森久保 寛

Hiroshi MORIKUBO

珪肺労災病院放射線科

Dept. of Radiology, Rousai Hospital for Silicosis

キーワード : Quantitative diagnosis, Computer-aided diagnosis, Breast tumor, Image processing, Multivariate statistical analysis

We are developing a system to quantitativly diagnose breast tumors using an image processing computer. This report discusses a computer-aided diagnosis system for the differential diagnosis of breast cancer and fibroadenoma.
The 11 parameters (quantitative features of tumors) are calculated by image analyses for a B-mode ultrasonography of a breast tumor. Four of these parameters are associated with texture analysis of internal echo images. The remaining 7 parameters are associated with geometric analysis of mass pictures. These parameters are described according to diagnostic criteria such as shape, border, DW ratio, internal echo, etc.
In 83 cases (33 of breast cancer, 50 of fibroadenoma) the parameters were statistically analyzed. Nine parameters in 11 were significant in the differential diagnosis. Seven parameters in 9 were analyzed by multivariate statistical methods. Discriminant analysis provided a diagnostic accuracy of 89.8%. Principal component analysis presented a well differentiated distribution for cases on a two-dimmensional coordinate (first-second principle factor).
We believe that this image analysis system can be used for the computer-aided diagnosis of breast tumors.