Abstract:Defining the boundaries of tumors (segmentation) on clinical imaging is required for the accurate delivery of radiation therapy. Previous work by our group has highlighted the utility of ADC maps from diffusion weighted MRI in the manual segmentation of cervical cancer. An auto-segmentation method would shorten the time required and provide a standardized means for generating contours. We now present a novel hybrid algorithm based on the watershed transform as a solution to this problem. An extension of the watershed algorithm including two pre-processing steps, the normalized discrete derivative calculation of the ADC map with the Sobel operator, and over-segmentation minimization with an adjustable H-minima transform, was used to auto-segment cervical tumors. This method was compared against manual contouring on ADC maps of 11 cervical cancer patients by comparing volumes and overlaps of these methods to metabolic tumor volumes delineated in registered FDG-PET scans, an imaging gold standard. Our method was also tested against a previously validated threshold method of 40% SUVmax on microPET images of 9 mice with implanted tumors based on how accurately the watershed or threshold delineated volumes matched volumes measured from calibrated images of tomographically sliced sections of these tumors after sacrifice, a histological gold standard. Conformity index, defined as the volume measured by the test method divided by the volume measured by the gold standard method (CFI = Vt/Vo), and Dice coefficient, defined as twice the overlap of the volumes of the test method and gold standard method divided by the sum (DC = 2*[Vt∩Vo]/[Vt+Vo]), were compared using paired t-tests. We compared the CFI of watershed contoured ADC maps (1.26 ± 0.20) and manual contoured ADC maps (1.24 ± 0.19) in reference to FDG-PET (p = .14) as well as the DC of watershed contoured ADC maps (0.60 ± 0.04) and manual contoured ADC maps (0.64 ± 0.04) in reference to FDG-PET (p = .71). Similarly in the mice models, we compared the CFI of watershed contoured microPET (1.04 ± 0.08) and threshold contoured microPET (1.09 ± 0.09) with respect to histology (p = 0.86). Thus, contours generated using the watershed method were not significantly different at determining tumor volumes or overlap compared to manual contouring for ADC images or threshold autocontouring for PET images. We propose a novel watershed based hybrid algorithm for segmentation of ADC maps of cervical tumors in comparison with a gold standard imaging method, FDG-PET, in humans. This method was also cross-validated against a threshold method in microPET in mice to confirm accuracy in a second imaging modality. Overall, the proposed algorithm achieved similar accuracy to manual segmentation.

An Improved Graph Cut Segmentation Method for Cervical Lymph Nodes on Sonograms and Its Relationship with Node'S Shape Assessment

Ultrasonographic Segmentation of Cervical Lymph Nodes Based on Graph Cut with Elliptical Shape Prior

Sonographic Feature Extraction Of Cervical Lymph Nodes And Its Relationship With Segmentation Methods

Segmentation of 3D High-frequency Ultrasound Images of Human Lymph Nodes Using Graph Cut with Energy Functional Adapted to Local Intensity Distribution

Supervoxels Based Graph Cut for Medical Organ Segmentation

Segmentation of Thyroid Glands and Nodules in Ultrasound Images Using the Improved U-Net Architecture.

Segmentation of acetowhite region in uterine cervical image based on deep learning

Acetowhite region segmentation in uterine cervix images using a registered ratio image

A Modified Level Set Algorithm Based on Point Distance Shape Constraint for Lesion and Organ Segmentation

Improved Graph-Cut Segmentation for Ultrasound Liver Cyst Image

Application of Computer-Aided Diagnosis to the Sonographic Evaluation of Cervical Lymph Nodes

Evaluation Studies of Ultrasonic Image Features Extraction for Cervical Lymph Nodes

Computer-aided diagnosis of cervical lymph nodes on ultrasonography

Segmentation Based Features For Lymph Node Detection From 3-D Chest Ct

Automatic Detection and Segmentation of Axillary Lymph Nodes.

A Novel Approach to Segment and Classify Regional Lymph Nodes on Computed Tomography Images.

Centerline-Based Vessel Segmentation Using Graph Cuts

A Clinically Useful Watershed-Based Method of Auto-Segmenting Apparent Diffusion Coefficient (adc) Maps of Cervical Cancer

Improved image segmentation algorithm based on Normalized Cut

Iterative Graph Cuts Segmentation with Local Constraints

A Practical Segmentation Method for Automated Screening of Cervical Cytology