Clustering of non-convex patterns for medical imaging
Isak Gath, Anna Smolyak Iskoz · 2002
The study treats the problem of partitional clustering of non-convex patterns of arbitrary shape, with no definite analytical description of the data. A new method for calculating the distance is defined, according to a data induced metric principle. Calculation of the distance is carried out using weighted graphs. The new distance is introduced in the fuzzy k-means algorithm, and the modified fuzzy k-means algorithm is tested on synthetic data, and on MRI images of the heart.