New Machine-Learning-Driven Opportunities for AFM Data Acquisition and Analysis
See how atomic force microscopy experts are using machine automation and AI to make AFM an even more powerful tool.
In this four-part mini-symposium, our speakers discuss their work and the latest advances using machine learning to automate AFM data acquisition and analysis. Their work underscores the potential of these tools to ensure the absence of artifacts and deliver new insights into AFM data.
Topics discussed include:
- Selecting and examining regions of interest within a sample through automated experimentation;
- Applying deep learning to correlate nanoscale mechanical property maps with bulk properties;
- Identifying artifacts in AFM images via machine learning and computer vision approaches; and
- Developing an open-source automated pipeline for analysis of AFM image data and classification of molecular structures.
Presenter's Abstracts
Deep Learning for Live Cell Shape Detection and Automated AFM Navigation
Presented by: Juan (Djuenne) Ren, Ph.D., Associate Professor, Department of Mechanical Engineering, Iowa State University
Atomic force microscopy (AFM) provides a platform for high-resolution topographical imaging and the mechanical characterization of a wide range of samples, including live cells, proteins, and other biomolecules. However, performing force measurements and high-resolution imaging with AFM and data analytics are time-consuming and require special skill sets and continuous human supervision. Recently, researchers have explored the applications of artificial intelligence (AI) and deep learning (DL) in the bio-imaging field. However, the applications of AI to AFM operations for live-cell characterization are little-known. In this talk, we will introduce our recently developed DL framework for automatic sample selection based on the cell shape for AFM navigation during AFM biomechanical mapping. We also established a closed-loop scanner trajectory control for measuring multiple cell samples at high speed for automated navigation. With this, we achieved a 60× speed-up in AFM navigation and reduced the time involved in searching for the particular cell shape in a large sample. Our innovation directly applies to many bio-AFM applications with AI-guided intelligent automation through image data analysis together with smart navigation.
Find out more about the technology featured in this webinar or our other AFM solutions:
Machine Learning Tools to Classify, Predict Structure-Property Relationships, and Detect Artifacts in AFM Images
Presented by: Dalia Yablon, Ph.D., SurfaceChar LLC
Machine learning techniques enhance acquisition and analysis of various microscopy-based techniques. We explore its role here in 3 distinct SPM applications: image classification, predicting structure-property relationships, and artifact detection. A series of polymer blends that varied in microstructure and bulk mechanical properties were imaged with two different AFM modes: qualitative phase imaging mode as well as quantitative peak force QNM. Supervised machine learning was then used to analyze the various datasets to classify the image by sample as well as correlate the various sample images with bulk mechanical properties; the peak force QNM data had the best accuracy for the classification application, while the phase data was best at the correlation study. Finally, both ML and other computer vision approaches were used to detect artifacts in PFQNM data.
TopoStats - An Open, Automated Analysis Pipeline for AFM Image Processing
Presented by: Alice Pyne, Ph.D., UKRI Future Leaders Fellow & Senior Lecturer, Department of Materials Science and Engineering, University of Sheffield, UK
Atomic Force Microscopy (AFM) imaging is a powerful tool to study the surface topography and physical properties of materials including biological samples, polymers, and semiconductors at the nanoscale. However, the increasing amounts of AFM data generated coupled with a lack of automated analysis tools, and slow integration of machine learning (ML) pipelines limits the quantitative analysis of this unique and powerful data. Limiting factors for the design and integration of automated analysis tools in AFM include specific issues with processing of raw data and small datasets compared to e.g. Cryo EM.
We have developed TopoStats, a high-throughput, open-source Python package that automates the analysis of AFM image datasets, producing processed, clean AFM images, and extracting quantitative information. TopoStats is designed to reduce the burden of user oversight including the processing of single images one-at-a-time for downstream analysis. TopoStats reads and analyses raw AFM data, automating image segmentation, analysis of surface features, and extracting statistical information from complex datasets using traditional and machine learning algorithms. TopoStats performs a range of processing steps including image filtering and feature extraction to output processed, flattened data, including images, binary and labeled masks, and visualises them via a range of histograms and other statistical plots. This allows users to examine the statistical properties of images and extracted features, including surface roughness and topography and view these via automated data visualisation. TopoStats provides an efficient, robust and reliable way to analyse AFM images, saving significant time and effort compared to manual analysis methods and making data analysis in AFM more open, increasing reproducibility across the field.
Automating AFM Data Acquisition and Analysis with the Dimension Icon and FastScan
Presented by: Bede Pittenger, Ph.D., AFM Senior Staff Development Scientist, Bruker Nano Surfaces
Atomic Force Microscopy (AFM) is a powerful tool for characterizing the structure and properties of materials at the nanoscale, but collecting and analyzing AFM data is often time consuming, making it costly to acquire the volume of data needed to develop machine learning models.
In this presentation, we examine several methods to reduce the amount of user time needed for acquisition and analysis of AFM data with the Dimension platform. We discuss methods for automating the acquisition, mode and probe selection to maximize speed, and best practices for maintaining accuracy and avoiding artifacts. Finally, we consider methods to import the resulting AFM data into Python, allowing automation of analysis and use with popular libraries for machine learning like Scikit-learn, TensorFlow, and PyTorch.
Featured Products and Technology
Speakers
Juan (Djuenne) Ren, PhD., Associate Professor, Department of Mechanical Engineering, Iowa State University
Dalia Yablon, Ph.D., SurfaceChar LLC
Alice Pyne, Ph.D., UKRI Future Leaders Fellow & Senior Lecturer, Department of Materials Science and Engineering, University of Sheffield, UK
Bede Pittenger, Ph.D., Sr. Staff Development Scientist, AFM Applications, Bruker Nano Surfaces
