Background. MR-based volumetric analysis of regional brain volumes is gaining widespread acceptance in clinical and epidemiological settings. Regional analyses are usually accomplished with techniques such as operator-guided tracing, thresholding or topologic sampling. These techniques are generally accurate, but time consuming. The availability of automated segmentation algorithms would allow greater utility for MRI quantification. Aim. To develop an algorithm based on artificial intelligence (AI) that automatically separates brain from non-brain tissues as the first step to a completely automated regional segmentation algorithm. Method. The algorithm is divided into 3 parts. Part I uses an AI-based morphogenetic neuron model to generate a mathematical model of the brain image. Part II uses local splines to identify the brain borders and Part III extracts the brain from the skull. The algorithm is completely automatic. Results. An example of the algorithm is shown below. The raw image is in the center, to the left the image after processing with the artificial intelligence algorithm, and to the right is a segmented image based on the threshold method of DeCarli et al. (1992) after manual separation of brain from non-brain tissues. The algorithm works equally well on different slices of the same subject and on homologous slices of different subjects Further validation of the method is ongoing. Conclusions. The method is promising to automatically segment the brain from non-brain tissue without operator intervention.
|Original language||English (US)|
|Issue number||4 SUPPL.|
|State||Published - 2000|
ASJC Scopus subject areas
- Clinical Neurology