A pipeline of unsupervised image analysis methods for extraction of geometrical features from retinal fundusimages has previously been developed. Features related to vessel caliber, tortuosity and bifurcations, have beenidentified as potential biomarkers for a variety of diseases, including diabetes and Alzheimer's. The currentcomputationally expensive pipeline takes 24 minutes to process a single image, which impedes implementationin a screening setting. In this work, we approximate the pipeline with a convolutional neural network (CNN)that enables processing of a single image in a few seconds. As an additional benefit, the trained CNN is sensitiveto key structures in the retina and can be used as a pretrained network for related disease classification tasks.Our model is based on the ResNet-50 architecture and outputs four biomarkers that describe global propertiesof the vascular tree in retinal fundus images. Intraclass correlation coefficients between the predictions of theCNN and the results of the pipeline showed strong agreement (0.86 - 0.91) for three of four biomarkers andmoderate agreement (0.42) for one biomarker. Class activation maps were created to illustrate the attentionof the network. The maps show qualitatively that the activations of the network overlap with the biomarkersof interest, and that the network is able to distinguish venules from arterioles. Moreover, local high and lowtortuous regions are clearly identified, confirming that a CNN is sensitive to key structures in the retina.
展开▼
