In this paper, we describe a system implemented to automatically classify and quantitatively measure the extent of a lung disease called Scleroderma using High Resolution Computed Tomography (HRCT) imagery. Scleroderma is a disease characterized by a slowly developing fibrosis in the lungs of its victims. Early diagnosis of the extent of the disease using CT imagery can be especially difficult, as many of its visual features are weak and very subtle. These visual subtleties can lead to differences in analysis between radiologists when gauging the exact extent of the disease. Not having an absolute standard with which to judge the extent of the disease can hinder evaluations on the effectiveness of new treatments applied during the disease's early stages. We have attempted to remove the subjective component by designing a robust system which aids radiologists in measuring the extent of the disease during it's earliest stages. This system employs a bank of 17 Maximum Likelihood classifiers trained on the variety of tissue types typically seen within Scleroderma HRCT imagery. The system also employs several heuristic constraints. These constraints are used to mimic some of the decision making processes that radiologists typically employ during their analysis. Results of this classifier system are demonstrated on a series of HRCT exams of patients in the early stages of the disease. These results were found to compare favorably with physiological tests performed on these patients. This research was done as a collaborative effort between Los Alamos National Laboratory (LANL) and the Radiology Department at National Jewish Center for Immunology and Respiratory Medicine (NJCIRM).
R. Fortson, D. Lynch, and J. Newell. Automated Segmentation of Scleroderma in High Resolution CT Imagery. Los Alamos Technical Report LA-UR-95-2401 Los Alamos National Laboratory, Los Alamos, NM, 1995. [ Abstract | PDF (461 KB) ]
