Region Based Parallel Hierarchy Convolutional Neural Network for Automatic Facial Nerve Paralysis Evaluation.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neurology

Background:

• Facial nerve paralysis (FNP), including Bell's palsy, is a common neuromotor dysfunction affecting facial muscle movement and expression.
• Current FNP grading relies on subjective clinical judgment, leading to inconsistent evaluations.
• Existing computer-aided methods often require complex and expensive imaging equipment, hindering accessibility for rehabilitation.

Purpose of the Study:

• To develop an objective and intelligent method for quantitative grading of facial nerve paralysis (FNP).
• To overcome the limitations of subjective clinical assessment and complex imaging techniques in FNP evaluation.
• To improve diagnostic accuracy and provide detailed paralysis analysis using deep learning.

Main Methods:

• Proposed a parallel hierarchy convolutional neural network (PHCNN) integrated with a Long Short-Term Memory (LSTM) network.
• The model analyzes region-based asymmetric facial features and temporal variations in image sequences.
• Facial segmentation into two palsy regions was employed to enhance feature learning and reduce confounding factors like age wrinkles.

Main Results:

• The PHCNN-LSTM model demonstrated superior performance in quantitatively assessing FNP.
• The method accurately discriminates FNP from normal faces, reducing the impact of variations in facial features.
• Experiments on benchmark datasets (YouTube Facial Palsy Database, Extended Cohn-Kanade Database) showed the method outperforms state-of-the-art deep learning approaches.

Conclusions:

• The proposed PHCNN-LSTM network offers an effective and accurate solution for objective FNP grading.
• This intelligent approach has the potential to significantly improve facial functional rehabilitation assessment.
• The method provides a more reliable and accessible alternative to subjective evaluations and complex imaging systems.