Even faster retinal vessel segmentation via accelerated singular value decomposition

Retinal blood vessel segmentation plays a vital role in medical image analysis since the appearance of vessels would contribute in the diagnosis, treatment, and evaluation for various diseases in ophthalmology and other fields, such as cardiology and neurosurgery. Among the state-of-the-art blood vessel segmentation techniques, the Hessian-based multi-scale filter has been widely used and shown its superior performance in the accuracy and visual effect. However, its execution time still remains a challenge due to the employment of eigenvalue decomposition in this approach. Bearing this in mind, we propose an accelerated matrix decomposition mechanism, which could be used to boost not only the original Hessian-based multi-scale approach but also the singular value decomposition-based algorithms. To evaluate the proposed method, we conducted comparison experiments between state-of-the-art techniques and our method. Experimental results show the superior performance of the proposed approach over state of the arts especially in execution time.