A Local Flow Phase Stretch Transform for Robust Retinal Vessel Detection

Abstract

This paper presents a new method for reliably detecting retinal vessel tree using a local flow phase stretch transform (LF-PST). A local flow evaluator is proposed to increase the local contrast and the coherence of the local orientation of vessel tree. This is achieved by incorporating information about the local structure and direction of vessels, which is estimated by introducing a second curvature moment evaluation matrix (SCMEM). The SCMEM evaluates vessel patterns as only features having linearly coherent curvature. We present an oriented phase stretch transform to capture retinal vessels running at various diameters and directions. The proposed method exploits the phase angle of the transform, which includes structural features of lines and curved patterns. The LF-PST produces several phase maps, in which the vessel structure is characterized along various directions. To produce an orientation invariant response, all phases are linearly combined. The proposed method is tested on the publicly available DRIVE and IOSTAR databases with different imaging modalities and achieves encouraging segmentation results outperforming the state-of-the-art benchmark methods.

Separable Paravector Orientation Tensors for Enhancing Retinal Vessels

“…In Table III, we report the results of the proposed method and benchmark methods on IOSTAR database. The benchmarks are WSF [2], LAD-OS [11], B-COSFIRE [12], the local flow phase stretch transform (LFPST) [15], the isotropic undecimated wavelet filter (IUWF) [19], and the local phase filter (LPF) [14]. The results of B-COSFIRE, IUWF, and LPF on IOSTAR dataset are taken from the published results in [2].…”

“…ACC SP SE Proposed 0.959 0.974 0.777 WSF [2] 0.948 0.967 0.772 LAD-OS [11] 0.951 0.974 0.754 B-COSFIRE [12] 0.941 0.967 0.761 LFPST [15] 0.957 0.975 0.750 IUWF [19] 0.911 0.921 0.726 LPF [14] 0.928 0.935 0.757 achieves a prominent response in distinguishing the closely adjacent vessels, outperforming the benchmark methods. Fig 14 shows the comparison of the proposed method and benchmark methods on a patch image from STARE dataset associated with uneven illumination, which causes background intensity variation, making the intensity values of vessel pixels comparable to that of background pixels.…”

“…Both categories of supervised methods require an intensive training procedure conducted on a large amount of manually annotated labels for achieving vessel segmentation task. The vessel enhancement techniques can be generally categorized into Gaussian-matched responses [6]- [8], mathematical morphology [4], [9], neurophysiology inspired operators [10]- [13], local phase maps [2], [14], [15], multiscale line detectors [16], [17], [52], wavelet transforms [18], [19], tensor methods [20]- [23], and combined ones [24], [53].…”