Abstract
Cancer classification using high-throughput mass spectrometry data for early disease detection and prevention has recently become an attractive topic of research in bioinformatics. Recently, several studies have shown that the synergy of proteomic technology and pattern classification techniques is promising for the predictive diagnoses of several cancer diseases. However, the extraction of some effective features that can represent the identities of different classes plays a critical factor for any classification problems involving the analysis of complex data. In this paper we present the concept of a fuzzy fractal dimension that can be utilized as a novel feature of mass spectrometry data, We then applied vector quantization to model the class prototypes using the fuzzy fractal dimensions for classification. Using a simple vector-quantization based classification rule, the overall average classification rates of the proposed approach were found to be superior to some other methods. In bio-imaging classification, we applied vector quantization and Markov modeling methods for cell-phase classification using time-lapse fluorescence microscopic image sequences. However this method is not always effective because cell features are treated with equal weight of importance that may not be always true. We proposed a subspace vector-quantization method to overcome this drawback. The proposed method can automatically weight cell features based on their attribute importance in fuzzy clustering analysis. Two weighting algorithms based on fuzzy c-means and fuzzy entropy clustering were studied, whose performances improved the classification rates.