Purpose: Fast and accurate recontouring is essential in MRI-guided adaptive radiotherapy (MRgART). Auto-segmentation’s, e.g., from image registration or deep learning, are not always acceptable, especially for complex abdominal organs, consequently requiring manual editing. This work aims to develop a fast 3D contour editing tool to accelerate manual editing of inaccurate contours.
Methods: The 3D contour editing tool consisted of 3D and directional half paintbrush with or without edge detection and paintbrush smoothness when edge detection was turned on. The 3D paintbrush with edge detection was developed using a hierarchical triangular mesh approach. The edge detection, dictated by a sensitivity parameter, detected all radially outward thresholds of largest differences in displayed intensity and conformed the contour to the identified organ boundary. The paintbrush smoothness removed sharp edges along the contour using an active contour algorithm. In the absence of edge detection, paintbrush was executed as a cylinder capped with a half-sphere at each end. The 3D editing tool was tested on a variety of inaccurate MRI contours of abdominal organs.
Results: The 3D contour editing tool performed best with the sensitivity value of > 75 and smoothing factor of 100. For a slice spacing t, the brush diameter of (2-4)t allowed simultaneous editing of 2-4 slices. The reduction in editing times using the 3D tool were up to 40% than those with a conventional 2D slice-by-slice editing. For small and/or narrow organs with curved boundaries, the edge detection eliminated the need to change brush size. The half-paintbrush function was especially useful for some organs, e.g., duodenum and small bowel, where the superior to inferior directional editing saved 10% time.
Conclusion: The newly developed 3D contour tool can rapidly edit inaccurate auto-segmented contours of abdominal organs on MRI, substantially accelerating recontouring in MRgART.
Funding Support, Disclosures, and Conflict of Interest: MM and DC are employees of Elekta Inc.