Automatic technique for localizing externally attached fiducial markers in volume images of the head

A method for automatically localizing three-dimensional centroids of a predetermined number M of fiducial markers in at least one image volume, comprising steps of:

• searching the at least one image volume for candidate markers, said candidate markers including a number of three-dimensional connected voxels having similar intensities and being within a predetermined size range and representing each of said candidate markers by a candidate voxel containing the centroid of the candidate marker, including:
• applying a median filter within each of a plurality of slices of the at least one image volume;
• forming a new smaller image volume for each slice by subsampling;
• determining an intensity threshold value;
• producing a binary image from said new smaller image of foreground and background voxels in response to said intensity threshold value;
• filling in any holes in the binary image;
• removing small components and small protrusions on large components appearing in the binary image;
• selecting components in the removed binary image which are most similar to a fiducial marker, said number of components being larger than said predetermined number M;
• and
• providing said selected components as the candidate voxels;
• examining a three-dimensional region in the at least one image volume around each of the candidate voxels and determining which of said candidate markers represented by said candidate voxels most likely correspond to said fiducial markers in said at least one image volume including for each of said candidate voxels:
• determining whether a size and shape of a set of foreground voxels connected to said candidate voxel is similar to a foreground component of a marker;
• discarding the candidate voxel if the size and shape are not determined to be similar;
• determining a centroid of the set of foreground voxels;
• repeating said determining steps until a same centroid has been determined twice in succession;
• discarding the candidate voxel if the determining steps are implemented a predetermined number of times;
• and
• if a same centroid has been determined twice in succession, recording said same centroid as a final centroid and recording a corresponding intensity of said final centroid.