Effects of Image Enhancement Techniques on CNNs Based Algorithms for Segmentation of Blood Vessels: A Review

Image quality can either boost or reduce segmentation sensitivity and predictive accuracy of micro vessels from retinal fundus images. The main cause of difficulty in segmentation task particularly in medical imaging such as retinal fundus images, can be majorly attributed to image degradation as a result of noise, low contrast, uneven illumination, obscurity and so on. Image quality degradation leads to loss of information and can affect both the classification and segmentation accuracy of Convolutional Neural Networks (CNNs) as well as diagnosis and analysis. Regardless of how good a CNNs model is, blur and noise can reduce the pixels classification accuracy by exhibiting partial biasness. To this end, this paper aims at reviewing the existing CNNs based method for the segmentation of blood vessels from the retinal fundus image. The review focuses majorly on the image enhancement technique(s) used to address the image quality degradation at pre-processing, processing and postprocessing stages and examines the effects on the CNNs based method performance. The review will aid a critical analysis of the effects of these enhancement techniques, particularly on sensitivity and Area under the ROC curve (AUC) metrics. The percentage of these two metrics are good and dependable yardstick to a successful model in terms of the blood vessels segmentation performance evaluation.