Ziel/Aim:
Dynamic renal studies are an important part of nuclear medicine imaging. Making an
accurate medical diagnosis requires the patient to remain motionless for the duration
of the study, which is not always achievable, especially in case of children, due
to the long imaging process. Correcting these displacements manually frame by frame
can be a tedious task. Our goal is to find an automatic and fast motion correction
algorithm that minimizes the motion of kidneys between 2D image frames with various
intensity scales between frames and studies.
Methodik/Methods:
We propose a rigid body registration algorithm which registers the time frames pairwise,
choosing one as fixed image and the following as moving image. We repeat this process
until there are no more frames left. We initialize the starting frame by segmenting
the kidneys on every time frame and choosing the image from which the kidney shapes
start to stagnate. Taking advantage of the small intensity differences between two
neighboring frames, we use the Mean Squares Error metric with gradient descent optimization
method.
Ergebnisse/Results:
We tested our algorithm on 16 different dynamic renal studies (Tc-99 m-DTPA) acquired
on the AnyScan SPECT system. 3 studiescontained no serious motion, while 13 were moderately
or seriously affected by displacements. Using the proposed motion correction method,
the corrected maximum translations were 4.86 mm and 7.29 mm (x and y axis) in the
first group; 31.59 mm and 55.89 mm in the second group. On motion corrected studies
ROI definition on summation image was sufficient to contain the kidneys on each frame
in every case (16/16) which was verified by two specialists in nuclear medicine.
Schlussfolgerungen/Conclusions:
We conclude that our fully automatic method is capable of correcting even large motions
on dynamic renal images while not damaging the quality of those studies on which no
significant motion is present.
