Automated Patient Identity Recognition by Analysis of Chest Radiograph Features

Rationale and Objectives: The aim of this study was to develop a computerized scheme for automated identity recognition based on chest radiograph features. Materials and Methods: The proposed method was evaluated on a database consisting of 1000 pairs of posteroanterior chest radiographs. The method was based on six features: length of the lung field, size of the heart, area of the body, and widths of the upper, middle, and lower thoracic cage. The values for the six features were determined from a chest image, and absolute differences in feature values between the two images (feature errors) were used as indices of image similarity. The performance of the proposed method was evaluated by receiver operating characteristic (ROC) analysis. The discriminant performance was evaluated as the area Az under the ROC curve. Results: The discriminant performance Az of the feature errors for lung field length, heart size, body area, upper cage width, middle cage width, and lower cage width were 0.794 +/- 0.005, 0.737 +/- 0.007, 0.820 +/- 0.008, 0.860 +/- 0.005, 0.894 +/- 0.006, and 0.873 +/- 0.006, respectively. The combination of the six feature errors obtained an Az value of 0.963 +/- 0.002. Conclusion: The results indicate that combining the six features yields a high discriminant performance in recognizing patient identity. The method has potential usefulness for automated identity recognition to ensure that chest radiographs are associated with the correct patient.