Novel multimodality segmentation using level sets and Jensen‐Rényi divergence

Abstract

Purpose: Positron emission tomography (PET) is playing an increasing role in radiotherapy treatment planning. However, despite progress, robust algorithms for PET and multimodal image segmentation are still lacking, especially if the algorithm were extended to image-guided and adaptive radiotherapy (IGART). This work presents a novel multimodality segmentation algorithm using the Jensen-Rényi divergence (JRD) to evolve the geometric level set contour. The algorithm offers improved noise tolerance which is particularly applicable to segmentation of regions found in PET and cone-beam computed tomography. Methods: A steepest gradient ascent optimization method is used in conjunction with the JRD and a level set active contour to iteratively evolve a contour to partition an image based on statistical divergence of the intensity histograms. The algorithm is evaluated using PET scans of pharyngolaryngeal squamous cell carcinoma with the corresponding histological reference. The multimodality extension of the algorithm is evaluated using 22 PET/CT scans of patients with lung carcinoma and a physical phantom scanned under varying image quality conditions. Results: The average concordance index (CI) of the JRD segmentation of the PET images was 0.56 with an average classification error of 65%. The segmentation of the lung carcinoma images had a maximum diameter relative error of 63%, 19.5%, and 14.8% when using CT, PET, and combined PET/CT images, respectively. The estimated maximal diameters of the gross tumor volume (GTV) showed a high correlation with the macroscopically determined maximal diameters, with a R 2 value of 0.85 and 0.88 using the PET and PET/CT images, respectively. Results from the physical phantom show that the JRD is more robust to image noise compared to mutual information and region growing. Conclusions:The JRD has shown improved noise tolerance compared to mutual information for the purpose of PET image segmentation. Presented is a flexible framework for multimodal image segmentation that can incorporate a large number of inputs efficiently for IGART.

Pathology-validated PET image data sets and their role in PET segmentation

“…The segmentation tools evaluated against the pathology in the original publications were mostly simple threshold or adaptive threshold methods. More advanced segmentation methods, which promise to be able to handle realistic tumors with irregular shape and non-uniform activity, have been evaluated in later publications against some of the PSPVs reviewed here [47][48][49][50][51][53][54][55]. Important conclusions have been reached for these more advanced methods, as pointed out in the previous section.…”

“…They found that the wavelet transform-enhanced FCM resulted in a smaller mean error of the maximal diameter estimation also for the NSCLC lesions. Markel et al [51] also used the MAASTRO NSCLC data set [12] to evaluate their multimodality segmentation tool using level sets and JRD and found that JRD outperformed an SBR method when using only PET and noted further performance improvement when information from both PET and CT is used. Sharif et al [55] used the MAASTRO data set [12] to evaluate an artificial neural network approach.…”

“…Zaidi et al [50] used the Louvain LSCC data set [25] to compare the performance of nine algorithms including five threshold methods, a level set method, a stochastic expectationmaximization method, fuzzy clustering-based segmentation (FCM) and a spatial wavelet-based FCM (FCM-SW) and found FCM-SW to be most accurate. Markel et al [51] also used the Louvain LSCC data set [25] to evaluate a Fixed thresholds (SUV and % of maximum) [38,39,40] Adaptive threshold (signal-to-background ratio based) [25,46,50,52] Gradient [39,47] Fuzzy C-means [48,50] Active contours [49,50] Possibility theory [53] Multimodality using level sets [51] Esophagus Fixed threshold (SUV and % of maximum) [41,42] Visual [41,42] Non-small-cell lung cancer (NSCLC)…”

“…Threshold [27-29, 33, 43] Adaptive threshold [12,31,33,52] Gradient [47] Fuzzy C-means [48] Active contours [49] FLAB [54] Neural network [55] Multimodality using level sets [51] Cervical cancer Threshold [34] Colon, rectal and sigmoid cancer Threshold [39,45] Adaptive threshold [35,44] Gradient [35,39] multimodality segmentation tool using level sets and Jensen-Renyi divergence (JRD). They compared the results to those from Zaidi et al [50], and found that the JRD approach was second to the FCM-SW method.…”

A Segmentation Algorithm for Quantitative Analysis of Heterogeneous Tumors of the Cervix With 18 F-FDG PET/CT

“…Most of these methods are based on the assumption that the lesions should be homogeneous [20], and may not obtain a satisfied results on heterogeneous tumors usually. Therefore many researches are focused on improving the traditional region-based method to adapt to the heterogeneous tumors in recent years, and some of them obtained stratified results [21] The fourth kind of methods, represented by level set and active contours [22][23][24][25], are based on the boundary of the object. Though how to locate the boundaries in the noisy and low-resolution PET images is challenging, these methods could overcome the restricted condition of homogeneity and obtain more accurate results.…”