Computerized Classification of Mammary Gland Patterns in Whole Breast Ultrasound Images

  • Authors:

  • Yuji Ikedo;Takako Morita;Daisuke Fukuoka;Takeshi Hara;Hiroshi Fujita;Etsuo Takada;Tokiko Endo

  • Affiliations:

  • Department of Intelligent Image Information, Division of Regeneration and Advanced Medical Sciences, Graduate School of Medicine, Gifu University, Gifu, Japan;Department of Mammary Gland, Chunichi Hospital, Japan;Technology Education, Faculty of Education, Gifu University, Gifu, Japan;Department of Intelligent Image Information, Division of Regeneration and Advanced Medical Sciences, Graduate School of Medicine, Gifu University, Gifu, Japan;Department of Intelligent Image Information, Division of Regeneration and Advanced Medical Sciences, Graduate School of Medicine, Gifu University, Gifu, Japan;Center of Medical Ultrasonics, Dokkyo Medical University School of Medicine, Tochigi, Japan;Department of Radiology, National Hospital Organization, Nagoya Medical Center, Japan

  • Venue:
  • IWDM '08 Proceedings of the 9th international workshop on Digital Mammography
  • Year:
  • 2008

Quantified Score

Hi-index 0.00

Visualization

Abstract

Several whole breast ultrasound (US) scanners have recently been developed for breast cancer screening. In ultrasonographic screening techniques that utilize scanners, assessment of the mammary gland pattern in US images by a radiologist is required. We developed a method of mammary gland analysis to automatically classify whole breast US images into three categories: mottled pattern (MP), intermediate pattern (IP), and atrophic pattern (AP). Our database included 50 patients who underwent US of the entire breast, and they were classified as 12 MP, 24 IP, and 14 AP cases. First, we extracted a volume of interest (VOI) including mammary gland regions. Following this, we extracted image features, i.e., the average pixel value (APV), the number of small hypoechoic regions (SHR), and Haralick's texture features, from the VOI. Finally, a canonical discriminant analysis with APV, SHR, and four texture features was applied for classification of mammary gland patterns. The performance of this classification method was 82.0% (41/50). We found that it is possible to classify whole breast ultrasound images based on mammary gland patterns. The classification method can be applied to estimate the risk of breast cancer based on US images, and it could also be applied in computer-aided diagnosis (CAD) systems for the detection of ultrasonographic breast cancer.