抄録
われわれは眼底画像の血管像をコンピュータで解析することによって,狭窄の程度を定量的に評価し,眼科医の診断を文援するコンピュータ支援診断システムを開発してきた.本システムでは,眼底画像から血管の抽出を行い,動静脈の分類を行った後,動静脈血管径比によって動脈狭窄を検出する.動静脈血管径比の算出には動静脈を正確に分類することが重要である.しかし,特徴量には画像の位置に依存してばらつきがあり,分類精度に悪影響を与えていた.そこで,トップハット変換した画像を用いて分類する手法を考案した.83枚の眼底画像に本手法を適用した結果,真陽性率76%のときに偽陽性数は画像1枚あたり1.4本であり,本手法の有効性を確認した.
The purpose of this study is to develop a computer-aided diagnosis (CAD) system for early detection of arteriolar narrowing on retinal images. Our scheme consists of extraction of blood vessel, classification of artery and vein, and detection of arteriolar narrowing by the artery-vein diameter ratio. The classification of artery and vein is important for improving the performance of the detection. However, the features used in the classification depend on the position of the image and the variation of the features negatively affects the classification accuracy. In order to solve this issue, we propose a technique to reduce the variation of the features by use of top-hat image in the feature extraction. Eighty-three images were used to estimate our scheme. As a result, the sensitivity for detection arteriolar narrowing was 76% with 1.4 false-positive vessels per image. Therefore, the technique we proposed in this study may be useful for detection of arteriolar narrowing.
