Premium
A computerized analysis of masses on mammograms (in Japanese)
Author(s) -
Goto Miki
Publication year - 2000
Publication title -
medical physics
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.473
H-Index - 180
eISSN - 2473-4209
pISSN - 0094-2405
DOI - 10.1118/1.1287651
Subject(s) - sponge spicule , mammography , pixel , artificial intelligence , radiology , computer aided diagnosis , computer science , breast cancer , computer vision , nuclear medicine , medicine , cancer , anatomy
Two new schemes for analysis of mammographic images were developed, a diagnostic logic for classifying masses on mammograms and a multistage pendulum filter for detecting spicules around the mass. To construct a computer‐aided diagnosis scheme for detecting breast cancer requires a knowledge of the diagnostic process. We investigated the classification logic for the diagnosis of breast masses on mammograms in cooperation with a radiologist specialized in mammography. We tested our classification logic by use of 99 mammograms in which the diagnosis had been confirmed. The accuracy for classifying the mass as malignant or benign was very high (sensitivity of 84% and specificity of 96%). Existence of spicules around a mass is one of the important signs which characterize malignant tumors. We developed an automated method for detecting spicules. The detection was performed with a newly developed “multistage pendulum filter.” A “spicule value,” which indicates the probability of “spicule presence” around the mass, was calculated. This filter was evaluated on 71 digitized mammograms. Digitized mammograms were obtained with a laser film scanner with a 0.1 mm×0.1 mm pixel size and 1024 gray levels. The digitized data were then reduced to image with effective pixel size of 0.2 mm for detecting spicules. It detected most of the spicules seen by a radiologist. The sensitivity was 89% and the specificity was 80%, which demonstrates the effectiveness of our method.