A New Approach for Evaluating and Improving the Performance of Segmentation Algorithms on Hard-to-Detect Blood Vessels

TL;DR

The paper introduces Local Vessel Salience (LVS) to quantify the difficulty of segmenting individual blood vessel regions and Low-Salience Recall (LSRecall) to assess segmentation accuracy on hard-to-detect vessels. A salience-based augmentation method is proposed to create more challenging samples that improve robustness of CNNs in low-salience regions, while preserving or revealing vascular topology. Evaluation on fluorescence microscopy data shows that high Dice/recall can mask regional failures, and LSRecall provides complementary insight into segmentation quality, with augmentation yielding substantial LSRecall gains for difficult vessels. This framework enables more nuanced comparisons of segmentation methods and their ability to preserve vascular topology, with potential extensions to differentiable LSRecall and explicit topology-focused data generation.

Abstract

Many studies regarding the vasculature of biological tissues involve the segmentation of the blood vessels in a sample followed by the creation of a graph structure to model the vasculature. The graph is then used to extract relevant vascular properties. Small segmentation errors can lead to largely distinct connectivity patterns and a high degree of variability of the extracted properties. Nevertheless, global metrics such as Dice, precision, and recall are commonly applied for measuring the performance of blood vessel segmentation algorithms. These metrics might conceal important information about the accuracy at specific regions of a sample. To tackle this issue, we propose a local vessel salience (LVS) index to quantify the expected difficulty in segmenting specific blood vessel segments. The LVS index is calculated for each vessel pixel by comparing the local intensity of the vessel with the image background around the pixel. The index is then used for defining a new accuracy metric called low-salience recall (LSRecall), which quantifies the performance of segmentation algorithms on blood vessel segments having low salience. The perspective provided by the LVS index is used to define a data augmentation procedure that can be used to improve the segmentation performance of convolutional neural networks. We show that segmentation algorithms having high Dice and recall values can display very low LSRecall values, which reveals systematic errors of these algorithms for vessels having low salience. The proposed data augmentation procedure is able to improve the LSRecall of some samples by as much as 25%. The developed methodology opens up new possibilities for comparing the performance of segmentation algorithms regarding hard-to-detect blood vessels as well as their capabilities for vascular topology preservation.

Figures (9)

• Figure 1: Two fluorescence microscopy images of the mouse cortex showing blood vessels with varying appearances freitas2022unbiased.
• Figure 2: Illustration of the calculation of the LVS index. A blood vessel and the respective medial axis are represented, respectively, in gray and red. For a point $p$ in the medial axis, the two closest contour points $p_{c1}$ and $p_{c2}$ are shown. Pixels belonging to the set $S_v$ are shown in light and dark blue. Background pixels belonging to the set $S_b$ are shown in green.
• Figure 3: Illustration of the parameters involved in the salience augmentation. (a) The central point $p_c$ (red), initial and final points $p_1$ and $p_2$ (blue) and the points $p_{d1}$ and $p_{d2}$ defining the discontinuity region (green) are shown. Parameters $l$ and $l_d$ set the length of the salience augmentation and discontinuity region. (b) Intensity preservation factor along the MAS. Relevant positions are shown with the same colors as the points in (a).
• Figure 4: Illustration of the background identification method for creating a vessel discontinuity. The discontinuity region $S_a$ along the vessel is indicated in dark grey and in blue. The inner and outer contours $C_i$ and $C_o$ of $S_a$ are indicated, respectively, in blue and orange. A candidate background region is indicated in dark grey and as a dashed blue line. The pixel intensities $I_o$ are obtained from $C_o$ and the intensities $I_i$ are obtained from the pixels indicated in the dashed blue line.
• Figure 5: Local vessel salience for some samples of the dataset. The first row of images shows the original samples, the second row shows the LVS values calculated for each vessel pixel. The third row indicates in red regions where the LVS is smaller than 0.2. The regions are indicated only at the blood vessels medial axes for easier comparison with the original samples.