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

Abstract

As positron-emission tomography (PET) images have low spatial resolution and much noise, accurate image segmentation is one of the most challenging issues in tumor quantification. Tumors of the uterine cervix present a particular challenge because of urine activity in the adjacent bladder. Here, we propose and validate an automatic segmentation method adapted to cervical tumors. Our proposed methodology combined the gradient field information of both the filtered PET image and the level set function into a level set framework by constructing a new evolution equation. Furthermore, we also constructed a new hyperimage to recognize a rough tumor region using the fuzzy c-means algorithm according to the tissue specificity as defined by both PET (uptake) and computed tomography (attenuation) to provide the initial zero level set, which could make the segmentation process fully automatic. The proposed method was verified based on simulation and clinical studies. For simulation studies, seven different phantoms, representing tumors with homogenous/heterogeneous-low/high uptake patterns and different volumes, were simulated with five different noise levels. Twenty-seven cervical cancer patients at different stages were enrolled for clinical evaluation of the method. Dice similarity coefficients (DSC) and Hausdorff distance (HD) were used to evaluate the accuracy of the segmentation method, while a Bland-Altman analysis of the mean standardized uptake value (SUVmean) and metabolic tumor volume (MTV) was used to evaluate the accuracy of the quantification. Using this method, the DSCs and HDs of the homogenous and heterogeneous phantoms under clinical noise level were 93.39 ±1.09% and 6.02 ±1.09 mm, 93.59 ±1.63% and 8.92 ±2.57 mm, respectively. The DSCs and HDs in patients measured 91.80 ±2.46% and 7.79 ±2.18 mm. Through Bland-Altman analysis, the SUVmean and the MTV using our method showed high correlation with the clinical gold standard. The results of both simulation and clinical studies demonstrated the accuracy, effectiveness, and robustness of the proposed method. Further assessment of the quantitative indices indicates the feasibility of this algorithm in accurate quantitative analysis of cervical tumors in clinical practice.

Radiomic and clinical predictors of cachexia in non-small cell lung cancer patients treated with immunotherapy

“…The primary lung tumors of PET and CT images were semi-automatically segmented with an improved level-set method based on the gradient fields (28), and 30 randomly selected nodules in the training cohort were segmented twice. After spatial registration using a rigid transformation by maximizing the Dice Similarity Coefficients on the condition that the maximal axial cross sections of the tumor nodules were aligned, fusion images were generated from the fused PET and CT images on a voxel-wise basis (29).…”

Cervical Cell Segmentation from Overlapped Cells using Fuzzy C-Means Clustering

“…Various results in Fig.2.1 show some captured results of a tested image generated by these processes. Extended depth of Field (EDF) in fig 2.2 shows a single image where all objects were focused on a discrete wavelet transforms [2]. In combining of all cellular objects in focus the Stack of focal plane images to produce a different view of single EDF image.…”

“…Analyze of the sensitivity, specificity and predictive values of the results using colposcopic directed biopsy as reference was done. In basic test of identifying cervical cancer is Pap screening test done every time to see pre-destructive changes in a specimen of cells, magnifying lens slide gives visual assessment of uterine cervix [2]. If every female adheres to current HPV vaccination programs the total number of female deaths from cervical cancer globally will drop.…”

Spatial Regularization and Adaptive Distance Metric Methods through DST for Tumor Segmentation

“…Compared to all the methods discussed above the clustering methods are more applicable for unsupervised segmentation because of their diversity nature. The FCM [2], [8] method has more stable segmentation compared to other clustering methods. there many version of FCM came into existence including the spatial information ,here in this paper the method called Evidential c-Means (ECM) [3], is proposed considering the spatial information directly modeled through DST [4], [6], to get appropriate and further enhancement in the performance for low quality MRI images.…”