Investigating the Visual Cues of CNNs for Vascular Segmentation: A Case Study in Microscopy and Fundus Imaging

2026-07-25

AI Summary

This study investigates the reliance of convolutional neural networks (CNNs) on visual cues—such as intensity, texture, and shape—in vascular segmentation tasks to enhance model interpretability and reliability. Through controlled experiments involving pixel shuffling, contour sparsification, and receptive field manipulation, the authors systematically evaluate multiple CNN architectures on both fluorescence microscopy and fundus images. They propose a generalizable quantification framework that reveals several counterintuitive findings: intensity cues are more critical than texture, effective receptive fields tend to be relatively small, and models retain high accuracy even when texture or intensity information is removed. Moreover, shape alone proves insufficient for reconstructing complete vascular structures, and global contextual information in fundus images yields only marginal performance gains.

Abstract

Vascular segmentation is a standard procedure for clinical diagnosis, yet the specific visual features determining model decisions remain poorly understood. This paper investigates the visual cues Convolutional Neural Networks (CNNs) use to segment blood vessels across two distinct imaging domains: fluorescence microscopy and retinal fundus photography. We employ a series of experiments to quantify the influence of shape, texture, and receptive field on segmentation performance. First, we isolate texture and intensity by evaluating performance on patches subjected to pixel shuffling and normalization. Second, we assess global shape relevance by training models on sparse contours and centerlines. Lastly, we quantify the required spatial context by systematically varying the network's theoretical and effective receptive fields. Within the scope of the evaluated datasets, we found that pixel intensity is more relevant than texture, though networks maintain surprisingly high accuracy even when both cues are removed. Furthermore, CNNs struggle to extrapolate full vessel geometry from shape cues alone, typically relying on a relatively small effective receptive field of around 20 pixels, though global context provides a modest benefit for fundus images. While specific to the modalities studied, this methodology offers a quantitative foundation to audit and refine deep learning systems in vascular imaging.

Problem

• vascular segmentation
• visual cues
• convolutional neural networks
• interpretability
• medical imaging

Innovation

• visual cues
• vascular segmentation
• receptive field
• CNN interpretability
• texture vs. intensity