Fuzzy logic based classification of nailfold capillary images in healthy, hypertensive and diabetic subjects

Nailfold Capillaroscopy is a simple, non-invasive clinical diagnostic aid, which is used for purposes of prognosis, therapeutic study and research. The capillary morphological changes are available for Normal, Hypertensive and Diabetes Mellitus Type 2 subjects. Individual capillary-wise features such as inter-capillary distance, capillary width, capillary length and tortuousity, as well as image-wise features such as capillary density and presence of hemorrhages are used to define a set of six abnormality parameters for a given image. Abnormality parameters defined are Rarefaction, Avascularity, Wide capillaries, Elongated capillaries, Tortuosity and Microhemorrhages. These abnormality parameters are graded as Absent, Mild, Moderate and Severe based on the morphological feature value. Finally, the abnormality parameters are combined using Fuzzy Inference System to classify a given image as healthy, hypertensive or diabetic. Overall classification accuracy is 83.3%, with 81.25% for Normal, 64.25% for hypertensive and 95.8% for diabetic.