Ezard, T. | Search bioRxiv

bioRxiv2024

How many specimens make a sufficient training set for automated 3D feature extraction?

Deep learning has emerged as a robust tool for automating feature extraction from 3D images, offering an efficient alternative to labour-intensive and potentially biased manual image segmentation methods. However, there has been limited exploration into the optimal training set sizes, including assessing whether artificial expansion by data augmentation can achieve consistent results in less time and how consistent these benefits are across different types of traits. In this study, we manually segmented 50 planktonic foraminifera specimens from the genus Menardella to determine the minimum number of training images required to produce accurate volumetric and shape data from internal and external structures. The results reveal unsurprisingly that deep learning models improve with a larger number of training images and that data augmentation can enhance network accuracy by up to 8.0%. Notably, predicting both volumetric and shape measurements for the internal structure poses a greater challenge compared to the external structure, due to low contrast between different materials and increased geometric complexity. These results provide novel insight into optimal training set sizes for precise image segmentation of diverse traits and highlight the potential of data augmentation for enhancing multivariate feature extraction from 3D images. Subject CategoryLife Sciences - Computer Science Subject Areascomputational biology, Artificial Intelligence

evolutionary biology↗

bioRxiv2018

Predictability of intraspecific size variation in extant planktonic foraminifera

The size structure of plankton communities is an important determinant of their functions in marine ecosystems. However, few studies have quantified how organism size varies within species across space. Using a recently-digitised museum collection, we investigate at high intraspecific resolution how planktonic foraminifera, an important microfossil group, vary in size across the tropical and subtropical oceans of the world. We measured 3799 individuals of nine species in 53 seafloor sediments and analysed potential size biases in the museum collection. For each site, we obtained corresponding local values of mean annual sea-surface temperature (SST), net primary productivity (NPP), and relative abundance of each species. Given former studies, we expected species to reach largest shell sizes under optimal environmental conditions. At species level, we find the expected pattern; however, at population level, species greatly differ in how much size variation is explained by SST, NPP and/or relative abundance. While some species show a high predictability of size variation given one single variable (Trilobatus sacculifer, Globigerinella siphonifera, Pulleniatina obliquiloculata, Globorotalia truncatulinoides), other species showed either weak or no relationships between size and the studied covariates (Globigerinoides ruber, G. conglobatus, Neogloboquadrina dutertrei, G. menardii, Globoconella inflata). By incorporating intraspecific variation and sampling broader geographical ranges compared to previous studies, we conclude that shell size variation in planktonic foraminifera species cannot be consistently predicted by the environment. Our results caution against the general use of size as a proxy for planktonic foraminifera environmental optima, and emphasise the need for more studies analysing their within-species size variation. More generally, our findings show that contrasting results can be obtained when analysing patterns at different organisational levels, and highlights the utility of natural history collections.

ecology↗