Correlative light and electron microscopy of poly(ʟ-lactic acid) spherulites for fast morphological measurements using a convolutional neural network. (22nd December 2021)

• Record Type: Journal Article
• Authors: Konyuba, Yuji; Marubayashi, Hironori; Haruta, Tomohiro; Jinnai, Hiroshi
• Is Part Of: Microscopy. Volume 71:Number 2(2022)
• Page Start: 104
• Page End: 110
• Publication Date: 2021-12-22
• Subjects: poly(ʟ-lactic acid) -- correlative light and electron microscopy -- transmission electron microscopy -- polarized optical microscopy -- spherulite -- convolutional neural network; Microscopy -- Periodicals; 502.825
• DOI: 10.1093/jmicro/dfab058
• Languages: English
• ISSNs: 2050-5698

Abstract

Abstract: Polarized optical microscopy (POM) and transmission electron microscopy (TEM) are widely used for imaging polymer spherulite structures. TEM provides nanometer resolution to image small spherulites of sub-micrometer in diameter, while POM is more suitable for observing large spherulites. However, high-resolution images with a large field of view are challenging to achieve due to the deformations of ultrathin sectioned samples used for TEM observations. In this study, we demonstrated that correlative light and electron microscopy (CLEM) combining POM and TEM could effectively characterize the spherulite structures in a wide range from nanometer to several hundred micrometers that neither TEM nor POM alone could cover. Furthermore, the deformations of the TEM ultrathin sections were corrected by referencing to the POM images at the same position of the sample, and large-area TEM images without deformations were successfully produced. The spherulite structures of poly(ʟ-lactic acid) were successfully analyzed using CLEM and TEM in a wide range of spatial scales at the same field of view. The large-area TEM image (250 µm × 250 µm), consisting of 702 TEM images stitched together, was subjected to machine learning to extract the essential structural information of spherulites. Analysis using the convolutional neural network, a well-known algorithm You Only Look Once (YOLO), demonstrated that it was practical to accurately obtain the diameter distribution and the space-filling factor (relative crystallinity) of the spherulites. This study presents a new approach for acquiring high-resolution images with a large field of view and processing the images at a fast speed.