Enhanced head-skull shape learning using statistical modeling and topological features.

Skull prediction from the head is a challenging issue toward a cost-effective therapeutic solution for facial disorders. This issue was initially studied in our previous work using full head-to-skull relationship learning. However, the head-skull thickness topology is locally shaped, especially in the face region. Thus, the objective of the present study was to enhance our head-to-skull prediction problem by using local topological features for training and predicting. Head and skull feature points were sampled on 329 head and skull models from computed tomography (CT) images. These feature points were classified into the back and facial topologies. Head-to-skull relations were trained using the partial least square regression (PLSR) models separately in the two topologies. A hyperparameter tuning process was also conducted for selecting optimal parameters for each training model. Thus, a new skull could be generated so that its shape was statistically fitted with the target head. Mean errors of the predicted skulls using the topology-based learning method were better than those using the non-topology-based learning method. After tenfold cross-validation, the mean error was enhanced 36.96% for the skull shapes and 14.17% for the skull models. Mean error in the facial skull region was especially improved with 4.98%. The mean errors were also improved 11.71% and 25.74% in the muscle attachment regions and the back skull regions respectively. Moreover, using the enhanced learning strategy, the errors (mean ± SD) for the best and worst prediction cases are from 1.1994 ± 1.1225 mm (median: 0.9036, coefficient of multiple determination (R2): 0.997274) to 3.6972 ± 2.4118 mm (median: 3.9089, R2: 0.999614) and from 2.0172 ± 2.0454 mm (median: 1.2999, R2: 0.995959) to 4.0227 ± 2.6098 mm (median: 3.9998, R2: 0.998577) for the predicted skull shapes and the predicted skull models respectively. This present study showed that more detailed information on the head-skull shape leads to a better accuracy level for the skull prediction from the head. In particular, local topological features on the back and face regions of interest should be considered toward a better learning strategy for the head-to-skull prediction problem. In perspective, this enhanced learning strategy was used to update our developed clinical decision support system for facial disorders. Furthermore, a new class of learning methods, called geometric deep learning will be studied.

A statistical shape modeling approach for predicting subject-specific human skull from head surface.

Human skull is an important body structure for jaw movement and facial mimic simulations. Surface head can be reconstructed using 3D scanners in a straightforward way. However, internal skull is challenging to be generated when only external information is available. Very few studies in the literature focused on the skull generation from outside head information, especially in a subject-specific manner with a complete skull. Consequently, this present study proposes a novel process for predicting a subject-specific skull with full details from a given head surface using a statistical shape modeling approach. Partial least squared regression (PLSR)-based method was used. A CT image database of 209 subjects (genders-160 males and 49 females; ages-34-88 years) was used for learning head-to-skull relationship. Heads and skulls were reconstructed from CT images to extract head/skull feature points, head/skull feature distances, head-skull thickness, and head/skull volume descriptors for the learning process. A hyperparameter turning process was performed to determine the optimal numbers of head/skull feature points, PLSR components, deformation control points, and appropriate learning strategies for our learning problem. Two learning strategies (point-to-thickness with/without volume descriptor and distance-to-thickness with/without volume descriptor) were proposed. Moreover, a 10-fold cross-validation procedure was conducted to evaluate the accuracy of the proposed learning strategies. Finally, the best and worst reconstructed skulls were analyzed based on the best learning strategy with its optimal parameters. The optimal number of head/skull feature points and deformation control points are 2300 and 1300 points, respectively. The optimal number of PLSR components ranges from 4 to 8 for all learning configurations. Cross-validation showed that grand means and standard deviations of the point-to-thickness, point-to-thickness with volumes, distance-to-thickness, and distance-to-thickness with volumes learning configurations are 2.48 ± 0.27 mm, 2.46 ± 0.19 mm, 2.46 ± 0.15 mm, and 2.48 ± 0.22 mm, respectively. Thus, the distance-to-thickness is the best learning configuration for our head-to-skull prediction problem. Moreover, the mean Hausdorff distances are 2.09 ± 0.15 mm and 2.64 ± 0.26 mm for the best and worst predicted skull, respectively. A novel head-to-skull prediction process based on the PLSR method was developed and evaluated. This process allows, for the first time, predicting 3D subject-specific human skulls from head surface information with a very good accuracy level. As perspective, the proposed head-to-skull prediction process will be integrated into our real-time computer-aided vision system for facial animation and rehabilitation. Graphical abstract.