Analysis of Anatomical Tree-shape Structures
Several structures in the human physiology follow a tree shaped topological morphology. Examples of such tree-topology anatomical structures are dendritic extensions of neurons, intrathoracic airway trees, the blood vessel system and the breast ductal network. In medical image analysis, the topological properties of these structures are usually associated with function, pathology, or the development stages of a disease and can be used to assist medical diagnosis. A primary object of my research is the automated analysis of the morphological variability of these tree-like structures aiming at the extraction of descriptive features that correspond to topological patterns and discriminative characteristics among groups of images. These features characterize and represent the original trees in medical images, capturing properties such as the branching frequency, the tortuosity, and the spatial distribution of branching.
Segmentation of Medical Images
Medical image segmentation algorithms are critical components of medical image analysis systems. With my collaborators, we have proposed an innovative methodology for identifying vessel anatomical structures in galactography images. Our hybrid approach employs the Canny edge detector to obtain an initial boundary of the vessels and then the fuzzy connectedness algorithm is applied locally to each point of discontinuity. A voting scheme using gray level intensity and the Euclidean distance estimates the best connection point. Finally, the image regions are labeled with an adaptive thresholding technique. The experimental results demonstrate the effectiveness of the proposed methodology compared to state-of-the-art techniques.
Ensemble Learning
The idea of ensemble learning is to employ multiple models to obtain better predictive performance than could be obtained from any of the individual models. Bagging, which stands for bootstrap aggregating, is one of the earliest, most intuitive and perhaps the simplest ensemble based algorithms, with a surprisingly good performance. According to this technique, the individual classifiers are combined by taking a simple majority vote of their decisions. Another useful ensemble learning technique is boosting which creates an ensemble of classifiers by resampling the data. However, in boosting, resampling is strategically geared to provide the most informative training data for each consecutive classifier.