Application of form features in digital cell analysis of non-Hodgkin's lymphomas

The form of a biological cell nucleus can be characterized by the well-known circularity factor, which is derived from the area of an object and its circumference. More sophisticated form features are introduced, which are calculated from the curvature of an object ("bending energy") or from invariant moments. To investigate the sensitivity of the various form features on controlled changes of form and the behaviour under rotation and scaling, algebraic curves similar to the form of real nucleus profiles are generated. Analysis of the shape characteristics of biological cells requires an extraction of the boundaries of the nuclei. This was performed by an edge detection algorithm using eight gradient masks followed by a contour tracing procedure with feedback. The suitability of the introduced form features for classification of different nucleus profiles of non-Hodgkin's lymphomas (NHL) was tested using a Bayes classifier.