[Comparison of Detectability for the New Image Processing and Conventional Image Processing by ROC Analysis: A Phantom Study of Chest Radiography].

The purpose of this study is to compare the detectability of diseases the new image processing and the conventional image processing by receiver operating characteristic (ROC) analysis and to show the usefulness of the new image processing. Radiographs with and without nodular cancer models in the chest phantom were used for observation samples. Totally 200 radiographs were evaluated by 10 radiological technologists (each readers had over 20 years or under 4 years of experience). The mean area under the curve (AUC) calculated from the over 20 years group was 0.754 for the new processing and 0.771 for the conventional processing (p value=0.651, 95% confidence interval=-0.084/0.049 (lower bound/upper bound)). On the other hand, the average AUC calculated from under 4 years group was 0.819 for the new processing and 0.678 for the conventional processing (p value= 0.041, 95% confidence interval=0.019/0.262 (lower bound/upper bound)). New image processing provides high detectability in less than 4 years group compared to conventional processing.