Combined fuzzy logic and random walker algorithm for PET image tumor delineation

A physics-guided modular deep-learning based automated framework for tumor segmentation in PET

An important need exists for reliable positron emission tomography (PET) tumor-segmentation methods for tasks such as PET-based radiation-therapy planning and reliable quantification of volumetric and radiomic features. To address this need, we propose an automated physics-guided deep-learning-based three-module framework to segment PET images on a per-slice basis. The framework is designed to help address the challenges of limited spatial resolution and lack of clinical training data with known ground-truth tumor boundaries in PET. The first module generates PET images containing highly realistic tumors with known ground-truth using a new stochastic and physics-based approach, addressing lack of training data. The second module trains a modified U-net using these images, helping it learn the tumor-segmentation task. The third module fine-tunes this network using a small-sized clinical dataset with radiologist-defined delineations as surrogate ground-truth, helping the framework learn features potentially missed in simulated tumors. The framework was evaluated in the context of segmenting primary tumors in 18F-fluorodeoxyglucose (FDG)-PET images of patients with lung cancer. The framework's accuracy, generalizability to different scanners, sensitivity to partial volume effects (PVEs) and efficacy in reducing the number of training images were quantitatively evaluated using Dice similarity coefficient (DSC) and several other metrics. The framework yielded reliable performance in both simulated (DSC: 0.87 (95% confidence interval (CI): 0.86, 0.88)) and patient images (DSC: 0.73 (95% CI: 0.71, 0.76)), outperformed several widely used semi-automated approaches, accurately segmented relatively small tumors (smallest segmented cross-section was 1.83 cm2), generalized across five PET scanners (DSC: 0.74 (95% CI: 0.71, 0.76)), was relatively unaffected by PVEs, and required low training data (training with data from even 30 patients yielded DSC of 0.70 (95% CI: 0.68, 0.71)). In conclusion, the proposed automated physics-guided deep-learning-based PET-segmentation framework yielded reliable performance in delineating tumors in FDG-PET images of patients with lung cancer.

A Novel Framework for Automated Segmentation and Labeling of Homogeneous Versus Heterogeneous Lung Tumors in [18F]FDG-PET Imaging

PURPOSE

Determination of intra-tumor high-uptake area using 2-deoxy-2-[¹⁸F]fluoro-D-glucose ([¹⁸F]FDG) positron emission tomography (PET) imaging is an important consideration for dose painting in radiation treatment applications. The aim of our study was to develop a framework towards automated segmentation and labeling of homogeneous vs. heterogeneous tumors in clinical lung [¹⁸F]FDG-PET with the capability of intra-tumor high-uptake region delineation.

PROCEDURES

We utilized and extended a fuzzy random walk PET tumor segmentation algorithm to delineate intra-tumor high-uptake areas. Tumor textural feature (TF) analysis was used to find a relationship between tumor type and TF values. Segmentation accuracy was evaluated quantitatively utilizing 70 clinical [¹⁸F]FDG-PET lung images of patients with a total of 150 solid tumors. For volumetric analysis, the Dice similarity coefficient (DSC) and Hausdorff distance (HD) measures were extracted with respect to gold-standard manual segmentation. A multi-linear regression model was also proposed for automated tumor labeling based on TFs, including cross-validation analysis.

RESULTS

Two-tailed t test analysis of TFs between homogeneous and heterogeneous tumors revealed significant statistical difference for size-zone variability (SZV), intensity variability (IV), zone percentage (ZP), proposed parameters II and III, entropy and tumor volume (p < 0.001), dissimilarity, high intensity emphasis (HIE), and SUV_min (p < 0.01). Lower statistical differences were observed for proposed parameter I (p = 0.02), and no significant differences were observed for SUV_max and SUV_mean. Furthermore, the Spearman rank analysis between visual tumor labeling and TF analysis depicted a significant correlation for SZV, IV, entropy, parameters II and III, and tumor volume (0.68 ≤ ρ ≤ 0.84) and moderate correlation for ZP, HIE, homogeneity, dissimilarity, parameter I, and SUV_min (0.22 ≤ ρ ≤ 0.52), while no correlations were observed for SUV_max and SUV_mean (ρ < 0.08). The multi-linear regression model for automated tumor labeling process resulted in R ² and RMSE values of 0.93 and 0.14, respectively (p < 0.001), and generated tumor labeling sensitivity and specificity of 0.93 and 0.89. With respect to baseline random walk segmentation, the results showed significant (p < 0.001) mean DSC, HD, and SUV_mean error improvements of 21.4 ± 11.5 %, 1.4 ± 0.8 mm, and 16.8 ± 8.1 % in homogeneous tumors and 7.4 ± 4.4 %, 1.5 ± 0.6 mm, and 7.9 ± 2.7 % in heterogeneous lesions. In addition, significant (p < 0.001) mean DSC, HD, and SUV_mean error improvements were observed for tumor sub-volume delineations, namely 5 ± 2 %, 1.5 ± 0.6 mm, and 7 ± 3 % for the proposed Fuzzy RW method compared to RW segmentation.

