application) The goal of the proposed research is to develop, as aids to radiologists, computerized schemes for the detection and classification of pulmonary nodules in digital chest images (including conventional posterior-anterior images, energy-subtracted images and CT images), and masses and parenchymal distortions in digital mammograms. These methods have the potential to increase diagnostic accuracy in the detection of cancer by reducing the """"""""miss-rates"""""""" associated with unaided radiologists' readings. The investigators plan to (1) create a database in order to investigate the distinguishing characteristics of lung nodules and breast masses, (2) develop digital filters for enhancing and suppressing nodules in PA chest images in order to facilitate the removal of normal structured anatomic background, (3) develop feature-extraction techniques for nodules in PA, dual-energy and CT chest images, (4) investigate pattern-recognition techniques for masses and parenchymal distortions in bilateral digital mammograms, (5) develop a method for the classification of breast masses based on the spectral characteristics of spiculations and (6) evaluate the efficacy of the computer-aided schemes compared with unaided radiologists' performance using ROC analysis and FROC analysis.
Showing the most recent 10 out of 27 publications
