Improved Visualization in Medical Endoscopy
Radial Distortion (RD) is a problem that persists in medical endoscopy because of the small size of the lenses. The distortion causes a nonlinear geometric deformation of the image, with the points being moved radially towards the center. The figure below shows an endoscopic image of a planar checkerboard pattern. RD can be easily recognized because straight lines are projected into curves in the image periphery. It can also be observed that strong distortion severely affects the perception of relative size and depth, and the user is hardly able to tell that the imaged scene is planar.
Endoscopic image of planar checkerboard pattern
We developed a versatile and low cost system for superior visualization in clinical endoscopy by rendering geometrically correct perspective images. This is achieved by calibrating the endoscopic camera from a single image (insere link para calibration) and use the projection parametrization for compensating the image distortion via image warping. The figure below shows a scheme of the proposed execution pipeline.
The development of the system led to new methods and models for the following problems:
(i) intrinsic camera calibration in the OR with minimum user intervention;
(ii) robust segmentation of the circular region;
(iii) inference of the relative rotation between lens probe and camera-head using uniquely image information;
(iv) on-line updating of the camera calibration during the clinical procedure.
In addition, we proposed an optimized hybrid CPU+GPU solution for the computation and this, along with carefully designed algorithms, makes it possible correct HD video inputs in real time. Furthermore, and since the approach is scalable, it will be suitable for execution on future GPU generations that are likely to have more cores. It will therefore easily support more complex processing or the processing of larger HD images. The video illustrates the different features of our solution
Superior visualization in clinical endoscopy solution
In addition we provide correction results in real clinical videos (please not that these videos were obtained by A/D conversion which explains the relative poorer quality)
The presented work was partially/totally funded by The Portuguese Agency for Innovation (ADI) through the QREN co-promotion project n 2009/003512 sponsored by the Operational Program "Factores de Competitividade" and the 10 European Fund for Regional Development.