ImageDataExtractor: A Tool To Extract and Quantify Data from Microscopy Images.

Area of Science:

• Materials Science
• Data Science
• Nanotechnology
• Scientific Image Analysis

Background:

• Data-driven materials discovery relies on large datasets of chemical structure and property information.
• Microscopy images are crucial for characterizing nanoparticle size and shape but are often fragmented and qualitative in scientific literature.
• Existing data sources lack the quantitative, large-scale information needed for advanced materials prediction.

Purpose of the Study:

• To develop a toolkit for automated extraction of quantitative data from microscopy images.
• To enable the generation of large datasets of nanoparticle size and shape information.
• To support data-driven materials discovery by providing accessible, analyzed image data.

Main Methods:

• The ImageDataExtractor toolkit automatically identifies and extracts microscopy images from scientific documents.
• It decodes scale bar information using optical character recognition and super-resolution convolutional neural networks.
• Particle detection and profiling are performed using thresholding, segmentation, polygon fitting, and edge correction algorithms.

Main Results:

• The toolkit successfully automates the extraction of quantitative particle size and shape data from microscopy images.
• Evaluation metrics show precision and recall greater than 80% for most image processing steps.
• Precision exceeds 80% for all critical data extraction steps.

Conclusions:

• ImageDataExtractor can generate large-scale, quantitative particle datasets essential for data-driven materials discovery.
• The toolkit's high-throughput capability addresses the need for structured data from fragmented microscopy literature.
• The tool is available under the MIT license, promoting wider use in scientific research.