A Deep Learning Approach for Predicting Subject-Specific Human Skull Shape from Head Toward a Decision Support System for Home-Based Facial Rehabilitation

Prediction of human skull shape from head is a complex and challenging engineering task for the development computer-aided vision system. Skull-to-face generation has been commonly performed in forensic facial reconstruction. Classical statistical approaches were usually used. However, head-to-skull relationship still misunderstood. Recently, novel deep learning (DL) models have showed their efficiency robustness large range applications. The present study aimed to develop approach based on reconstruct head. A workflow was developed evaluated. database computed tomography (CT) images 209 subjects established training testing purposes. Three-dimension (3-D) geometries reconstructed then respective descriptors (head/skull volumes, sampling feature points point-to-center distances, head-skull thickness, Gaussian curvatures) extracted. Two (regression neural network long-short term memory (LSTM)) implemented evaluated with different configurations. 10-fold cross-validation performed. Finally, best worst predicted cases analyzed discussed. mean errors better accuracy level regression model according long short-term model. error between DL-predicted shapes CT-based ranges 1.67 mm 3.99 by using configuration. volume deviation smaller than 5%. suggested that allows be given good accuracy. This opens new avenues rapid visual sensors (e.g. Microsoft Kinect) toward system mimic rehabilitation. As perspectives, muscle will incorporated into workflow. Then, movements tracked animated evaluate optimize rehabilitation exercises. • Predicting subject-specific automatically shape. Feature relationship. 3D geometries.