Ze Jin

Computer-assisted framework for machine-learning–based delineation of GTV regions on datasets of planning CT and PET/CT images

We have proposed a computer-assisted framework for machine-learning–based delineation of gross tumor volumes (GTVs) following an optimum contour selection (OCS) method. The key idea of the proposed framework was to feed image features around GTV contours (determined based on the knowledge of radiation oncologists) into a machine-learning classifier during the training step, after which the classifier produces the ‘degree of GTV’ for each voxel in the testing step. Initial GTV regions were extracted using a support vector machine (SVM) that learned the image features inside and outside each tumor region (determined by radiation oncologists). The leave-one-out-by-patient test was employed for training and testing the steps of the proposed framework. The final GTV regions were determined using the OCS method that can be used to select a global optimum object contour based on multiple active delineations with a LSM around the GTV. The efficacy of the proposed framework was evaluated in 14 lung cancer cases [solid: 6, ground-glass opacity (GGO): 4, mixed GGO: 4] using the 3D Dice similarity coefficient (DSC), which denotes the degree of region similarity between the GTVs contoured by radiation oncologists and those determined using the proposed framework. The proposed framework achieved an average DSC of 0.777 for 14 cases, whereas the OCS-based framework produced an average DSC of 0.507. The average DSCs for GGO and mixed GGO were 0.763 and 0.701, respectively, obtained by the proposed framework. The proposed framework can be employed as a tool to assist radiation oncologists in delineating various GTV regions.

Computer-assisted delineation of lung tumor regions in treatment planning CT images with PET/CT image sets based on an optimum contour selection method.

To assist radiation oncologists in the delineation of tumor regions during treatment planning for lung cancer, we have proposed an automated contouring algorithm based on an optimum contour selection (OCS) method for treatment planning computed tomography (CT) images with positron emission tomography (PET)/CT images. The basic concept of the OCS is to select a global optimum object contour based on multiple active delineations with a level set method around tumors. First, the PET images were registered to the planning CT images by using affine transformation matrices. The initial gross tumor volume (GTV) of each lung tumor was identified by thresholding the PET image at a certain standardized uptake value, and then each initial GTV location was corrected in the region of interest of the planning CT image. Finally, the contours of final GTV regions were determined in the planning CT images by using the OCS. The proposed method was evaluated by testing six cases with a Dice similarity coefficient (DSC), which denoted the degree of region similarity between the GTVs contoured by radiation oncologists and the proposed method. The average three-dimensional DSC for the six cases was 0.78 by the proposed method, but only 0.34 by a conventional method based on a simple level set method. The proposed method may be helpful for treatment planners in contouring the GTV regions.

Impact of pixel-based machine-learning techniques on automated frameworks for delineation of gross tumor volume regions for stereotactic body radiation therapy

The aim of this study was to investigate the impact of pixel-based machine learning (ML) techniques, i.e., fuzzy-c-means clustering method (FCM), and the artificial neural network (ANN) and support vector machine (SVM), on an automated framework for delineation of gross tumor volume (GTV) regions of lung cancer for stereotactic body radiation therapy. The morphological and metabolic features for GTV regions, which were determined based on the knowledge of radiation oncologists, were fed on a pixel-by-pixel basis into the respective FCM, ANN, and SVM ML techniques. Then, the ML techniques were incorporated into the automated delineation framework of GTVs followed by an optimum contour selection (OCS) method, which we proposed in a previous study. The three-ML-based frameworks were evaluated for 16 lung cancer cases (six solid, four ground glass opacity (GGO), six part-solid GGO) with the datasets of planning computed tomography (CT) and 18F-fluorodeoxyglucose (FDG) positron emission tomography (PET)/CT images using the three-dimensional Dice similarity coefficient (DSC). DSC denotes the degree of region similarity between the GTVs contoured by radiation oncologists and those estimated using the automated framework. The FCM-based framework achieved the highest DSCs of 0.79±0.06, whereas DSCs of the ANN-based and SVM-based frameworks were 0.76±0.14 and 0.73±0.14, respectively. The FCM-based framework provided the highest segmentation accuracy and precision without a learning process (lowest calculation cost). Therefore, the FCM-based framework can be useful for delineation of tumor regions in practical treatment planning.

Computer-assisted target volume determination

The gross tumor volume (GTV) regions are the fundamental regions used to determine the clinical target volumes (CTVs) and planning target volume (PTV). The accuracy of the GTVs may affect tumor control and adverse events related to organs at risk or normal tissue. The PTV is the volume that includes the CTV plus CTV-to-PTV margin including the internal margin (IM) and the setup margin (SM). This chapter introduces the computational segmentation approaches for GTV and computational determination of the CTV-to-PTV margin.

Automated method for extraction of lung tumors using a machine learning classifier with knowledge of radiation oncologists on data sets of planning CT and FDG-PET/CT images

We have developed an automated method for extraction of lung tumors using a machine learning classifier with knowledge of radiation oncologists on data sets of treatment planning computed tomography (CT) and 18F-fluorodeoxyglucose (FDG)-positron emission tomography (PET)/CT images. First, the PET images were registered with the treatment planning CT images through the diagnostic CT images of PET/CT. Second, six voxel-based features including voxel values and magnitudes of image gradient vectors were derived from each voxel in the planning CT and PET /CT image data sets. Finally, lung tumors were extracted by using a support vector machine (SVM), which learned 6 voxel-based features inside and outside each true tumor region determined by radiation oncologists. The results showed that the average DSCs for 3 and 6 features for three cases were 0.744 and 0.899, and thus the SVM may need 6 features to learn the distinguishable characteristics. The proposed method may be useful for assisting treatment planners in delineation of the tumor region.

