Maxillo Facial Fracture Detection System (MFDS) in Accident Victims with Deep Learning Techniques using Artificial Intelligence

Abstract

It is proposed that a novel Maxillo Facial Fracture Detection System (MFDS) be developed to identify traumatic fractures in patients, using the use of deep learning and transfer techniques. To classify future Computed Tomography (CT) scans as "fracture" or "no Fracture," we re-trained a Conventional Neural Network on CT scans after it had been initially trained on non-medical images. A total of 358 CT scans were used for training the model (150) patients were annotated as having a fracture, and 78 were annotated as having no fracture). Five patients were classified as having "no Fracture," and the remaining 38 were classified as having a fracture; these 50 patients made up the statistical analysis validation data set. The whole dateset used for validation included 60 CT images, 38 of which had "fracture" labels and 10 had no such labels, and both sets were used. Both a focus on individual slices and on grouped slices for patients was used in the tests. Patients were considered to have fractures if two successive slices had a fracture probability of more than 0.99. In terms of classifying Maxillo Facial fractures, the model has an 99% accuracy rate, as evidenced by the patients' outcomes. The MFDS model may not be able to fully take over the Radiologist's duties, but it can certainly help by lowering the likelihood of mistakes, protecting patients from harm by shortening the time it takes to make a diagnosis, and alleviating the disproportionate load of staying in the hospital.