A tensorflow implementation for Unsupervised vessel segmentation based on ADDA

DILITHIUM: Deep learning with less-to-no supervision for the segmentation of vessels in high-resolution 7T MRAs using unsupervised and weakly-supervised learning

unsupervised u-neural net using morse smale complex topological segmentation at various persistences to learn segmentation of optic vessel images. Contribution is the lack of needed hand segmentation used in the original Unet paper by Ronneberger et. al. via topological training set.

No description available

对论文《DeNVeR: Deformable Neural Vessel Representations for Unsupervised Video Vessel Segmentation》的复现

Retinal diseases are already the most common cause of childhood blindness worldwide. Accordingly, it would be extensively beneficial to populations and health-related communities if we can automate the procedure of diagnosis thoroughly or at least partially by exploiting capabilities of computer-aided diagnosis (CAD). This paper proposes two segmentation methods, a supervised method and an unsupervised one, which shall expertly tackle the problem of vessel segmentation in retinal fundus images. Dataset used in this research is the so-called DRIVE dataset which is public and has been established to enable comparative studies on segmentation of blood vessels, hence containing 2 groups that each group constitutes of 20 colour images for the purpose of train and test. Our supervised method has achieved a higher accuracy of 94.47% by exploiting support-vector-machine technique (SVM) as its intellect, and our unsupervised method has achieved an ample accuracy of 94.28%, with a response time of 1.65 second providing operators and/or systems with fast and reliable results.