Occipitomental View Development of an AI System for Quality Assessment of Occipitomental View Skull Radiograph

Abstract

Because each radiological technician uses different criteria to evaluate for quality assessment of occipitomental view skull radiograph and new radiological technician still lack confidence in making decisions in evaluating radiological image quality. Therefore, this research aims to develop an artificial intelligence system to help evaluation in the unique professional standards quality of occipitomental view (Water view). By collecting Occipitomental view (Water view) radiographic images from Picture Archiving and Communication systems (PACS) at the radiology department of Samutsakhon Hospital from October 1, 2023 to October 31, 2024, a total of 1,020 images are created into a database and labeled each image into 5 categories as follows: 1. the professional standards skull radiographs label as adequate 2. Skull radiograph with anatomy cut off 3. Skull radiograph with rotation 4. Skull radiograph with foreign bodies and 5. Skull radiograph with opened mouth. Then, a model from deep learning technique using the structure of a convolutional neural network is created. The skull radiographs are separated into 3 groups as training set to train the model, validation set to improve the efficiency of the model after training and test set to test the efficiency of the model with arranged into 80:10:10 ratio. The model efficiency performance to help evaluation in the quality assessment of occipitomental view skull radiograph is good with 86.0% accuracy.