RESUMEN
Botulinum toxin injection is a highly efficacious treatment for masseter muscle hypertrophy, while manual injection point localization based on experience can be non-quantitative, subjective and therefore suboptimal and even side-effect risky. To address this important while challenging task, in this paper we present a methodology of automatic ideal point localization for botulinum toxin injection based on automatic segmentation, measurement and quantitative analysis. Specifically, we first present a novel three-dimensional (3D) fully convolutional neural network for fully automatic mandible and masseter regions of interest (RoI) localization and segmentation from head computed tomography (CT) images. Given the segmentation results, the ideal injection points on the face are located using ray casting based automatic thickness measurement. We conducted experiments on an internal dataset consisting of head CT images acquired from 53 patients to evaluate the segmentation performance and localization reliability. The results demonstrate that the segmentation framework outperforms the state-of-the-art method by a significant margin, and the localization system provides intuitive, interactive user interface and reliable injection point decisions.