Diabetic peripheral neuropathy assessment through texture based analysis of corneal nerve images

Mini EURO Conference on Improving Healthcare: new challenges, new approaches

物理学;医药卫生

Silva, Susana F.^1 ; Gouveia, Sofia^2 ; Gomes, Leonor^2 ; Negrão, Luís^3 ; Quadrado, Maria João^1,4 ; Domingues, José Paulo^1,5 ; Morgado, António Miguel^1,5

IBILI - Institute for Biomedical Imaging and Life Sciences, Faculty of Medicine, University of Coimbra, Coimbra, Portugal^1

Department of Endocrinology, Diabetes and Metabolism, Coimbra Hospital University Center, Coimbra, Portugal^2

Department of Neurology, Coimbra Hospital University Center, Coimbra, Portugal^3

Department of Ophthalmology, Coimbra Hospital University Center, Coimbra, Portugal^4

Department of Physics, Faculty of Sciences and Technology, University of Coimbra, Coimbra, Portugal^5

关键词: Classification technique;    Diabetic neuropathy;    Morphometric parameters;    Nerve segmentations;    Noninvasive methods;    Peripheral neuropathy;    Texture analysis;    Texture features;   

Others  :  https://iopscience.iop.org/article/10.1088/1742-6596/616/1/012002/pdf
DOI  :  10.1088/1742-6596/616/1/012002

学科分类：卫生学

来源: IOP

【 摘 要 】

Diabetic peripheral neuropathy (DPN) is one common complication of diabetes. Early diagnosis of DPN often fails due to the non-availability of a simple, reliable, non-invasive method. Several published studies show that corneal confocal microscopy (CCM) can identify small nerve fibre damage and quantify the severity of DPN, using nerve morphometric parameters. Here, we used image texture features, extracted from corneal sub-basal nerve plexus images, obtained in vivo by CCM, to identify DPN patients, using classification techniques. A SVM classifier using image texture features was used to identify (DPN vs. No DPN) DPN patients. The accuracies were 80.6%, when excluding diabetic patients without neuropathy, and 73.5%, when including diabetic patients without diabetic neuropathy jointly with healthy controls. The results suggest that texture analysis might be used as a complementing technique for DPN diagnosis, without requiring nerve segmentation in CCM images. The results also suggest that this technique has enough sensitivity to detect early disorders in the corneal nerves of diabetic patients.