Constructing an automatic diagnosis and severity-classification model for acromegaly using facial photographs by deep...

Area of Science:

• Artificial Intelligence
• Medical Imaging
• Endocrinology

Background:

• Acromegaly diagnosis is often delayed due to insidious onset, leading to severe complications.
• There is a critical need for convenient and effective acromegaly screening methods.

Discussion:

• A novel deep learning model was developed for automatic acromegaly diagnosis and severity classification using facial photographs.
• The model was trained on 2148 images with severity scores ranging from 1 to 3.
• Performance evaluation demonstrated a 90.7% prediction accuracy on an internal test dataset.

Key Insights:

• The developed AI model achieved higher accuracy (90.7%) than ten junior internal medicine physicians (89.0%) in diagnosing acromegaly.
• Facial photographs serve as a viable data source for AI-driven acromegaly assessment.
• The model's accuracy suggests its potential as a valuable clinical screening tool.

Outlook:

• The model's high accuracy and potential health economic benefits indicate a promising future for clinical application.
• Further validation and integration into healthcare systems could significantly improve acromegaly detection rates.
• This AI approach may facilitate earlier intervention and better patient outcomes for acromegaly.