Department of Radiological Technology, Saitama Saiseikai Kawaguchi General Hospital, 5-11-5 Nishikawaguchi, Kawaguchi City, Saitama, 332-8558, Japan.
Department of Radiological Sciences, Tokyo Metropolitan University Graduate School, Human Health Sciences, 7-2-10 Higashi-Ogu, Arakawa-ku, Tokyo, 116-8551, Japan.
Radiol Phys Technol. 2020 Jun;13(2):177-186. doi: 10.1007/s12194-020-00566-3. Epub 2020 May 6.
This paper describes the auto-analysis of the mammary gland visualized on mammography images to eliminate the subjective evaluation error between physicians using pixel values and image similarity, including pattern recognition. The mammography images including the heterogeneously dense and extremely dense images were divided into two groups based on the result of the subjective breast classification as the dense breast, and non-dense breast. One hundred and thirty images obtained during screening were set as search images, and 101 evaluation images were classified using zero-mean normalized cross-correlation. Concerning the conventional method, we employed the variance histogram analysis method of Yamazaki et al. The concordance rate for the subjective breast classification result obtained using the conventional and proposed methods was 79.2% and 89.1%. The image similarity evaluation method, which analyzes the pattern of the pixel values, enabled the breast classification while eliminating ambiguity in the subjective breast classifications among physicians.
本文描述了对乳腺钼靶图像进行自动分析，以消除医师使用像素值和图像相似性进行主观评估时的误差，包括模式识别。将包括不均匀致密和极度致密图像的乳腺钼靶图像根据主观乳腺分类结果分为两组，即致密乳腺和非致密乳腺。将 130 张筛查图像作为搜索图像，使用零均值归一化互相关对 101 张评估图像进行分类。对于传统方法，我们采用了 Yamazaki 等人的方差直方图分析方法。传统方法和提出的方法在主观乳腺分类结果上的符合率分别为 79.2%和 89.1%。这种分析像素值模式的图像相似性评估方法能够在消除医师主观乳腺分类歧义的同时进行乳腺分类。
