Facial Analysis in Acromegaly Using Machine Learning: Toward Earlier Diagnosis

Abstract

Context: Acromegaly is a rare and progressive disorder often diagnosed late due to its insidious onset and gradually evolving facial features. Early detection remains a critical unmet need to reduce disease-associated morbidity and mortality.

Objective: This study aimed to develop and evaluate machine learning models that can identify acromegaly-specific facial features using prediagnostic photographs, potentially enabling earlier diagnosis.

Methods: A total of 489 facial photographs from 92 patients with acromegaly and 254 images from 88 controls were analyzed. A 2-stage pipeline was implemented: (1) deep feature extraction using a pretrained VGG-Face model followed by support vector machine (SVM) classification, and (2) an interpretable model using 5 landmark-based facial measurements. Separate data sets were created using prediagnosis, postdiagnosis, and combined images to evaluate model performance.

Results: The best classification results were obtained from the prediagnosis data set (mean 7.47 years before diagnosis), with an area under the curve (AUC) of 0.982 and accuracy of 91.5%. Interpretability analyses highlighted maxillary, nasal, and orbital regions as key facial zones. The interpretable model, using facial ratios, achieved moderate accuracy (AUC = 0.776) while providing clinical insight into contributing features such as face width-to-height ratio and philtrum height.

Conclusion: Our findings demonstrate that acromegaly-related facial features can be detected years before clinical diagnosis using machine learning. By combining high-performance deep models with interpretable approaches, this study supports the potential for artificial intelligence-based facial screening tools to aid in early detection of acromegaly.

Keywords: acromegaly; early diagnosis; facial analysis; machine learning.