Method for Human Identification Based on Radiological Head Images and Its Robustness to Distortions

Computed tomography (CT) is a diagnostic method used to obtain images of internal structures in the human body, including the skull. It assists experts in identifying head injuries, hemorrhages, brain tumors, and other pathologies. Frequently, patients are unconscious and without documentation, leaving specialists unable to access their medical history or identify contraindications. However, if prior radiological studies of the patient exist in the storage system, these records can be used to identify the patient and restore their medical history in a short timeframe, thereby increasing the likelihood of effective treatment. In the information system, each image is linked to the patient’s identification data. Because of operator errors during data entry, studies may be assigned to incorrect identifiers. This work aims to develop a method for patient identification based on radiological images, which can detect such data entry errors and assist in database cleansing (deduplication tasks). The developed method achieved 100% identification accuracy on a test set of 862 images from 238 individuals. The method was also evaluated for its robustness to anonymization procedures and craniofacial surgeries. The effect of template detail representation on recognition accuracy was investigated.