乳房Ｘ線写真の胸壁側に存在する欠損腫りゅう陰影への扇型モデルによる自動検出法

DOI

書誌事項

タイトル別名
  • Automated Detection of Partial Loss Masses Existing Near Chest Wall on Mammograms Using a Sector-Shaped Model.

抄録

We have been developing automated detection algorithms to detect masses on mammograms. In our algorithm, we devised an adaptive thresholding technique for detecting masses, but there was a problem because our system failed to detect the masses with a partial loss of region. To deal with this problem, we propose a new method in this study. The partial loss mass is identified by similarity with the sector-shaped mass pattern models. To calculate the similarity, four features are applied, such as (1) average of pixel values, (2) standard deviation of pixel values, (3) template matching based on the sector-shaped mass pattern models, and (4) concentration feature by density gradient. Our automated mass detection system was improved by combining a previous method with a new one. To evaluate the new method for detecting the masses with a partial loss of region, we examined 335 mammograms. When the searching area is limited near the chest wall, the TP (True Positive) rate by applying our new method is higher than that by using the previous method with fewer false positives per image. Our system improved this by combining the previous method with the new one; the detection sensitivity of the partial loss masses was improved from 57% to 96%, though the number of false positives was increased to 1.09 per image, from 0.88. These results indicated that the new technique effectively improved the detection performance of our computer-aided diagnosis system.

収録刊行物

詳細情報 詳細情報について

  • CRID
    1390282679531707008
  • NII論文ID
    130004327124
  • DOI
    10.11239/jsmbe1963.39.305
  • ISSN
    21855498
    00213292
  • 本文言語コード
    ja
  • データソース種別
    • JaLC
    • CiNii Articles
  • 抄録ライセンスフラグ
    使用不可

問題の指摘

ページトップへ