality and quantification of the 3D brain and 2D lungs PET images can be significantly improved by Fourier-wavelet (FTW) restoration filtering.