A neural network based 3D/3D image registration quality evaluator for the head-and-neck patient setup in the absence of a ground truth
The axial and the sagittal views of the FBCT and the CBCT of a head-and-neck patient. The FBCT and the CBCT images are shown in the top and the bottom panels, respectively. The ROI is indicated by the white boxes and the isocenter as the circled cross hairs located in the ROI.
Definitions of the distinctiveness of the optimum and the mirror symmetry features.
Schematic plot of the structure of the neural network used as the registration quality evaluator.
Scatter plots of all the data over two features: DO(1) and DO(5). The networks were constructed using (a) mutual information and the whole image content; (b) mutual information and bony landmarks; (c) mean-squared difference and the whole image content; and (d) mean-squared difference and bony landmarks.
Same as Fig. 4, except that the data were plotted over two other features: MS(1) and MS(5).
Cost function (i.e., the negated mutual information computed using the whole image content) profiles probed along the x, y, and z axes from the selected point-of-solutions. The profiles centered on the successful solutions from two different patients are shown in (a) and (b). The profiles centered on the local minimum solutions are shown in (c) and (d).
An example of the misregistration caused by local minimum trapping. The target FBCT images were shown on the left panels and the registered CBCT images on the right. The top panels give the axial views and the sagittal views show on the bottom.
Range of random disturbances to the initial transformation parameters and the registration success rates when generating training and test data. Abbreviations: information; difference.
Definitions of the criteria that were used to evaluate the performance of a classifier. Abbreviations: positive; positive; negative; negative; predictive value; predictive value.
Classifier performance when the mutual information and the whole image content were used. Abbreviations: See Table II.
Classifier performance when the mutual information and bony landmarks were used. Abbreviations: See Table II.
Classifier performance when the mean-squared difference and the whole image content were used. Abbreviations: See Table II.
Classifier performance when the mean-squared difference and bony landmarks were used. Abbreviations: See Table II.
Article metrics loading...
Full text loading...