Multimodality imaging in radiotherapy treatment planning and the definition of the biophysical target as an integration of information from different imaging modalities.
Multimodality image analysis framework showing the intertwining between image segmentation and image registration.
A block diagram of the proposed concurrent multimodality segmentation method. The method starts by preprocessing of the different images to reduce noise and correct for artifacts. Next, the images are fused using a coregistration method (rigid or deformable). Then, the concurrent segmentation starts by identifying the region-of-interest (a sphere is used unless prior knowledge is emphasized) and iteratively segmenting the biophysical target by integrating information from various imaging modalities until the convergence criteria is satisfied.
Deformable image segmentation by the level set method. (a) Representation of the level set surface at time of the evolving function. (b) A projected view showing the evolution direction. The function evolves at a velocity proportional to the curvature and inversely proportional to the image gradient. In this example, is represented with a signed Euclidean distance transform of value . The contour is extracted at (with negative values inside the contour and positive values outside). This principle is generalized to multimodality images.
Analysis of lung PET/CT case. (a) A fused PET/CT displayed in CERR with manual contouring shown of the subject’s right gross tumor volume. The contouring was performed separately for CT (in orange), PET (in green), and fused PET/CT (in red) images. (b) The MVLS algorithm was initialized with a circle (in white) of 9.8 mm diameter, evolved contour is steps of ten iterations (in black), and the final estimated contour (in thick red). The algorithm converged in 120 iterations in a few seconds. The PET/CT ratio weighting was selected as 1:1.65. (c) MVLS results is shown along with manual contour results on the fused PET/CT. Note the agreement of fused PET/CT manual contour and MVLS . (d) MVLS contour superimposed on CT (top) and PET (bottom) separately.
Analysis of cervix PET/CT. (a) Coregistered PET/CT after correction for small hardware misalignment with mutual information increasing from 1.22 to 1.23. The maximum SUV contour (in green) is shown in the region-of-interest. (b) The MVLS algorithm is initialized with a circle (in white) of 15.9 mm diameter, curve evolution in steps of ten iterations (in blue), and the final estimated contour (in thick black). The algorithm converged in 30 iterations. This quick convergence is by part due to the deformed circular shape of the tumor and the PET sharp gradient. The PET/CT ratio weighting was selected as 1:1.25. (c) MVLS result is shown along with SUV contour on the fused PET/CT. Note the high agreement as expected. (d) MVLS contour superimposed on CT (top) and PET (bottom). The PET/CT ratio weighting was selected as 1:1.25.
A 3D generalization of the MVLS algorithm in the case of PET/CT cervix. Different slices of the 3D volume of the PET/CT data are shown. (a) The data were fused for display (foreground PET and background CT). The MVLS algorithm is initialized with a sphere (in light blue) of 15.9 mm diameter, curve evolution in steps of ten iterations (in magenta), and the final estimated contour (in thick dark blue). The algorithm converged in 30 iterations. (b) MVLS estimated contour superimposed on CT. (c) MVLS estimated contour superimposed on PET.
Analysis of prostate MRI/CT. (a) Checkerboard display of the original MRI/CT. (b) Coregistered MRI/CT using a NMI algorithm, which increased NMI from 1.07 to 1.11. Right: selected region-of-interest containing the prostate which is chosen as the target for multimodality analysis. (c) Overlaid display of MRI/CT (foreground is MR and background in gray scale is the CT). In the figure, we show initially drawn contour (in white). Note in this case using an arbitrary shape might lead to erroneous results. This initial contour is emphasized in the MVLS algorithm as prior knowledge of the prostate shape. Evolved contour is shown (in blue) after ten iterations and the final estimated contour (in thick red). The algorithm converged in 50 iterations. A 1.5:1 MR to CT importance weighting was used in this example. In (d) and (e) we show the MVLS result superimposed on CT (d) and MR (e) for comparison. Note that using other multispectral MR images beside the could potentially provide further useful information to the algorithm.
Physical phantom study. (a) A picture of the anthropomorphic head phantom showing the internal target spheres used to evaluate the MVLS concurrent segmentation algorithm. (b) Registered PET/CT in CERR. (c) Registered MRI/CT in CERR.
Sample physical phantom results using the MVLS. Different slices of the 3D volume are shown. (a) The MVLS algorithm is initialized with a sphere (in white) in the fused PET/CTMR domain, curve evolution is shown in steps of ten iterations (in blue), and the final estimated contour (in thick black). The algorithm converged in 30 iterations. Note the accurate capturing of the sphere boundary (black circle) despite the fact that parts of the initial contour lie outside the sphere while others lie inside the sphere. In addition, the algorithm can adapt to topological changes (capturing the hole in the middle of the sphere—or necrotic regions in case of tumors for example—if desired). Only the exterior boundary, however, was used for evaluation purposes. For illustration, the estimated contour is superimposed separately on (b) CT, (c) PET, and (d) MR.
Phantom study evaluation summary of the different metrics (mean, standard deviation) for the four balls boundary detection using MVLS.
Article metrics loading...
Full text loading...