3D Shape Modeling for Cell Nuclear Morphological Analysis and Classification

Alexandr A. Kalinin, Ari Allyn-Feuer, Alex Ade, Gordon-Victor Fon, Walter Meixner, David Dilworth, Syed Sameed Husain, Jeffrey R. de Wet, Gerald A. Higgins, Gen Zheng, Amy L. Creekmore, John Wiley, James E. Verdone, Robert W. Veltri, Kenneth J. Pienta, Donald S. Coffey, Yongqun He, Ivo D. Dinov

Abstract

Quantitative analysis of morphological changes in a cell nucleus is important for the understanding of nuclear architecture and its relationship with pathological conditions such as cancer. However, dimensionality of imaging data, together with a great variability of nuclear shapes, presents challenges for 3D morphological analysis. Thus, there is a compelling need for robust 3D nuclear morphometric techniques to carry out population-wide analysis. We propose a new approach that combines modeling, analysis, and interpretation of morphometric characteristics of cell nuclei and nucleoli in 3D. We used robust surface reconstruction that allows accurate approximation of 3D object boundary. Then, we computed geometric morphological measures characterizing the form of cell nuclei and nucleoli. Using these features, we compared over 450 nuclei with about 1,000 nucleoli of epithelial and mesenchymal prostate cancer cells, as well as 1,000 nuclei with over 2,000 nucleoli from serum-starved and proliferating fibroblast cells. Classification of sets of 9 and 15 cells achieved accuracy of 95.4% and 98%, respectively, for prostate cancer cells, and 95% and 98% for fibroblast cells. To our knowledge, this is the first attempt to combine these methods for 3D nuclear shape modeling and morphometry into a highly parallel pipeline workflow for morphometric analysis of thousands of nuclei and nucleoli in 3D.

Introduction

The shape of a cell's nucleus is far more than an aesthetic detail; it is deeply linked to gene regulation, cell differentiation, and disease states such as cancer. As the field moves toward understanding the "4D Nucleome"—how the 3D architecture of the nucleus evolves over time—there is an urgent need for tools that can quantify these changes accurately.

Traditional morphometric tools often rely on 2D "slices" or noisy voxel-based 3D representations. These methods frequently lose fine geometric detail or fail to handle the inherent noise in raw microscopy data. To address these limitations, this research presents a robust, high-throughput 3D pipeline designed for the morphological analysis of cell nuclei and nucleoli, utilizing advanced geometric descriptors to achieve high-accuracy phenotype classification.

Methodology

The proposed approach is a multi-stage pipeline integrated into the LONI Pipeline workflow manager, designed to handle massive datasets through parallel processing. The workflow transitions from raw imaging data to interpretable biological insights through four primary stages: data preparation, surface reconstruction, feature extraction, and classification.

Overview of the 3D shape modeling and morphometry pipeline.

Robust Surface Reconstruction

A core challenge in 3D analysis is the "jagged" nature of raw voxel data. Instead of analyzing these raw voxels directly, the pipeline employs Laplace-Beltrami (LB) eigen-projection. This technique treats the nuclear boundary as a smooth manifold (specifically a "genus zero" surface). By projecting the boundary onto a smooth basis, the method filters out spurious noise from the segmentation process while preserving the essential topology and geometric integrity of the nucleus.

Comparison of raw 3D binary masks (A, C) and reconstructed smooth surfaces (B, D).

Morphometric Descriptors and Feature Extraction

Once the surface is reconstructed, the system extracts six key 3D geometric features:

  1. Volume and Surface Area: Basic size metrics.
  2. Mean Curvature and Curvedness: Measures of how "bumpy" or sharp the surface features are.
  3. Shape Index: A descriptor of the local shape (e.g., spherical, cap-like, or saddle-like).
  4. Fractal Dimension: A measure of the surface's structural complexity.

These features were compared against Spherical Harmonics (SPHARM), the current gold standard. The results indicated that these direct 3D morphometric measures consistently outperformed SPHARM across various machine learning algorithms.

