胸部X線像における肺腫瘤影のコンピュータによる画像認識的解析に関する研究

概要

論文の詳細を見る
In order to clarify the correlation between the interpretation by observers and the computerized detection of pulmonary nodules, both an interpreting experiment and a computerized image analysis of conventional chest radiographs of 192 cases, each of which contained a pathologically proven tumor nodule, were performed. Two kinds of directional contrast filters (DCFs) were developed for this study, one for detecting nodular densities, DCF(N), and the other for detecting linear or vascular densities, DCF(V). Three major groups of candidates detected by DCF(N) were tumor nodules (181 areas), overlapping blood vessels (1045 areas), and overlapping ribs (730 areas). The mean values and gradients of DCF(N) output of nodule candidates were significantly different from those of blood vessel candidates and rib candidates. As for the nodules, computerized detection was considered as confirmed if included in the first to the twentieth ranked candidates, as in 170 (88.5%) of the 192 cases. Significant correlation was seen between the rank of the candidates and the DCF(N)-detected area, and between the rank of the candidates and the mean value of DCF(N) output of the central portion of the candidates. Factors affecting the detectability of pulmonary nodules in the experiment of observer interpretation of chest radiographs were the area, density, vascular convergence of nodules, and the degree of overlay of nodules with posterior ribs. These results were also recognized by the data of computerized image analysis. These findings suggest that the computerized image recognition of chest radiographs with DCF(N) and DCF(V) finds some of the characteristic features of lung nodules and might have a role in the development of a computerized support system for detecting pulmonary nodules in chest radiographs.
札幌医科大学の論文
1989-08-01

胸部X線像における肺腫瘤影のコンピュータによる画像認識的解析に関する研究

スポンサーリンク

概要

札幌医科大学 | 論文

スポンサーリンク