On machine learning analysis of atomic force microscopy images for image classification, sample surface recognition.

Area of Science:

• Surface Science and Nanotechnology
• Biophysics and Computational Biology
• Materials Science and Engineering

Background:

• Atomic Force Microscopy (AFM) generates multidimensional datasets of surface physicochemical properties.
• Traditional analysis of AFM data is challenging due to its complexity and high dimensionality.
• Machine Learning (ML) offers a promising avenue for analyzing AFM data, but deep learning methods often require large datasets.

Purpose of the Study:

• To explore and present machine learning (ML) methods suitable for analyzing atomic force microscopy (AFM) data, particularly when dealing with small databases.
• To demonstrate the application of these ML methods for classification and recognition tasks using AFM imaging.
• To provide a general framework for ML analysis tailored to AFM data, emphasizing statistical significance.

Main Methods:

• Utilized machine learning (ML) algorithms, focusing on methods beyond deep learning neural networks, for the analysis of atomic force microscopy (AFM) image data.
• Developed and applied a general template for ML analysis specific to AFM data.
• Included a focus on assessing the statistical significance of the obtained results, with a simple method described for this purpose.

Main Results:

• Successfully applied ML methods to analyze and classify the surfaces of biological cells, demonstrating effectiveness with limited AFM image data.
• Validated the potential for ML-driven AFM analysis in diverse fields including medical imaging, materials processing, forensic science, and art authentication.
• Provided a practical example of identifying cell phenotypes using the proposed ML approach.

Conclusions:

• Machine learning (ML) provides a powerful and accessible tool for analyzing complex, multidimensional data from atomic force microscopy (AFM), even with small datasets.
• The presented ML approach offers broad applicability across various scientific and technical domains requiring surface characterization and classification.
• Emphasizing statistical significance in ML analysis of AFM data is crucial for reliable and interpretable results.