Projects
The proposed projects are:
Incorporating Semantic-Based Features into Semantic-Based Image Retrieval
Computerized Detection of Liver Lesions in Computed Tomography
Project 1 (Furst):Liver Segmentation Using Active Learning. Active learning is a current area of research in machine learning, in which a machine learning algorithm improves its classification as additional labeled data becomes available. Our project applies active learning to the task of liver segmentation, in particular to growing segmented regions from seed regions, using the new growth as feedback to the algorithm.
Project 2 (Furst):Multi-organ Based Segmentation. The goal of our research is to create a robust automatic segmentation algorithm that segments the spleen and liver together that would aid radiologists in diagnosis. Our project explores multi-organ segmentation using texture features, intensity values, and probability maps. This concept can be easily extended to other organs such as kidneys or heart.
Project 3 (Raicu): Incorporating Semantic-Based Features into Semantic-Based Image Retrieval. My project goal is to improve our current content-based image retrieval system by incorporating semantic-based characteristics. After developing a system that accurately retrieves images that correspond to those that are semantically similar, we will attempt to incorporate relevance feedback into the system.
Project 4 (Raicu): Effect of CT Parameters on Lung Nodule Interpretation. Our project goal is to identify the CT parameters that affect the mapping of low-level image features to semantic characteristics. By taking into account the CT parameters in our semantic mapping, we hope to develop a model that can predict semantic characteristics regardless of what CT scanner the image came from.
Project 5 (Raicu): Multi-view Learning for Lung Nodule Classification. The purpose of my research is to evaluate the performance of a Multi-view learning approach to classification of lung nodules for semantic characteristics. I am investigating application of multi-view learning techniques such as feature set partitioning, co-training, and feature subset reduction, as well as various machine learning techniques for classification, including decision trees.
Project 6 (Channin): Radiology Playbook. The project focuses on creating an ontology of radiology terms and procedure steps from RadLex and Northwestern Memorial Hospital's imaging protocols. This ontology will be highly structured and will be used to drive a web application and interface which radiologists, physicians and technicians nationwide can access. We also hope to standardize a naming convention for MR protocols using a new classification grammar developed in Germany.
Project 7 (Channin): Integrating XIP into a FOSS Workstation. This project seeks to integrate the eXtensible Imaging Platform tool into the ClearCanvas Workstation so as to present the output of the XIP Builder tool in a useful environment that is familiar to radiologists, and to allow for the quick and easy modification of the way medical images are displayed.
Project 8 (Armato): Classification of Temporal Subtraction Images. The goal of this research is to use mutual information to quantify registration accuracy of temporally subtracted images in order to detect misregistration before radiologists view the distorted images. This research determines the effect of a lung mask region size, included bin sizes as well as using a texture-based algorithm to detect change before the use of mutual information and can be applied to a wide range of registration techniques.
Project 9 (Suzuki): 3D Computerized Detection of Liver Lesions in Computed Tomography. The goal of this project is to develop a 3D algorithm that will scan liver CT images and identify possible locations of hepatocellular carcinoma (HCC), which is a common, deadly liver cancer. We have developed a program that segments the liver from a series of CT images and attempts to locate HCC using a 3D watershed segmentation.