bill@ur-laser.UUCP (Bill Lampeter) (04/04/84)
I have developed an "expert" system that analyzes chest ra-
diographs for tumors. This system was tested on 37 films
that contain nodules. It is capable of finding the nodules
in 92% of the films. In studies of mass screenings of ra-
diographs by radiologists it was found that the radiologists
miss 25-30% of all nodules < 1cm. A Rib Expert determines
whether a candidate nodule (possible tumor) is a rib. A No-
dule Expert, a linear-discriminant-based pattern recognizer,
classifies candidate nodules. All candidate nodules that are
classified as any type of nodule are presented to a radiolo-
gist for further inspection. Radiologists can recognize no-
dules as such once they are pointed-out. If you are in-
terested in this work or want leads to other methods of au-
tomated chest film analysis, which are listed in the bi-
bliographies, contact Peggy Meeker ((716)275-7737, {allegra,
seismo}!rochester!peg) at the Computer Science Dept at the
University of Rochester and request the following TRs:
Lampeter, W.A. "Design, tuning, and performance evaluation
of an automated pulmonary nodule detection system." TR-120,
Computer Science Department, University of Rochester, Ro-
chester NY, 1983.
Lampeter, W.A. "Three image experts which help distinguish
tumors from non-tumors," TR-123, Computer Science Depart-
ment, University of Rochester, Rochester NY, 1984.
Other works of possible interest:
Ballard, D. H., J. Sklansky. "Tumor detection in radio-
graphs," Computers in Biomedical Research, 6, 299-321,
1973.
Jagoe, J.R., "Reading chest radiographs for pneumoconiosis
by computer," Brit. J. Ind. Med., 32, 267-272, 1975.
Toriwaki, J. et.al. Pattern recognition of chest x-ray im-
ages. Comp Grap Pat Recog, 2, 252-271, 1973.
Bill Lampeter
Department of Radiology
School of Medicine and Dentistry
University of Rochester
(716) 275-5101 or (716) 275-3194
{seismo, allegra}!rochester!ur-laser!bill