Scale and Curvature Invariant Ridge Detector for Tortuous and Fragmented Structures

Abstract

Abstract. Segmenting dendritic trees and corneal nerve fibres is challenging due to their uneven and irregular appearance in the respective image modalities. State-of-the-art approaches use hand-crafted features based on local assumptions that are often violated by tortuous and pointlike structures, e.g., straight tubular shape. We propose a novel ridge detector, SCIRD, which is simultaneously rotation, scale and curvature invariant, and relaxes shape assumptions to achieve enhancement of target image structures. Experimental results on three datasets show that our approach outperforms state-of-the-art hand-crafted methods on tortuous and point-like structures, especially when captured at low resolution or limited signal-to-noise ratio and in the presence of other non-target structures.

Accelerating Convolutional Sparse Coding for Curvilinear Structures Segmentation by Refining SCIRD-TS Filter Banks

“…As a consequence, detection performance may degrade significantly. We addressed these modelling issues in our previous work [13], [23] by proposing a novel ridge detector, SCIRD, which adds curvature and contrast invariance to that of previous HCFs (i.e., scale, rotation and elongation).…”

“…Most of the existing methods for automated curvilinear structure segmentation rely on hand-crafted filters (henceforth HCFs) designed to model local geometric properties of ideal tubular shapes [1], [4], [7], [13], [23]. However, HCFs often require manual parameter tuning (e.g.…”

“…However, HCFs often require manual parameter tuning (e.g. width, length and orientation) [1], [4], [7], [13], which does not guarantee optimal performance. Today research is moving towards fully/deep learning architectures (henceforth, DLA) given their excellent results on several challenging tasks [25]- [29].…”

“…For the modelling part, we propose a new formulation of the recent SCIRD ridge detector [13], [23], denoted as SCIRD-TS, which improves the detection of thin structures. Moreover, we formulate the problem of identifying the optimal set of SCIRD-TS filters to be used to initialise the CSC optimisation as a compression task, and motivate the adoption of a K-means algorithm to perform this task.…”

Vessel Extraction for AS-OCT Angiography

“…1 (B). After that, the Scale and Curvature Invariant Ridge Detector (SCIRD) [3] is employed to enhance the vessel map. SCIRD is simultaneously rotation, scale and curvature invariant, and removes the assumption of locally straight tubular structures based on the curved-support Gaussian model [5].…”