Deep Learning Based Enhancement and Detection of Capillaries from Nailfold Capillaroscopy Images

Nailfold Capillaroscopy (NFC) images facilitate the visualization of capillaries present in the nailfold. This technique is utilized to diagnose various microvascular conditions based on capillary morphology. Medical image processing, in conjunction with deep learning applications, plays a significant role in detecting regions of interest from images for diagnosing various diseases. The capillary images acquired from NFC are of poor quality and low contrast, thus necessitating preprocessing techniques for enhancing the visibility of capillaries. This work focuses on the development of capillary enhancement and detection methods using deep learning. In preprocessing, bilateral filter, median filter, and Wiener filter were applied for image smoothing. Super-resolution-CNN and Enhanced CNN were employed to enhance the capillaries, and for capillary detection, YOLO was applied to the NFC image. The applied techniques were evaluated using performance metrics, and analysis revealed that the median filter demonstrated superior performance with a PSNR value of 37.69, MSE of 0.0001, and SSIM of 0.99. Capillaries were enhanced more effectively with Enhanced Convolutional Neural Network (CNN), yielding a PSNR of 16.54 and MSE of 0.0121. Furthermore, this work utilized the YOLO V8 model to automatically detect capillaries from the nailfold capillaroscopy image.