Cellulose nanofiber diameter distributions from microscopy image analysis: effect of measurement statistics and operator

Shaun Ang, Judith Raj Narayanan, Wriju Kargupta, Victoria Haritos, Warren Batchelor

Research output: Contribution to journalArticleResearchpeer-review

16 Citations (Scopus)


Abstract: In the field of cellulose nanofibers (CNF), the determination of their nano-scale fibre diameters is crucial for characterising the material and helping end-users compare different products available on the market. Yet, measuring CNF diameters from microscopy images is often a tedious task due to the limited field of view at high magnification, high fibre counts and complex entanglement of fibre networks. In this work, the effect of measuring all fibres in an entire image, 100 representative fibres in an image and all fibres in a reduced area of an image were investigated. Two CNF samples were analysed: a coarser more heterogeneous fibre sample and a finer more homogeneous fibre sample. Four high magnification images were analysed per sample. Different people (operators) were given the same task of measuring the fibres to study the effect of operator bias. All fibres from all images were first measured by Operator 1 (main operator) to determine the baseline distribution. Operators 2 and 3 were tasked with selecting and measuring 100 representative fibres from each image. The results showed that when the operators were asked to select fibres, they showed consistent bias towards larger fibres thus rendering the results unreliable. A more promising approach was found when Operators 2 and 3 analysed all fibres within a reduced area within an image. This showed a maximum deviation from the baseline fibre diameter distribution of 9 nm for the coarser more heterogeneous sample and 2 nm for the finer more homogeneous sample. Statistical analysis showed that approximately 150 fibres needed to be measured in order to obtain representative results, provided all fibres within the selected area were measured. Kruskal–Wallis statistical testing also showed that the technique of measuring all fibres in a reduced area of an SEM image yielded more representative results to the baseline distribution compared to the technique of measuring 100 representative fibres. The results here provide the foundation for future accurate measurements of CNF diameter distributions.

Original languageEnglish
Pages (from-to)4189–4208
Number of pages20
Issue number8
Publication statusPublished - May 2020


  • Cellulose nanofiber
  • Fibre diameter
  • Histogram
  • SEM
  • Statistics

Cite this