Experimental Setup and Datasets

The study utilized the 3D Cell Nuclear Morphology Microscopy Imaging Dataset, focusing on two primary cell types:

  • Fibroblasts: Primary human cells in proliferating (PROLIF) and serum-starved (SS) states.
  • Prostate Cancer (PC3): Cells undergoing Epithelial-to-Mesenchymal Transition (EMT), a critical process in cancer metastasis.

To validate the pipeline's accuracy, synthetic data (cubes, spheres, and ellipsoids) were also used to ensure geometric measures were calculated correctly.

Results and Discussion

The Predictive Power of the Nucleolus

While many studies focus exclusively on the nucleus, this research highlights the importance of the nucleoli—the sub-compartments within the nucleus. In prostate cancer cells, the spatial distribution and geometric moments of nucleoli were among the most important features for distinguishing between epithelial and mesenchymal states. Specifically, nucleolar "Curvedness" and "Fractal Dimension" were critical discriminators.

Collective Classification

To account for the natural heterogeneity within cell populations, the researchers implemented a "collective classification" strategy. Rather than classifying single cells, the model classifies sets of cells (e.g., groups of 9 or 15). This approach mirrors clinical pathology, where tissue "neighborhoods" are evaluated rather than isolated cells. This strategy saw accuracy jump from approximately 70% for single cells to a remarkable 98% for sets of 15 cells in both fibroblast and prostate cancer datasets.

Conclusion

This study demonstrates that true 3D shape modeling, combined with robust surface reconstruction and collective classification, significantly enhances our ability to distinguish between complex cell phenotypes. By integrating these methods into a massively parallel pipeline, the researchers have provided a scalable framework for population-wide 3D nuclear analysis. Future work may involve relaxing topological constraints to model more complex structures and integrating deep learning to complement these hand-crafted geometric descriptors.

Summary updated

Abstract

Quantitative analysis of morphological changes in a cell nucleus is important for the understanding of nuclear architecture and its relationship with pathological conditions such as cancer. However, dimensionality of imaging data, together with a great variability of nuclear shapes, presents challenges for 3D morphological analysis. Thus, there is a compelling need for robust 3D nuclear morphometric techniques to carry out population-wide analysis. We propose a new approach that combines modeling, analysis, and interpretation of morphometric characteristics of cell nuclei and nucleoli in 3D. We used robust surface reconstruction that allows accurate approximation of 3D object boundary. Then, we computed geometric morphological measures characterizing the form of cell nuclei and nucleoli. Using these features, we compared over 450 nuclei with about 1,000 nucleoli of epithelial and mesenchymal prostate cancer cells, as well as 1,000 nuclei with over 2,000 nucleoli from serum-starved and proliferating fibroblast cells. Classification of sets of 9 and 15 cells achieved accuracy of 95.4% and 98%, respectively, for prostate cancer cells, and 95% and 98% for fibroblast cells. To our knowledge, this is the first attempt to combine these methods for 3D nuclear shape modeling and morphometry into a highly parallel pipeline workflow for morphometric analysis of thousands of nuclei and nucleoli in 3D.
BibTeX
@inproceedings{kalinin2018d,
  author = {Alexandr A. Kalinin and Ari Allyn-Feuer and Alex Ade and Gordon-Victor Fon and Walter Meixner and David Dilworth and Syed Sameed Husain and Jeffrey R. de Wet and Gerald A. Higgins and Gen Zheng and Amy L. Creekmore and John Wiley and James E. Verdone and Robert W. Veltri and Kenneth J. Pienta and Donald S. Coffey and Yongqun He and Ivo D. Dinov},
  title = {{3D Shape Modeling for Cell Nuclear Morphological Analysis and Classification}},
  booktitle = {Scientific},
  year = {2018},
  doi = {10.1038/s41598-018-31924-2},
}
Figure