Nasolabial Folds Extraction based on Neural Network for the Quantitative Analysis of Facial Paralysis

Abstract

Nasolabial folds play an important role in diagnosing facial paralysis. The higher the asymmetry of nasolabial folds is, the more severe the facial paralysis will be. But the judgment of asymmetry of nasolabial folds depends on subjective clinical experience of medical experts, and lacks objective quantitative assessment. Moreover, it is difficult for traditional image processing methods to extract nasolabial folds for quantitative computation. In this paper, we propose a method that combines object detection network with semantic segmentation network to extract nasolabial folds. Firstly, Faster-Regions with Convolutional Neural Network is adopted to recognize the nasolabial region from whole face images. Secondly, Global Convolutional Network is adopted to segment nasolabial folds from the nasolabial region. Our method(91%) has outperformed traditional methods(64%). We computed the length, depth and direction of nasolabial folds to evaluate the asymmetry. With further calculation, a quantitative relationship between the asymmetry of nasolabial folds and the severity of facial paralysis is established to help doctors with better diagnosis and follow-up rehabilitation training.