Development and Validation of a Deep Learning-Based Facial Weakness Score for Objective Assessment in...

Area of Science:

• Neurology
• Biomedical Engineering
• Medical Imaging

Background:

• Facioscapulohumeral muscular dystrophy (FSHD) causes significant facial weakness, impacting communication and quality of life.
• Current manual assessments of facial weakness in FSHD are subjective and lack inter-observer reliability.
• Objective clinical outcome measures for facial weakness in FSHD are critically needed.

Purpose of the Study:

• To develop and validate a deep learning-based facial weakness score (DLFWS) for objective assessment in FSHD.
• To establish DLFWS as a reliable clinical outcome measure for facial weakness.
• To explore the utility of DLFWS in tracking disease progression over time.

Main Methods:

• A convolutional neural network analyzed video recordings of 122 FSHD patients and 56 controls performing facial exercises.
• The DLFWS was trained by comparing video frames with manual facial weakness scores (MFWS) from experienced observers.
• Performance was evaluated using Pearson correlation and intraclass correlation coefficients (ICC), with 64 patients followed for 5 years.

Main Results:

• DLFWS strongly correlated with MFWS (mean Pearson r=0.79, max=0.85).
• The DLFWS demonstrated excellent test-retest reliability (ICC=0.90).
• No significant facial weakness progression was observed over the 5-year follow-up period.

Conclusions:

• The DLFWS offers a reliable and objective method for assessing facial weakness in FSHD.
• This automated tool has potential for standardized clinical and research applications in FSHD.
• DLFWS can facilitate more consistent evaluation of facial muscle function in this population.