An Automated Atom Probe Tomography Cluster Detection Approach Using Transfer Learning.

Area of Science:

• Materials Science
• Data Analysis
• Computational Science

Background:

• Atom probe tomography (APT) enables atomic-scale visualization of solute distribution in materials.
• Quantitative cluster analysis in APT is challenging due to subjective parameter selection in algorithms like HDBSCAN.
• This sensitivity hinders reproducibility and accuracy in APT data interpretation.

Purpose of the Study:

• To develop an automated, image-based deep learning workflow for parameter selection and cluster detection in APT data analysis.
• To reduce reliance on manual parameter tuning for improved consistency and scalability.
• To enable faster and more objective analysis of complex spatial patterns in APT datasets.

Main Methods:

• Projecting 3D APT point clouds onto 2D planes for analysis.
• Utilizing pretrained convolutional neural networks (ConvNeXt-Tiny, ResNet-50) with transfer learning.
• Employing deep learning outputs to guide K-means clustering and estimate HDBSCAN parameters (min cluster size, min sample points).

Main Results:

• Successfully automated parameter selection for clustering algorithms in APT data.
• Demonstrated reduced subjectivity and improved consistency in cluster detection.
• Validated the feasibility of using image-based deep learning for complex spatial pattern interpretation.

Conclusions:

• The proposed deep learning-aided workflow significantly enhances the objectivity and efficiency of APT data analysis.
• This method improves the reproducibility and scalability of quantitative cluster analysis in materials science.
• The workflow and code are publicly available to foster further research and application.