An Efficient Deep Learning and Statistical Modelling Pipeline for Porosity Analysis in Foraminifera
(2026) The Swedish Climate Symposium 2026- Abstract
- Foraminifera are single-celled eukaryotes; most possess a perforated calcite test (shell) that fossilizes
in marine sediments. During calcification, morphological features reflect environmental conditions,
particularly, foraminiferal porosity is a promising proxy for reconstructing past ocean oxygenation.
However, traditional analysis relies on averages from small parts of the shell in 2D slices due to shell
curvature, which masks ontogenetic (growth-related) variations. We present an efficient pipeline
combining deep learning and statistical modelling to resolve chamber-specific porosity in 3D
micro-computed tomography ( micro -CT) scans of the benthic species Elphidium clavatum, enabled
by a streamlined... (More) - Foraminifera are single-celled eukaryotes; most possess a perforated calcite test (shell) that fossilizes
in marine sediments. During calcification, morphological features reflect environmental conditions,
particularly, foraminiferal porosity is a promising proxy for reconstructing past ocean oxygenation.
However, traditional analysis relies on averages from small parts of the shell in 2D slices due to shell
curvature, which masks ontogenetic (growth-related) variations. We present an efficient pipeline
combining deep learning and statistical modelling to resolve chamber-specific porosity in 3D
micro-computed tomography ( micro -CT) scans of the benthic species Elphidium clavatum, enabled
by a streamlined annotation workflow. Our approach utilizes a multi-planar 2D U-Net for voxel-wise
segmentation, preserving structural detail from grayscale data, followed by t-SNE and HDBSCAN to
cluster pores based on spatial distribution. We analyze 122 specimens from the Baltic Sea, spanning the
Last Interglacial to the present. The proposed statistical modelling framework allows for rigorous
testing of morphological heterogeneity across growth stages. Preliminary results highlight that pores in
specific chambers, notably the penultimate chamber, frequently deviate from whole-shell means. These
findings suggest that aggregate metrics may bias paleoenvironmental inferences, and our pipeline offers
a robust tool for decoding environmental fluctuations recorded within foraminiferal tests. (Less)
Please use this url to cite or link to this publication:
https://lup.lub.lu.se/record/31490c21-c95e-493b-b3e3-1370925180bd
- author
- Wu, Hanqing
LU
; Choquel, Constance
; Ni, Sha
LU
; Filipsson, Helena L.
LU
and Pirzamanbin, Behnaz
LU
- organization
- publishing date
- 2026-05-21
- type
- Contribution to conference
- publication status
- unpublished
- subject
- conference name
- The Swedish Climate Symposium 2026
- conference location
- Lund, Sweden
- conference dates
- 2026-05-20 - 2026-05-22
- project
- A big data approach to environmental change: Statistical quantification of 3D microfossil images
- language
- English
- LU publication?
- yes
- id
- 31490c21-c95e-493b-b3e3-1370925180bd
- date added to LUP
- 2026-06-04 11:33:59
- date last changed
- 2026-06-08 16:23:57
@misc{31490c21-c95e-493b-b3e3-1370925180bd,
abstract = {{Foraminifera are single-celled eukaryotes; most possess a perforated calcite test (shell) that fossilizes <br/>in marine sediments. During calcification, morphological features reflect environmental conditions, <br/>particularly, foraminiferal porosity is a promising proxy for reconstructing past ocean oxygenation.<br/>However, traditional analysis relies on averages from small parts of the shell in 2D slices due to shell <br/>curvature, which masks ontogenetic (growth-related) variations. We present an efficient pipeline <br/>combining deep learning and statistical modelling to resolve chamber-specific porosity in 3D <br/>micro-computed tomography ( micro -CT) scans of the benthic species Elphidium clavatum, enabled <br/>by a streamlined annotation workflow. Our approach utilizes a multi-planar 2D U-Net for voxel-wise <br/>segmentation, preserving structural detail from grayscale data, followed by t-SNE and HDBSCAN to <br/>cluster pores based on spatial distribution. We analyze 122 specimens from the Baltic Sea, spanning the <br/>Last Interglacial to the present. The proposed statistical modelling framework allows for rigorous <br/>testing of morphological heterogeneity across growth stages. Preliminary results highlight that pores in <br/>specific chambers, notably the penultimate chamber, frequently deviate from whole-shell means. These <br/>findings suggest that aggregate metrics may bias paleoenvironmental inferences, and our pipeline offers <br/>a robust tool for decoding environmental fluctuations recorded within foraminiferal tests.}},
author = {{Wu, Hanqing and Choquel, Constance and Ni, Sha and Filipsson, Helena L. and Pirzamanbin, Behnaz}},
language = {{eng}},
month = {{05}},
title = {{An Efficient Deep Learning and Statistical Modelling Pipeline for Porosity Analysis in Foraminifera}},
year = {{2026}},
}