Understanding the Influence of Receptive Field and Network Complexity in Neural-Network-Guided TEM Image Analysis

TL;DR

This study systematically investigates how neural network architecture, specifically receptive field and complexity, impacts segmentation performance in TEM image analysis, providing guidance for customizing models based on image resolution and contrast features.

Contribution

It clarifies the roles of receptive field and network complexity in TEM image segmentation, offering practical insights for optimizing neural networks for different TEM imaging conditions.

Findings

Receptive field has minimal impact on low-resolution TEM segmentation.

Receptive field is crucial for high-resolution TEM with phase contrast.

Network complexity effects vary with image resolution and contrast type.

Abstract

Trained neural networks are promising tools to analyze the ever-increasing amount of scientific image data, but it is unclear how to best customize these networks for the unique features in transmission electron micrographs. Here, we systematically examine how neural network architecture choices affect how neural networks segment, or pixel-wise separate, crystalline nanoparticles from amorphous background in transmission electron microscopy (TEM) images. We focus on decoupling the influence of receptive field, or the area of the input image that contributes to the output decision, from network complexity, which dictates the number of trainable parameters. We find that for low-resolution TEM images which rely on amplitude contrast to distinguish nanoparticles from background, the receptive field does not significantly influence segmentation performance. On the other hand, for high-resolution TEM images which rely on a combination of amplitude and phase contrast changes to identify nanoparticles, receptive field is a key parameter for increased performance, especially in images with minimal amplitude contrast. Our results provide insight and guidance as to how to adapt neural networks for applications with TEM datasets.