Section: New Results

Algorithms for row registration to improve quality of Tissue MicroArray (TMA) images

Participants : Hoai Nam Nguyen, Charles Kervrann.

Row jittering is a common problem arising in medical imaging devices such as CT (Computer Tomography) and MRI (Magnetic Resonance Imaging) scanners due to errors of synchronization during image acquisition process. On scanners designed and developed by Innopsys, the problem becomes more challenging mainly because the pixel displacement is non constant along each row (Fig. 14 ) and possibly sub-pixel (i.e. non integer translation). To overcome this drawback, we first proposed a window-based algorithm to approximate the translation at each pixel by selecting the value that best minimizes a matching criteria over a finite set of possible sub-pixel translations. We obtained satisfying results with this method on real data with fast computation time (see Fig. 14 ). Furthermore, this matching criteria has been considered as a data fidelity term and was combined to a regularization term to promote a smooth solution and correct small artifacts which were not removed with the window-based method. To minimize the energy functional, we have adopted the quadratic relaxation technique and proximal method. This algorithm is slower and is initalized by the window-based algorithm to produce very encouraging results and elimination of all undesirable artifacts (see Fig. 14 ).

Collaborators: Vincent Paveau and Cyril Cauchois (Innopys).

Figure 14. Illustration of the two-row registration algorithms. First row : full width input image (by courtesy of Innopsys). Second row : zooms on input image (blue boxes). Third row : corrected with window-based algorithm. Fourth row : corrected with variational method.