Detection of Radiographic Abnormalities in Mammograms by Means of Optical Scanning and Computer Analysis

Abstract

Listen

Translate article

Mammography is now established as a relatively routine method of examination of the breasts. Although it is not yet accepted as a screening procedure for breast cancer there is evidence that it could play a role in early detection and salvage of life if applied on a wide enough scale (1–3). Because of the problems inherent in the routine viewing of large numbers of examinations of presumably asymptomatic patients we have proposed the automation of reading of the radiographs by means of optical scanning and computer interpretation. Our work has, we believe, demonstrated the feasibility of this method. Films of the breasts were made, using bolus immersion technic. The breast was immersed in either mineral oil or alcohol, and translateral decubitus films were obtained. The resulting radiograph was converted to a positive image and scanned by a converted radio-facsimile scanner. This scanner carries the print on a rotating drum and advances the photocell by means of a lead-screw to measure the reflected light from the print. A facsimile scanner was used, since we thought it unwise to invest in an expensive transparency scanner without knowing the specifications which would be needed. The basic programing could be done with immersion mammograms and subsequently refined when a scanner capable of reading negatives became available. The fascimile scanner has a maximum resolution of 180 lines per inch and a usable gray scale of 32, although only 16 levels were employed in the program. The voltage signal from the scanner is passed through the analog to digital interface of the CDC 160A computer and recorded on digital tape. Background density of the films was measured and used to correct all measured densities to a standard value. The film was divided into 64 small rectangles, in an 8 × 8 array; the size and location of these rectangles for each film was arranged in a standardized manner relative to the chest wall, the skin, and the nipple. This technic insures that each rectangle contains a fixed percentage of the total tissue area (rather than a fixed area) and is located in the same part of the tissue. Comparison of two breasts with different shapes and sizes can then be done more uniformly on the basis of the rectangles. The distribution of optical densities for each segment is characterized by 4 vectors which measure the distribution of density elements. The first of these is referred to as the area vector and tells us, for each of the 16 density levels, what fraction of the scan points within the rectangle contain that density level. The second vector requires a further slicing of the rectangle into 16 strips. One can then state for each of the 16 density levels the number of slices within the rectangle which contain any scan points at the particular density level. The third vector measures the uniformity of the distribution of each density level within the rectangle.