A Novel Approach to Classify Nailfold Capillary Images in Indian Population Using USB Digital Microscope

Nailfold capillaroscopy is a simple, non-invasive clinical aid used for the diagnosis of vascular dysfunction. The method proposed emphasizes on the USB Digital Microscope which is a cheaper and useful alternative to commercially available expensive videocapillaroscopes which produces high quality images. As the process of identifying anomalies in nailfold capillaries is a tedious and time-consuming process this article proposes a completely automated system to diagnose anomalies. The nailfold capillary images are pre-processed to highlight the important features and remove noise in the images. The processed images are then used to train machine learning models. This article then assays and compares the performance of Logistic Regression classifier, fully connected neural network, Convolutional Neural Network (CNN) and Random Forest classifiers by evaluating their classification accuracy, sensitivity and specificity. The results prove Logistic Regression to be most accurate with a low classification error rate of 10.64%. while, a substantial classification accuracy of 72% was obtained with a small dataset by using bottleneck features of a deep CNN.