Pattern Classification Methods for Analysis and Visualization of Brain Perfusion CT Maps - Computational Intelligence Paradigms in Advanced Pattern Classification - page 165

Fig. 9 Example visualization of data acquired from patience with suspicious of brain stroke
after identification of potential brain lesions with algorithm described in this article. (A) Set
of different images. (B) Volumetric data from head - neck protocol. (C) Detailed view of
CBV map with perfusion lesions marked as white regions, in the left side of the screen de-
scription of image done by DMD algorithm. (D) CT image and the same CT image with
prognosis for brain tissues.
Author's solution was tested on off - the - shelf computer with Intel Core 2
Duo CPU 3.00 GHz processor, 3.25 GB RAM, Nvidia GeForce 9600 GT graphic
card with 32 - bit Windows XP Professional. Creative pd 1120 USB web cam was
used as a video capture device. Because volume rendering is the most time de-
manding algorithm system performance tests was done for AR visualization of
single volume (the time of rendering 2D square image is relatively small).
The speed of marker detection algorithm was 30 frames per second (fps) for
320 x 240 resolution of the camera and 15 fps for 640 x 480. Those are the max-
imal speed of image capture for tested web camera so the fps limit of marker
detection software could not be found without better (faster) camera.
The speed of 3D rendering depends of the size of rendered object. The smaller
the object is (less number of pixel to compute) the rendering process run faster. In
author's research virtual object was localize in the nearest position to the camera
as it is possible in order to generate the maximal possible size of the rendered
model. Tests were performed on three 3D models generated from computed tomo-
graphy (CT) data acquired from three different patients. Data was stored in collec-
tion of DICOM files. The size of rendered volume was 256 x 256 x 248 pixels,