Towards Protocol-Adaptive, Small Dataset-Based Deep Learning Techniques for Upper-Airway Segmentation in Magnetic Resonance Imaging