Overall procedure for the diaphragm detection algorithm.
(a) Example of an original image within its ROI, with a diagram illustrating some of the variables in Eqs. (1) and (2). (b) Image filtered by LoG alone; (c) image filtered by LoG and multiplied by the edge direction term ; (d) image filtered by anisotropic diffusion, with ten iterations, a diffusion constant of 250 (image intensity range 500 to 1000), and multiplied by .
Root-mean-square difference between expert and algorithm determined apex positions for four different diaphragm detection algorithms on 21 patient data sets. The results labeled “accuracy” are based on with higher sampling rates on the vertex position and parabola curvature.
Examples of diaphragm detection for (a) DHT; (b) ; and (c) ACM.
Diaphragm tracking result for DHT alone. The rectangle is the ideal bounding rectangle, while the tips of the black and white arrows represent the algorithm and expert determined IHDA, respectively. The parabolic diaphragm contour was generated by the algorithm. (a) ED1: High contrast, accurate localization; (b) ED1, accurate localization; (c) RJ4, diaphragm in collimator region; (d) WB2, contour located between two diaphragms; (e) RJ4, diaphragm in collimator region and interference of the other diaphragm; (f) WB1, confused by overlapping diaphragm; (g) WB2, confused by overlapping diaphragm; and (h) RJ1, interfering edge superior to the diaphragm.
DHT trajectories for four CBCT scans. Manual (lines) and DHT hemidiaphragm apex craniocaudal positions using the default (circles) and slightly modified (cross marks) energy function parameters. Default parameters are , , , , , and . (a) ED1, (b) RJ4, (c) WB1, default and modified (, ), and (d) WB2, default and modified .
Statistics of RMS error, in mm, in 21 patient data sets.
Article metrics loading...
Full text loading...