The proposed projects are:
Project 1 (Furst): “Evaluation of multi-valued probabilistic classification systems where uncertainty exists” This research investigates methods to evaluate machine learning classification problems where uncertainty exists in a multi-valued ground truth, and the classifier outputs a distribution of probabilities. Methods to explore will include distribution comparisons, uncertainty-weighted metrics, ranking comparisons, and modified ROC analysis. Issues in dealing with the classification and evaluation of uncertain labels and multi-valued probabilistic outputs will be a main focus.
Project 2 (Furst): “Prediction of Chronic Fatigue Syndrome”. CFS (Chronic Fatigue Syndrome) is a chronic condition with symptoms that are severe, but often difficult to detect upon physical examination. They include debilitating fatigue, headaches and unrefreshing sleep. The objective is to determine which symptoms and patient conditions define the CFS and develop a system that will be able to differentiate between CFS and regular fatigue.
Project 3 (Raicu): “Content-Based Image Retrieval”. The objective is to create a content-based image retrieval (CBIR) system for lung nodules whose similarity results correspond to the human visual perception of similarity. New similarity metrics that mimic the human perception of similarity are investigated using quantitative diagnostic features as the input signal and semantic concepts as the teaching signal. The expected outcome will be a CBIR system that can be used in conjunction with Computer-Aided Diagnosis (CAD) systems to provide both ‘visual aids’ and semantic diagnostic characterizations.
Project 4 (Raicu):” Three-Dimensional Analysis of Lung Nodules”. This project investigates the use of 3D image features - intensity-based and morphological-based - for classifying the semantic characteristics of lung nodules present in computer tomography (CT) scans. Various classification techniques will be explored and the effectiveness of 3D vs. 2D image features will be evaluated with respect to classification performance.
Project 5 (Armato): “Automated Detection of Interval Change in Temporal Subtraction Chest Images”. Temporal subtraction images of the chest provide radiologists with a powerful tool for the enhanced visualization of pathologic change between temporally sequential chest radiographs. The utility of these images could be further improved through the development of computer-aided diagnostic methods designed to automatically detect regions of change in temporal subtraction images. In particular, regions of actual pathologic change must be distinguished from misregistration artifacts. Image gray-level statistics and texture measures will be calculated within the “lung mask” region of the temporal subtraction image, which will be decomposed into an array of regions-of-interest. The quantitative information extracted in this manner will be used to identify regions that overlap foci of pathologic change.
Project 6 (Suzuki): “Probabilistic Atlases for human bodies in FDG-PET”. The goal of the project is to overlay knowledge of human anatomy on the FDG-PET images in order to allow for more accurate segmentation. Several algorithms to register hundreds of FDG-PET images are experimented in particular for chest and head, thus creating an atlas of human anatomy in those regions.
Project 7 (Suzuki):“Computer-Aided Diagnosis for Early Detection of Colon Cancer in CT Colonoscopy”. Colorectal cancer is the second leading cause of cancer deaths in the U.S. “Virtual colonoscopy” or CT colonography (CTC) is a new test used for screening for colorectal carcinoma through the acquisition of a CT scan of the colon. Computer-aided diagnosis (CAD) for the detection of polyps has the potential to improve diagnostic performance of radiologists in CTC. CAD automatically detects polyps in CTC and displays the locations of suspicious polyps for radiologists to review. Our purpose in this research project is to improve the performance of our CAD scheme in terms of the sensitivity as well as specificity of the detection of polyps in CTC. Students would become familiar with diagnostic imaging and computer algorithms, in particular Gabor filters and the massively trained artificial neural network (MTANN).