An ellipsoid convex enhancement filter for detection of asymptomatic intracranial aneurysm candidates in CAD frameworks

PURPOSE Various kinds of enhancement filters have been developed in computer-aided diagnostic (CAD) frameworks for asymptomatic intracranial aneurysms in magnetic resonance angiography (MRA). However, many bending or branching portions on vessels are also enhanced by the conventional filters as false positives in 3.0 T MRA, which can visualize smaller vessels compared with 1.5 T MRA. To overcome this problem, this study focused on developing an ellipsoid convex enhancement (ECE) filter, which can selectively enhance aneurysms while reducing false positive contrasts on bending or branching portions on vessels, for detection of asymptomatic intracranial aneurysm candidates in CAD frameworks. METHODS The ECE filter was mathematically designed to enhance various convex regions in the intensity space such as convex aneurysms, in which the ratio of the shortest and longest diameters for aneurysms corresponds to the ratio of reciprocals of the square roots of the first and third eigenvalues of a Hessian matrix. The proposed ECE filter was evaluated by measuring an average contrast for false positive models and free-response receiver operating characteristic curves between two simple CAD frameworks using the ECE and conventional filters based on a leave-one-out-by-patient test. MRA images for thirty patients (male: 10, female: 20; age: 48-86 yr, mean: 69.2) with 31 unruptured aneurysms (longest diameter: 2.0-5.5 mm, mean: 3.7 mm) were selected for this study. RESULTS The average contrast for false positive models was reduced by 51.4% using the ECE filter, compared with the conventional filter for the convex regions with ratios of the shortest and longest diameters less than 0.4. The number of false positives per case was decreased from 41.1 to 22.8 on average at a sensitivity of 87% by using the ECE filter. CONCLUSIONS The ECE filter would be useful for boosting the performance of the CAD framework of asymptomatic intracranial aneurysms by providing higher contrast aneurysms and lower contrast false positives such as bending or branching portions on vessels.

TU-H-CAMPUS-JeP2-03: Machine-Learning-Based Delineation Framework of GTV Regions of Solid and Ground Glass Opacity Lung Tumors at Datasets of Planning CT and PET/CT Images

PURPOSE In radiation treatment planning, delineation of gross tumor volume (GTV) is very important, because the GTVs affect the accuracies of radiation therapy procedure. To assist radiation oncologists in the delineation of GTV regions while treatment planning for lung cancer, we have proposed a machine-learning-based delineation framework of GTV regions of solid and ground glass opacity (GGO) lung tumors following by optimum contour selection (OCS) method. METHODS Our basic idea was to feed voxel-based image features around GTV contours determined by radiation oncologists into a machine learning classifier in the training step, after which the classifier produced the degree of GTV for each voxel in the testing step. Ten data sets of planning CT and PET/CT images were selected for this study. The support vector machine (SVM), which learned voxel-based features which include voxel value and magnitudes of image gradient vector that obtained from each voxel in the planning CT and PET/CT images, extracted initial GTV regions. The final GTV regions were determined using the OCS method that was able to select a global optimum object contour based on multiple active delineations with a level set method around the GTV. To evaluate the results of proposed framework for ten cases (solid:6, GGO:4), we used the three-dimensional Dice similarity coefficient (DSC), which denoted the degree of region similarity between the GTVs delineated by radiation oncologists and the proposed framework. RESULTS The proposed method achieved an average three-dimensional DSC of 0.81 for ten lung cancer patients, while a standardized uptake value-based method segmented GTV regions with the DSC of 0.43. The average DSCs for solid and GGO were 0.84 and 0.76, respectively, obtained by the proposed framework. CONCLUSION The proposed framework with the support vector machine may be useful for assisting radiation oncologists in delineating solid and GGO lung tumors.

TH-CD-206-11: An Ellipsoid Convex Enhancement Filter Based Computer-Aided Diagnostic Framework of Intracranial Aneurysms in MRA Images

PURPOSE For early detections of asymptomatic intracranial aneurysms, computer aided diagnostic (CAD) frameworks of intracranial aneurysms in magnetic resonance angiography (MRA) images were developed with various enhancement filters. However, discrimination between aneurysms and false positive structures, such as bending or branching portions of vessels could still be a challenging task for the enhancement filters. Therefore, we focused on developing an ellipsoid convex enhancement (ECE) filter based CAD framework, which is capable of enhancing aneurysms while reducing the effect on bending and branching portions of vessels, for detection of intracranial aneurysm candidate. METHODS The proposed ECE filter was designed for selectively enhancing aneurysms with certain ellipsoid convex shapes, whose ratios of the shortest to longest diameters correspond to the ratio of reciprocals of the square roots of the first and third eigenvalues of Hessian matrices. This filter was applied to artificial images with Gaussian ellipsoid models and clinical MRA images of 30 patients (Male: 10, Female: 20; age: 48 to 86 years, mean: 69.2) with 31 unruptured aneurysms (longest diameter: 2.0 to 5.5 mm, mean: 3.7 mm) based on a leave-one-out-by-patient test. RESULTS The proposed ECE filter reduced the average contrast for false positive Gaussian ellipsoid models by 51.4%. The number of false positives per patient decreased from 41.1 to 22.8 at a sensitivity of 87% by using the ECE filter. CONCLUSION The ellipsoid convex enhancement filter based CAD framework of intracranial aneurysms in MRA images would be useful by providing a higher contrast between aneurysms and false positive structures such as bending or branching portions of vessels.

Good relationships between computational image analysis and radiological physics

Good relationships between computational image analysis and radiological physics have been constructed for increasing the accuracy of medical diagnostic imaging and radiation therapy in radiological physics. Computational image analysis has been established based on applied mathematics, physics, and engineering. This review paper will introduce how computational image analysis is useful in radiation therapy with respect to radiological physics.