Abstract:

In order to avoid Xradiation harm that PET/CT does to patients, and to achieve better PET/MRI attenuation correction, we divide MRI into different tissues,such as air liquid, soft tissues and bones under the guidance of tissue segmentation method  by using fuzzy clustering algorithm. Then different organizations are given different linear attenuation coefficients so as to achieve better PET attenuation correction. The proposed method has three advantages: 1) benefiting from the guidance of historical knowledge, it tends to be effective in the situations when the data is insufficient or distorted by much noise; 2) the simple sampling strategy based on transfer learning greatly shortens the overall time of clustering, and at the same time ensures the robustness of the algorithm, thus suitable for medical MRI fast clustering; 3) as the historical MRI knowledge does not expose the raw data of the source domain, this algorithm is capable of protecting privacy of the source domain, and meets the basic requirements of medical diagnosis. Experimental studies on realworld datasets demonstrate these merits of our work.

Key words: PET/CT;Xray radiation;PET/MRI;transfer learning based fuzzy clustering;attenuation coefficient;historical data