Searchable abstracts of presentations at key conferences in endocrinology
Endocrine Abstracts (2011) 26 P248

ECE2011 Poster Presentations Pituitary (111 abstracts)

Detection of acromegaly by automatic face classification software

H J Schneider 1 , R Kosilek 1 , M Günther 2 , J Römmler 1 , G K Stalla 3 , C Sievers 3 , M Reincke 1 , J Schopohl 1 & R Würtz 2


1Medizinische Klinik, Innenstadt, Ludwig Maximilians University, Munich, Germany; 2Institut für Neuroinformatik, Ruhr-Universtät, Bochum, Germany; 3Neuroendocrinology Group, Max Planck Institute of Psychiatry, Munich, Germany.


Acromegaly is accompanied by increased morbidity and mortality. The delay between onset of first symptoms and diagnosis of the disease is 6 to 10 years. Acromegaly causes typical changes of the face. We hypothesized that face classification software might help distinguishing between subjects with and without acromegaly on regular photographs and, thus, might help improving early recognition of acromegaly.

Methods: We took frontal and side photographs of the faces of 57 patients with acromegaly (29 women, 28 men) and of 60 sex- and age-matched controls. We grouped patients into subjects with mild, moderate, and severe facial features of acromegaly by overall impression. We then analyzed all picture by using two principles of similarity analysis: analysis of texture using Gabor jets and analysis of geometry using facial landmarks. We used the leave-one-out cross-validation method to classify subjects by the software. Additionally we asked three acromegaly experts and three general internists to classify all pictures by visual impression.

Findings: The software correctly classified 71.9% of patients and 91.5% of controls, using a combination of texture and geometry analysis. Correct classification rates for patients by visual analysis were 63.2 and 42.1% by experts and general internists, respectively. Correct classification rates for controls were 80.8 and 87.0% by experts and internists, respectively.

Patients with mild facial features of acromegaly (n=24) were correctly classified at 58.3, 38.9 and 20.8% by software, experts, and internists, respectively. Similar classification rates were achieved by the software and experts for patients with moderate or severe features of acromegaly.

Interpretation: Acromegaly can be detected by computer software using photographs of the face. Classification accuracy is higher than by medical experts or general internists. This is particularly the case for patients with mild features of acromegaly.

Article tools

My recent searches

No recent searches.