CPC G16H 30/20 (2018.01) [G16H 30/40 (2018.01)] | 11 Claims |
1. A computer-implemented method for robust and automatic face de-identification from CT and MRI using one or more processors, the method comprising:
receiving DICOM files of patient bodies obtained by a CT or MRI system; extracting the related 3D images and sorting them along the superior-inferior axis based on the “Image Position (Patient)” attribute; converting CT images from stored units to output units through rescale intercept and rescale slope; applying intensity cropping; transforming the images to patient coordinate system; detecting the body using adaptive thresholding and consequently extracting the head from it on the basis of geometric properties; centering and aligning the head and image along the anterior-posterior axis if needed; filling any holes and cavities found in the segmented head; calculating the convex hull for the x-y plane; extracting the face anonymizing mask by subtracting the head from its convex hull; expanding the mask along the anterior-posterior direction; enhancing the face mask with its convex hull on y-z plane; calculating the border separating the mask from the face; embedding both the mask and the border into the original images; setting the intensity values of the mask and the border so to be indistinguishable from the corresponding head slice; applying a smoothing filter to enhance blending; transforming back the masked image to its original orientation, center and coordinate system; writing the processed images back to the original DICOM files.
|