Augmenting Vascular Disease Diagnosis by Vasculature-awareUnsupervised Learning

Abstract

Vascular diseases are among the leading causes of death and threaten human health worldwide. Imaging examination of vascular pathology with reduced invasiveness is challenging due to the intrinsic vasculature complexity and the non-uniform scattering from bio-tissues. Here, we report VasNet, a vasculature-aware unsupervised learning algorithm that augments pathovascular recognition from small sets of unlabeled fluorescence and digital subtraction angiography (DSA) images. The VasNet adopts the multi-scale fusion strategy with a domain adversarial neural network (DANN) loss function that induces biased pattern reconstruction, by strengthening the features relevant to the retinal vasculature reference while weakening the irrelevant features. VasNet delivers outputs of "Structure + X", where X refers to multi-dimensional features such as blood flows, the distinguishment of blood dilation and its suspicious counterparts, and the dependence of new pattern emergence on a disease progression, which may assist the discovery of novel diagnostics. Therefore, explainable imaging output from VasNet and other algorithm extensions hold the promise to revolutionize the practice of medical diagnosis, as it improves performance while reduces the cost on human expertise, equipment exquisite and time consumption.