Deformable Model-based PET Segmentation for Heterogeneous Tumor Volume Delineation

PET-guided radiation therapy treatment planning, clinical diagnosis, assessment of tumor growth and therapy response are dependent on the accurate delineation of the tumor volume. Several PET segmentation techniques have been proposed in the recent years. Most these techniques fail in the presence of heterogeneity in the lesion. In this work, an active contour model based on the work presented by Chan and Vese (2001) is proposed to handle the heterogeneity of the lesion uptake. In the proposed method, the fitting terms in the Chan-Vese formulation are modified by introducing extra input images, including the smoothed version of the original image, using anisotropic diffusion filtering (ADF) and a trous wavelet transform of the image to handle the heterogeneity of the lesion uptake and avoid getting stuck in local minima. The advantage of utilizing ADF for image smoothing is that it avoids blurring the object's edges and preserves the average activity within a region, which is an important property for accurate PET quantification. The a trous wavelet transform is utilized due to its easy implementation and better performance at high noise levels. The algorithm was evaluated using seven clinical datasets with T3-T4 laryngeal squamous cell carcinoma from Louvain database where the 3D histology served as reference for comparison. Further evaluation was performed using phantom studies. The proposed method is also compared with a number of commonly used segmentation techniques, including fixed thresholding by 40% of the maximum SUV, the thresholding technique proposed by Nestle et al., and a fuzzy clustering-based approach (FCM). The quantitative data analysis shows that the segmented volumes using the proposed method have the highest overlap with the histology volumes. Moreover, the relative errors of calculated volumes and classification errors are lowest when using the proposed approach. Therefore, the proposed PET segmentation technique seems suitable for accurate tumor volume delineation.