Artificial intelligence evaluation of computerized tomography images of patient suspected for cervical, thoracic and lumbar vertebral fracture in the emergency department

In the study, a computer-aided algorithm created with artificial intelligence was used recognize cervical, thoracic and lumbar vertebra fractures, which are an important cause of morbidity and mortality in adult patients, in the emergency department with computed tomography (CT) images at an early stage and to accurately determine their location. In the study, which was conducted as a retrospective observational methodological research, the data of 1466 patient who applied to Hacettepe University Faculty of Medicine Adult Emergency Department between April 1, 2019 and April 1, 2023 and were examined with computed tomography for suspected cervical, thoracic and lumbar vertebra fracture were analyzed. After the patient's image were recorded in DICOM format, they were evaluated by standardizing them on the algorithm written in Python programs. The presence and absence of fractures in the patients' computed tomography images by the designed artificial intelligence algorithm; If there is a fracture, its anatomical localitazion. (C1-C7, T1-T12, L1-L5). 42.77% of the patients included in the study were female (n=627) and 57.23% were male (n=839). The median age of the patients was 51.95. Based on the radiologist's interpretation as the gold standard; In terms of cervical vertebral fracture, the sensitivity of artificial intelligence analysis was 35.7%, the specificity was 98%. In terms of thoracic vertebral fracture, the sensitivity of the artificial intelligence analysis was 5.4%, the specificity was 99.5%. In terms of lumbar vertebral fracture, the sensitivity of artificial intelligence analysis was 11.4%, specificity 96.8%. For all cervical, thoracic and lumbar vertebra fractures, the positive predictive value was calculated as 26.7% and the negative predictive value was 97.3%. In conclusion, although the rapid development of AI systems is exciting, rigorous studies to guide AI in various clinical settings and a better understanding of its generalizability are needed to safely translate this important tool into widespread clinical practice.