Implementation method in blind deconvolution based tumor segmentation using simulated PET images

PET imaging is increasingly used in determining functional tumor volumes for therapy response assessment and treatment planning. Accurate measurement of metabolically active tumors may improve optimal delivery of radiation treatment. However, volume measurement of small tumors (<;2 cm) has been a much investigated subject due to Partial Volume Effect and limited resolution of the system. This research combined two methods in improving accuracy of small tumor volume estimation: the tumor targeted blind deconvolution and the resampling method, and assessed it using the thresholding method on simulated PET images of tumors sized 10.5 mm and 14 mm at Tumor to Background Ratios (TBRs) of 4:1 and 10:1 reconstructed with (AW)-OSEM algorithm. Results showed that the local blind deconvolution method significantly decreased volume error rate of small tumors (<;2 cm) compared with the original non-deconvolved images. The segmented mean volumes and SDs were 7.13 cc ± 2.64 cc and 1.54 cc ± 0.51 cc for the tumors in original images and locally deconvolved images respectively. The best segmented volume was measured in locally deconvolved and resampled image for the tumor size of 14 mm at TBR of 10. The estimated volume error rate was found as 0.74% for this tumor. In addition, the computation time is significantly reduced ~ 13-fold using the local extraction method with resampling. Unlike previous studies, we combined two methods and optimized iterations of deconvolution to enhance resolution of small tumor images. Our results may facilitate improvements in detection of small tumors with different shapes and heterogenous activity distributions.