Tissue Classification for PET/MRI Attenuation Correction Using Conditional Random Field and Image Fusion

Photon attenuation correction is a challenging task in the emerging hybrid PET/MRI medical imaging techniques because of the missing link between tissue attenuation coefficient and MRI signal. MRI-based tissue classification methods for attenuation correction have difficulties caused by the significantly different abilities of photon absorption in tissues with similar MRI signal, such as bone and air. We proposed a novel method of integrating the information from MRI and PET emission data to increase the tissue classification accuracy. A classifier based on conditional random field was trained using features extracted from fused MRI and uncorrected PET images. The efficacy of the proposed method was validated quantitatively on synthetic datasets. It was found that the inclusion of PET data improved the classifier’s performance in terms of classification accuracy and PET image reconstruction quality.