Recognition of regions in brain sections.

This paper addresses the problem of region identification in sequential brain sections and presents a recognition system that finds and tracks region boundaries in those sections. The characteristics of the areas of interest are unique in one sense because they are not stationary. Some regions are hardly discernible. In others, parts of the boundary are missing or so completely blurred that parts of the background may be considered as an extension of the region itself. Moreover, outliers are likely to exist in many cases. Due to the unique properties of brain regions, the emphasis is on robustification and efficiency. The region segmentation problem was expressed as a multi-hypothesis test seeking boundaries that maximize a performance criterion which is general in terms of blur and noise. Boundary candidates are restricted to an adaptive search area around a reference boundary which is usually the outcome of the algorithm from the previous section. The search for the maximum criterion uses a fast first order dynamic programing (DP) procedure, reducing the processing time. Outlier rejection techniques are integrated with the multi-hypothesis test to compensate for both outliers and noise. The result is the reference for the next section. Experimental results on boundary detection are presented. The algorithm is successful in tracing boundaries when the contrast is smaller than the noise power, and when parts of the outlines are missing.