Segmentation of masticatory muscles on ultrasonographic images using artificial intelligence in pediatric population

Abstract

Aim: The objective of this study is to utilize artificial intelligence for the segmentation of masticatory muscles in ultrasound images.

Materials and Methods: The study comprised a cohort of 60 pediatric patients with ultrasonographic images of the masseter, anterior temporal, and lateral pterygoid muscles, 120 images for each muscle, right and left, totaling 360 muscle images. Within the context of this research, the YOLOv8-Seg deep learning model was employed to automatically conduct the segmentation of the masseter, anterior temporal and lateral pterygoid muscles within ultrasonography images. In this study, an artificial intelligence algorithm (Roboflow, Inc., Des Moines, Iowa, USA) was developed to autonomously carry out the segmentation of the masseter, anterior temporal and lateral pterygoid muscles. A total of 120 images for each muscle group were randomly divided into training, validation and test sets.

Results: For the muscle segmentations on the test data, the true positive (TP), false positive (FP) and false negative (FN) values were 18, 0, 0 for masseter muscle, 18, 0, 0 for temporal muscle and 16, 1, 1 for lateral pterygoid muscle, respectively. The model's F1 score, precision and sensitivity values are 1.0, 1.0 and 1.0 for masseter muscle, 1.0, 1.0 and 1.0 for temporal muscle and 0.92, 0.94 and 0.94 for lateral pterygoid, respectively.

Conclusion: In summary, segmentation techniques based on deep learning for analyzing ultrasonography images of anatomical structures like masticatory muscles have great potential in clinical applications. Precise segmentation of muscles through this technology can play a crucial role in the diagnosis and follow-up of diverse medical conditions and diseases.