Abstract:
This dissertation presents an automatic method to trace the boundary of the tumor in PET image. A double-stage threshold that locates the local minimum between the Otsu’s threshold and the pixel with maximum intensity gray level within the image is proposed. The gray level in accordance with this scheme is chosen and embedded into the external force of a region-based active contour so that both algorithms are performed consecutively. The automated tumor contouring method is validated using the IEC/2001 torso phantom with six hot spheres (0.52-26.53 cc) insert and the variation of the source-to-background ratio (SBR). The results show that the tumor volumes segmented by automated algorithm are at higher accuracy than the traditional active contour. The accuracy of the detected volume is reduced in small sphere with low SBR. The least volume mismatch was at SBR 16 in the largest sphere (3.7 cm diameter) of 1.51 %. The average volume mismatch between the automated and manual method is -4.06 ± 6.35 % in clinically implemented with 10 esophageal cancer patients underwent whole-body 18F-FDG PET/CT imaging. The advantage of the study is not only to improve the precision and accuracy of PET tumor contouring, but also to use by radiation oncologist for radiation therapy planning. Furthermore, this method can contribute to clinical PET image analysis.
