Optical Microscopy and Deep Learning for Absolute Quantification of Nanoparticles on a Macroscopic Scale and Estimating Their Number Concentration

This study introduces a simple and absolute method called evaporated volume analysis (EVA) for determining the number concentration of nanoparticles. The technique involves depositing macroscopic volumes of nanoparticle dispersions onto a glass surface, allowing the solvent to evaporate, and subsequently scanning the entire dried droplet with an optical microscope. Stitched micrographs enable AI-assisted counting of thousands of nanoparticles per droplet. 

The method was validated using ∼60 nm Tm³⁺-doped photon-upconversion nanoparticles, achieving a combined relative standard uncertainty of 2.7%, and was further tested with ∼80 nm Nile red-doped polystyrene and ∼90 nm silver nanoparticles. Theoretical performance limits, including detection limits, quantification limits, and optimal working ranges, are also discussed.

The original article

Optical Microscopy and Deep Learning for Absolute Quantification of Nanoparticles on a Macroscopic Scale and Estimating Their Number Concentration

Antonín Hlaváček, Kateřina Uhrová, Julie Weisová, Hana Brožková, Naděžda Pizúrová

Anal. Chem. 2025, 97, 5, 2588–2592

https://doi.org/10.1021/jasms.4c00455

licensed under CC-BY 4.0

Many new substances─nanoparticles─have been recently prepared and advanced research and technology. They also compose our environment and bodies where part of them is of natural origin and the other part is contributed by human activity. Besides others, the absolute quantification of nanoparticles and estimating their number concentrations (i.e. the number of nanoparticles in a given volume) are important for standardization, trading, studying nanoparticle properties, researching their role in the environment, and evaluating their toxicity. (1,2) Available techniques for nanoparticle quantification measure either the ensemble property of nanoparticle dispersion or utilize nanoparticle counting. (1,2) The counting is absolute and more straightforward. (1,2) Two types of counting approaches─serial and parallel─can be recognized. In serial counting, the nanoparticles are detected individually when the dispersion streams through a detector. Optical detection, (3) resistive pulse sensing, (4) or single particle mass spectroscopy (5) can be used. In a parallel setting, microscopy is used for imaging and counting large numbers of nanoparticles in a single micrograph. For instance, optical microscopy was used for counting and tracking single nanoparticles in a free dispersion. (6)

However, these methods are limited by short detection times, which prevent us from counting small nanoparticles (∼30 nm) with generally low signals. (6,7) In serial counting, the detection time is limited by a need for sufficiently fast flow and diffusion of nanoparticles imposes limits in single-particle tracking. The problem of moving nanoparticles was solved by their embedding in gels (8) or resins. (9) However, the detection is still limited by a background signal (7) from the gel-enclosed solvent, gel matrix, or resin, and the complexity of instrumentation. (8,9) To overcome these limitations, specialized sampling approaches are developed when the nanoparticles are immobilized on a suitable substrate and the solvent is removed traceably. For instance, a nanopipette was designed for transmission electron microscopy (TEM), (10) and anisotropically collapsing gels were used for optical microscopy. (11)

Here, we present a simplistic approach (Scheme 1). A macroscopic volume of nanoparticle dispersion is dropped on a glass surface and the solvent is freely evaporated. An optical microscope scans the dried droplet, and all nanoparticles are counted. We call this method evaporated volume analysis (EVA) because all nanoparticles are counted after droplet volume evaporation. Indeed, the EVA is not a completely new technique and was used either for larger particles with specific imaging signals (e.g. 107 nm Eu chelate-doped polystyrene nanoparticles in time-resolved fluorescence microscope) (12) or for counting particles from microscopic droplets fitting the single field of view of the microscope (e.g. gold nanoparticles in piezo-dispensed microdroplets in scanning electron microscope). (13) On the other hand, theoretical limits, and uncertainty sources were not investigated. The EVA was also overlooked in reviews dedicated to the methods of nanoparticle quantification. (1,2) The reasons for this limited use can be (1) contradicting requirements for sensitivity and the field of view (tends to image either large particles or working with impractically small volumes/fields of view), (7) (2) the need for counting large numbers of nanoparticles (counting uncertainty decreases with number of counted nanoparticles), (3) nanoparticle spatial organization during the droplet drying (causing nanoparticle overlaps and subsequent miscounting). (14) We solved these problems by (1) stitching multiple micrographs of small fields of view ultimately covering macroscopic scales, (2) artificial intelligence automated the counting of nanoparticles, and (3) nanoparticle clustering was prevented by introducing ultralow gelling agarose. As a model, an aqueous dispersion of ∼60 nm Tm³⁺-doped photon-upconversion nanoparticles (UCNPs) coated in carboxylated silica shells (UCNP-COOHs) are quantified. Uncertainty sources are discussed and theoretical limits are analyzed. A wider applicability is tested by imaging ∼80 nm Nile red-doped polystyrene nanoparticles (NileNPs) and ∼90 nm silver nanoparticles (AgNPs). See the Supporting Information (SI) for nanoparticle synthesis (SI Note S1) and characterization (SI Note S2).

Anal. Chem. 2025, 97, 5, 2588–2592 - Scheme 1 a) Nanoparticles (blue dots) in ultra-low-gelling agarose b) After gelling and drying, c) the entire droplet is covered with a grit of 6×6 images, the images are stitched, and nanoparticles are counted

TEM showed ∼60 nm oleic acid-capped NaYF₄ UCNPs doped with 18% Yb³⁺ and 2% Tm³⁺ (Figure S1, SI Note S2). The UCNP-COOHs were formed by coating the UCNPs with a ∼ 5 nm thick carboxylated silica (15) (Figure 1a, Figure S2, SI Note S2) and were stored as a concentrated aqueous dispersion (11.8 ± 0.24 mg mL^–1). The sizes were confirmed by dynamic light scattering estimating hydrodynamic diameters at 69 and 78 nm for oleic acid-capped UCNPs and UCNP-COOHs, respectively (Figure 1b). The dispersion of UCNP-COOHs was virtually aggregate-free as confirmed by agarose gel electrophoresis (Figure 1c). Under the excitation of 976 nm, the UCNP-COOH emitted strongly at near-infrared (802 nm, Figure 1d), and in the epiphoton-upconversion microscope appeared as diffraction-limited spots with a full width at half-maximum of 1.8 px (1.2 μm, Figure 1e,f, SI Note S3, Figure S3).

Anal. Chem. 2025, 97, 5, 2588–2592 - Figure 1

Nanoparticle properties. (a) TEM image of UCNP-COOHs. (b) Hydrodynamic diameters of UCNPs with different surface modifications. (c) UCNP-COOH agarose electrophoresis. (d) The emission spectrum of UCNP-COOHs in water. (e) Epiphoton-upconversion microscopy of UCNP-COOHs in the dried droplet (exposure 5 s). (f) Nanoparticle spot cross sections as indicated in (e). (c-f) Excitation 976 nm. Emission (c) 764–822 nm, and (e,f) 800 ± 25 nm.

Conclusion

In conclusion, image stitching, artificial neural networks, and agarose gel refined the properties of EVA. The number concentration of ∼60 nm UCNPs was estimated with a relative standard uncertainty of 2.7%. Imaging fluorescent NileNPs and plasmonic AgNPs confirmed the wider applicability.