Metallic implants, such as hip implants, are known to induce streaking artifacts in CT images which can cause over/underestimation of the activity uptake in CT-based attenuation corrected PET images. Hence, metal artifact reduction (MAR) of CT images is essential in order to obtain accurate quantification of PET data. The proposed MAR technique replaces the projection bins of the virtual sinogram affected by metallic implants using a 2D cubic interpolation scheme. Since removing the affected projection bins renders the sinogram grid irregular, a Delaunay triangulated gridding together with Clough-Tocher cubic interpolation, which is compatible with this irregular grid, is used to substitute the values of the affected bins. A cylindrical phantom filled with uniform activity concentration incorporating metallic inserts and 30 clinical PET-CT studies containing hip prostheses were used to assess the performance of the proposed approach. The resulting images were compared to those obtained using the built-in MAR algorithm on a Siemens mCT64 PET/CT scanner. Phantom and clinical studies showed that the proposed algorithm performed considerably better than Siemens's method in the regions corresponding to dark streaking artifacts (underestimated regions), whereas it performed equally well compared to Siemens's method in the other regions. In the underestimated regions, the proposed method increased the uptake value up to 45%, whereas the Siemens's method kept almost the same uptake as the uncorrected PET images. In the overestimated regions both methods decreased the uptake by similar to 45%. The phantom experiment also revealed that the proposed approach is in better agreement with the actual activity concentration compared to both the uncorrected and corrected images using Siemens's method. It can be concluded that the proposed method allows more accurate attenuation correction of PET data thus preventing misinterpretation of activity uptake in regions adjacent to metallic objects.