Searchable abstracts of presentations at key conferences in endocrinology
Endocrine Abstracts (2016) 41 EP871 | DOI: 10.1530/endoabs.41.EP871

ECE2016 Eposter Presentations Pituitary - Clinical (83 abstracts)

Computer vision technology in the diagnosis of Cushing’s syndrome – advanced studies with a cohort matched by body mass index

Kathrin H. Popp 1, , Robert P. Kosilek 2 , Günter K. Stalla 1 , Mareike Stieg 1 , Christina M. Berr 2 , Martin Reincke 2 , Matthias Witt 2 , Rolf P. Würtz 3 & Harald J. Schneider 2


1Max-Planck Institut für Psychiatrie, Munich, Germany; 2Medizinische Klinik und Poliklinik IV, Munich, Germany; 3Ruhr-Universität, Bochum, Germany.


Introduction: Cushing’s syndrome (CS) is a rare disease characterized by clinical features that show overlap with the ‘metabolic syndrome’. Pilot studies regarding the use facial image analysis software as a novel diagnostic tool in acromegaly and CS have shown promising results. Distinguishing CS patients from patients that show similar features without true hypercortisolism remains a challenge in clinical practice. To address this particular problem, we evaluated the performance of this tool for CS with a larger cohort and included controls matched by BMI.

Methods: Eighty two (22 m., 60 f.) study subjects with confirmed CS and 98 (32 m., 66 f.) control subjects matched by age, gender and BMI were included. For the control group we screened patients with typical clinical signs (metabolic syndrome) but biochemically excluded CS. Standardized frontal and profile facial photographs were acquired using a regular digital camera. The images were analyzed with the software tool FIDA (facial image diagnostic aid). A grid of nodes was semi-automatically placed on relevant facial structures. Classification was done using a combination of Gabor wavelet transformation and geometrical analysis and a maximum likelihood classifier. Classification accuracy was calculated using a leave-one-one cross-validation procedure.

Results: The mean BMI and age of the study cohort, stratified by gender, did not differ significantly (t-test, P> 0.05). The correct classification rates were: 10/22 (45%) for male patients and 26/32 (81%) for male controls, 34/60 (57%) for female patients and 43/66 (65%) for female controls. The classification rates by etiology of CS were: Adrenal CS 3/8 (m), 7/13 (f); Cushing’s disease 4/8 (m), 18/37 (f) and Iatrogenic CS 2/3 (m), 8/9 (f).

Conclusion: Regarding the advanced problem of detecting CS patients within a BMI-matched cohort, we have found a satisfying classification accuracy by facial image analysis. Classification accuracy would most likely be significantly higher in a study cohort with healthy control subjects. Further studies might pursue a different combination of nodes and equations in the analysis for improving the method.

Article tools

My recent searches

No recent searches.