Toward patient identification using chest CT scan

Abstract

3D digital medical images are usually generated by computerized medical equipment such as CT scanners and MRI machines. Selected anatomic information and structures can be extracted as features from this type of volume data. These features, if unique enough, can be used to identify the patient. This can be viewed as a special field of biometrics-- identification of individuals using biological traits. As an example of implementation, we present a solution for same-patient decision in chest CT volume data. Our method uses the lung area profiles along the sagital, coronal and axial planes. The area profile curves of the two volume data are cross-correlated to find the best scale factors as well as the best offsets. The extension of this work is to evaluate the "goodness-of-fit" in order to classify two studies being or not from the same patient. We apply our method to 3 panels of pairs of the same patient and pairs of different patient and report classification performance.