Acromegaly facial changes analysis using last generation artificial intelligence methodology: the AcroFace system.

Area of Science:

• Artificial Intelligence in Medicine
• Medical Imaging Analysis
• Endocrinology

Background:

• Acromegaly is a rare endocrine disorder characterized by excessive growth hormone production.
• Early detection of acromegaly is crucial for timely intervention and management of complications.
• Facial features are known indicators of acromegaly, but objective and scalable detection methods are needed.

Purpose of the Study:

• To develop and evaluate the AcroFace system, an AI-based tool for the early detection of acromegaly using facial photographs.
• To explore the efficacy of combining visual/texture features with geometric information from facial images for acromegaly detection.

Main Methods:

• Development of the AcroFace system integrating Support Vector Machines (SVM) for geometric features and Convolutional Neural Networks (CNNs) for visual features.
• Training of various CNN models (ResNet-50, VGG-16, MobileNet, Inception V3, DensNet121, Xception) using expert-annotated facial images.
• Optimization of feature extraction and classification strategies for accurate acromegaly detection.

Main Results:

• The ResNet-50 model combined with Support Vector Regression (SVR) achieved the highest performance, with accuracy values of 75% (δ1) and 89% (δ3).
• Visual features demonstrated superior accuracy compared to geometric features alone.
• The validation cohort achieved high performance metrics: 0.90 precision, 0.93 accuracy, 0.92 F1-Score, 0.93 sensitivity, and 0.93 specificity.

Conclusions:

• The AcroFace system demonstrates robust performance in distinguishing facial traits associated with acromegaly.
• The AI system holds potential as a population-level screening tool for early acromegaly detection.
• Facial photograph analysis using AI offers a promising non-invasive approach for acromegaly screening.