Elastic Registration of two images using Natural Skin Markers
Registration of two sets of skin markers from different dates for the early detection of skin cancer.
Category: Computer Assisted Diagnosis
Technologies: Image Processing, Elastic Registration, VTK.
The objective of this module is the registration of two sets of skin markers obtained from images at two different instants of time. This is an essential element of one suite for the early detection of skin cancer. The developed algorithms contributes to the automation of a procedure for medical diagnosis in whole body photo imaging, which are used in the early detection of potentially deadly skin cancers. The main algorithm of this module was designed as a VTK class which houses a specialized Thin Plate Spline Transform.
The figure below shows two incomplete point clouds. The blue dots (become reds) are mapped to the black spots. An elastic transformation was applied to the blue dots and the red dots were obtained, superimposed on the black points.
The experiment in this figure contains new markers that appear to the right and other markers that disappear. The elastic registration is achieved by the two clouds of markers based on common markers in the two clouds. This experiment is important because it verifies that our method can work with different number of markers of the two clouds and it does not matter that new markers appear, and it does not matter if there are markers that disappear.
The following figure shows two images taken at different time instants. Skin markers was obtained with our own method of detection, based on the metaphor of a spider specialized in recognizing natural skin markers.
The next figure, at left, shows the new arrangement of the red markers, superimposed to the blue markers, after applying the elastic registration algorithm. At the right, the figure has been expanded to appreciate the small shifts of some markers. Some markers appear and disappear as was shown in the example of the two fish.
Elastic Registration of two Images Spaced in Time using Trac...
Elastic Registration of two images using Natural Skin Marker...