Facial Recognition Intensity in Disease Diagnosis Using Automatic Facial Recognition

Area of Science:

• Medical Diagnostics
• Artificial Intelligence
• Genetics

Background:

• Artificial intelligence (AI) is increasingly used in medicine for disease diagnosis using facial phenotypes.
• Quantitative evaluations of AI performance in facial phenotype-based diagnosis are lacking.

Purpose of the Study:

• To systematically review and quantitatively synthesize studies on AI for diagnosing heterogeneous diseases based on facial features.
• To introduce and evaluate a new index, Facial Recognition Intensity (FRI), for assessing diagnostic complexity.

Main Methods:

• Systematic identification and quantitative analysis of 20 studies involving 12,557 subjects across 7 diseases.
• Meta-regression and subgroup analyses to assess the impact of FRI and other factors on diagnostic accuracy.

Main Results:

• Pooled sensitivity of 89% and specificity of 92% for AI in facial phenotype-based diagnosis.
• Facial Recognition Intensity (FRI) significantly contributed to diagnostic accuracy (p=0.021).
• Increased training size and deep learning models improved accuracy for low-FRI diseases.

Conclusions:

• AI demonstrates significant potential for diagnosing diseases via facial phenotypes.
• The FRI index offers a novel way to understand and predict AI diagnostic performance.
• Further research into universal AI performance rules is warranted.