JoVE - Journal of Visualized Experiments
JoVE Visualize
Contact Us

Facial Analysis in Acromegaly Using Machine Learning: Towards Earlier Diagnosis

  • 0Division of Endocrinology and Metabolic Diseases, Department of Internal Medicine, Cerrahpasa Faculty of Medicine, Istanbul University-Cerrahpasa, Istanbul, Türkiye.
The Journal of Clinical Endocrinology and Metabolism +

|

August 21, 2025

|

August 21, 2025

View abstract on PubMed

Summary

This summary is machine-generated.

Machine learning models can detect acromegaly facial features in photographs years before diagnosis, aiding early detection. This AI approach uses facial analysis to identify the rare disorder, improving patient outcomes.

Area Of Science

  • Medical imaging analysis
  • Artificial intelligence in healthcare
  • Endocrinology

Background

  • Acromegaly is a rare, progressive disorder with subtle, evolving facial features leading to late diagnosis.
  • Early detection of acromegaly is crucial for reducing morbidity and mortality.

Purpose Of The Study

  • To develop and assess machine learning (ML) models for identifying acromegaly-specific facial characteristics in pre-diagnostic photographs.
  • To enable earlier diagnosis of acromegaly through AI-powered facial screening.

Main Methods

  • Analysis of 489 facial photographs from acromegaly patients and 254 from healthy controls.
  • A two-stage pipeline involving deep feature extraction (VGG-Face) with SVM classification and an interpretable landmark-based facial measurement model.
  • Evaluation of models using pre-diagnosis, post-diagnosis, and combined image datasets.

Main Results

  • The best performance was achieved using pre-diagnosis images (7.47 years prior), yielding an AUC of 0.982 and 91.5% accuracy.
  • Key facial regions identified include maxillary, nasal, and orbital areas.
  • An interpretable model using facial ratios showed moderate accuracy (AUC 0.776), highlighting features like face width-to-height ratio.

Conclusions

  • Machine learning can detect acromegaly facial features years before clinical diagnosis.
  • Combining deep learning with interpretable AI approaches shows promise for early acromegaly detection.
  • AI-based facial screening tools could significantly aid in the early identification of acromegaly.

Keywords: