Cell Image Segmentation for Diagnostic Pathology

Dorin Comaniciu(1) and Peter Meer(2)

(1)Imaging Research Department
Siemens Corporate Research
Princeton, NJ 08540

(2)ECE Department
Rutgers University
Piscataway, NJ 08854

The colors associated with a digitized specimen representing peripheral blood smear are typ­ ically characterized by only a few, non Gaussian clusters, whose shapes have to be discerned solely from the image being processed. Nonparametric methods such as mode based analysis [10], are particularly suitable for the segmentation of this type of data since they do not constrain the cluster shapes. This chapter reviews an efficient cell segmentation algorithm that detects clusters in the L u v color space and delineates their borders by employing the gradient ascent mean shift procedure [8, 9]. The color space is randomly tessellated with search windows that are moved till convergence to the nearest mode of the underlying probability distribution. After the pruning of the mode candidates, the colors are classified using the basins of attraction. The segmented image is derived by mapping the color vectors in the image domain and enforcing spatial constraints. The segmenter is the core module of the Image Guided Decision Support (IGDS) system [14, 13] which is discussed next. The IGDS architecture supports decision making in clinical pathology and provides components for remote microscope control and multiuser visualization. The primary and long term goal of the IGDS related research is to reduce the number of false negatives during routine specimen screening by medical technologists. The Decision­Support component of the system searches remote databases, retrieves and displays cases which exhibit visual features consistent to the case in question, and suggests the most likely diagnosis according to majority logic. Based on the Micro­Controller component the primary user can command a robotic microscope from the distance, obtain high­quality im­ ages for the diagnosis, and authorize other users to visualize the same images. The system has a natural man­machine interface that contains engines for speech recognition and voice feedback.

Advanced Algorithmic Approaches to Medical Image Segmentation: State-Of-The-Art Applications in Cardiology, Neurology, Mammography and Pathology . J. Suri, S. Singh and K. Setarehdan (Eds.), Springer, 2001, 541-558.
Return to Research: Content-based Retrieval       Return to List of Publications