Radiologists miss about 25-30% of all pulmonary nodules smaller than 1.0 cm. in mass screenings. A system for the automated detection of the pulmonary nodule based on that of Hallard has been designed, tuned, and tested on a 43 chest radiographs [Ballard, 1973). The goal of this system is to aid the radiologist in locating a pulmonary nodule by indicating a few sites in the radiograph that are most likely to be nodules. Computer image analysis programs that respond to specific types of anatomic features have been devised and are incorporated in a pattern recognizer, which uses linear discriminant analysis to classify the candidate nodule sites. Candidate nodule sites that are not classified as nodules are eliminated from the list of sites that are presented to the radiologist for inspection. The pattern recognizer was trained with the features from 2750 candidate nodules, which came from 37 films and another pattern recognizer was trained with the features from 402 candidate nodules from 9 films. This research demonstrates that pattern recognition techniques and procedurally driven image experts are capable of reducing the number of candidate nodule sites that a radiologist must inspect from at most 12 to at most 4 if he is to be 99% confident of having inspected any nodule detected by the system which was trained with 37 films. The radiologist must be willing to accept a film true positive rate of 88% (as opposed to a film true positive rate of 92%) for the convenience of having fewer points to inspect. These film true positive rates are derived from 37 films which contain nodules that were evaluated by the system. The particular contributions of this work lies in the implementation and testing of a spline filter, a preprocessing step, which removes background variations in the radiograph so that nodules are more visible; the development of Vascularity and Rib Experts which recognize these classes of candidate nodules; and in die implementation of the particular features that are extracted from the candidate nodule and used by the pattern classifier.

Library of Congress Subject Headings

Radiography, Medical; Medical photography; Pattern recognition systems

Publication Date


Document Type


Department, Program, or Center

School of Photographic Arts and Sciences (CIAS)


Francis, Ronald

Advisor/Committee Member

Ballard, Dana

Advisor/Committee Member

Brown, Christopher


Note: imported from RIT’s Digital Media Library running on DSpace to RIT Scholar Works. Physical copy available through RIT's The Wallace Library at: TR750.L35 1982


RIT – Main Campus