| Abstract | A semi-automatic image analysis program, SMART, was used to analyze transmission electron microscopy (TEM) images from four laboratories that participated in an interlaboratory comparison study by Meija et al. on CNC particle size measurement by TEM using conventional manual image analysis approaches. Detailed image-to-image comparisons found that the percentage of “correctly” identified CNCs by SMART was 58% to 78%, while manual was 70% to 87%, depending on TEM image quality from a given laboratory. SMART was able to parameterize image quality, and it was found that SMART had difficulties in CNC identification for images with a combination of higher noise, lower contrast, and higher CNC density. Overall, the SMART image analysis was consistent with the manual approach. SMART showed lower laboratory-laboratory variation as compared to manual, suggesting that the variability of analyst bias of manual approach was removed and demonstrates an opportunity with SMART to improve the standardization of CNC size characterization. An approach to estimate the likelihood of reaching a representative measurement for CNC particle size was developed. SMART area analysis found that less than 10% of CNCs were used in morphology characterization; to assess more CNC material, SMART was used to analyze CNC agglomerates as a proof-of-concept demonstration. The total SMART image analysis time for each laboratory, having between 115 and 244 images, was less than 15 min, after selection of appropriate parameters. The SMART code is now available for the public to use for free at Github™. |
|---|
