Semi-quantitative analysis and prediction of nailfold capillary images using response surface regression

Capillary images obtained from the finger nailfolds are very useful for diagnosis of the health condition of a person. Specific capillaroscopic patterns have been identified for healthy controls as well diseases such as Systemic Sclerosis and Raynaud's phenomenon. Here, six capillary morphological abnormalities act as inputs to response surface regression, generating an output whose range will indicate whether the person is Normal, Hypertensive or diabetic. The model is obtained using a training set of 30 images, verified by using another testing set of 30 images. The correlation coefficient of Capillary rarefaction is highest at 0.93 followed by Avascularity at 0.77, Wide capillaries with 0.59, Capillary Elongation with 0.44, Tortuousity with 0.45 and Microhemorrhages with 0.25. A comparative study of first and second order surface regression is carried out. The coefficient of determination R
²
is very high. The overall classification accuracy of 92.22% is obtained from the confusion matrices. Positive predictive index is 88.33% and negative predictive index is 94.16%.