CONCLUSION

We proposed and demonstrated an automatic framework for significantly improved segmentation and labeling of homogeneous vs. heterogeneous tumors in lung [¹⁸F]FDG-PET images.

Background based Gaussian mixture model lesion segmentation in PET

PURPOSE

Quantitative (18)F-fluorodeoxyglucose positron emission tomography is limited by the uncertainty in lesion delineation due to poor SNR, low resolution, and partial volume effects, subsequently impacting oncological assessment, treatment planning, and follow-up. The present work develops and validates a segmentation algorithm based on statistical clustering. The introduction of constraints based on background features and contiguity priors is expected to improve robustness vs clinical image characteristics such as lesion dimension, noise, and contrast level.

METHODS

An eight-class Gaussian mixture model (GMM) clustering algorithm was modified by constraining the mean and variance parameters of four background classes according to the previous analysis of a lesion-free background volume of interest (background modeling). Hence, expectation maximization operated only on the four classes dedicated to lesion detection. To favor the segmentation of connected objects, a further variant was introduced by inserting priors relevant to the classification of neighbors. The algorithm was applied to simulated datasets and acquired phantom data. Feasibility and robustness toward initialization were assessed on a clinical dataset manually contoured by two expert clinicians. Comparisons were performed with respect to a standard eight-class GMM algorithm and to four different state-of-the-art methods in terms of volume error (VE), Dice index, classification error (CE), and Hausdorff distance (HD).

RESULTS

The proposed GMM segmentation with background modeling outperformed standard GMM and all the other tested methods. Medians of accuracy indexes were VE <3%, Dice >0.88, CE <0.25, and HD <1.2 in simulations; VE <23%, Dice >0.74, CE <0.43, and HD <1.77 in phantom data. Robustness toward image statistic changes (±15%) was shown by the low index changes: <26% for VE, <17% for Dice, and <15% for CE. Finally, robustness toward the user-dependent volume initialization was demonstrated. The inclusion of the spatial prior improved segmentation accuracy only for lesions surrounded by heterogeneous background: in the relevant simulation subset, the median VE significantly decreased from 13% to 7%. Results on clinical data were found in accordance with simulations, with absolute VE <7%, Dice >0.85, CE <0.30, and HD <0.81.

CONCLUSIONS

The sole introduction of constraints based on background modeling outperformed standard GMM and the other tested algorithms. Insertion of a spatial prior improved the accuracy for realistic cases of objects in heterogeneous backgrounds. Moreover, robustness against initialization supports the applicability in a clinical setting. In conclusion, application-driven constraints can generally improve the capabilities of GMM and statistical clustering algorithms.

Generic and robust method for automatic segmentation of PET images using an active contour model

PURPOSE

Although positron emission tomography (PET) images have shown potential to improve the accuracy of targeting in radiation therapy planning and assessment of response to treatment, the boundaries of tumors are not easily distinguishable from surrounding normal tissue owing to the low spatial resolution and inherent noisy characteristics of PET images. The objective of this study is to develop a generic and robust method for automatic delineation of tumor volumes using an active contour model and to evaluate its performance using phantom and clinical studies.

METHODS

MASAC, a method for automatic segmentation using an active contour model, incorporates the histogram fuzzy C-means clustering, and localized and textural information to constrain the active contour to detect boundaries in an accurate and robust manner. Moreover, the lattice Boltzmann method is used as an alternative approach for solving the level set equation to make it faster and suitable for parallel programming. Twenty simulated phantom studies and 16 clinical studies, including six cases of pharyngolaryngeal squamous cell carcinoma and ten cases of nonsmall cell lung cancer, were included to evaluate its performance. Besides, the proposed method was also compared with the contourlet-based active contour algorithm (CAC) and Schaefer's thresholding method (ST). The relative volume error (RE), Dice similarity coefficient (DSC), and classification error (CE) metrics were used to analyze the results quantitatively.

RESULTS

For the simulated phantom studies (PSs), MASAC and CAC provide similar segmentations of the different lesions, while ST fails to achieve reliable results. For the clinical datasets (2 cases with connected high-uptake regions excluded) (CSs), CAC provides for the lowest mean RE (-8.38% ± 27.49%), while MASAC achieves the best mean DSC (0.71 ± 0.09) and mean CE (53.92% ± 12.65%), respectively. MASAC could reliably quantify different types of lesions assessed in this work with good accuracy, resulting in a mean RE of -13.35% ± 11.87% and -11.15% ± 23.66%, a mean DSC of 0.89 ± 0.05 and 0.71 ± 0.09, and a mean CE of 19.19% ± 7.89% and 53.92% ± 12.65%, for PSs and CSs, respectively.

CONCLUSIONS

The authors' results demonstrate that the developed novel PET segmentation algorithm is applicable to various types of lesions in the authors' study and is capable of producing accurate and consistent target volume delineations, potentially resulting in reduced intraobserver and interobserver variabilities observed when using manual delineation and improved accuracy in treatment planning and outcome evaluation.