NucMM Dataset: 3D Neuronal Nuclei Instance
Segmentation at Sub-Cubic Millimeter Scale

Zudi Lin^1† Donglai Wei^1† Mariela Petkova¹ Yuelong Wu¹ Zergham Ahmed¹ Krishna Swaroop K^2* Silin Zou¹
Nils Wendt^3* Jonathan Boulanger-Weill¹ Xueying Wang¹ Nagaraju Dhanyasi¹ Ignacio Arganda-Carreras^4,5,6
Florian Engert¹ Jeff W. Lichtman¹ Hanspeter Pfister¹

¹ Harvard University ² NIT Karnataka ³ Technical University of Munich
⁴ Donostia International Physics Center ⁵ University of the Basque Country ⁶ Ikerbasque, Basque Foundation for Science

† Equally contributed.

* Works are done as interns at Harvard University.

Abstract

Segmenting 3D cell nuclei from microscopy image volumes is critical for biological and clinical analysis, enabling the study of cellular expression patterns and cell lineages. However, current datasets for neuronal nuclei usually contain volumes smaller than 10^-3 mm³ with fewer than 500 instances per volume, unable to reveal the complexity in large brain regions and restrict the investigation of neuronal structures. In this paper, we have pushed the task forward to the sub-cubic millimeter scale and curated the NucMM dataset with two fully annotated volumes: one 0.1 mm³ electron microscopy (EM) volume containing nearly the entire zebrafish brain with around 170,000 nuclei; and one 0.25 mm³ micro-CT (uCT) volume containing part of a mouse visual cortex with about 7,000 nuclei. With two imaging modalities and significantly increased volume size and instance numbers, we discover a great diversity of neuronal nuclei in appearance and density, introducing new challenges to the field. We also perform a statistical analysis to illustrate those challenges quantitatively. To tackle the challenges, we propose a novel hybrid-representation learning model that combines the merits of foreground mask, contour map, and signed distance transform to produce high-quality 3D masks. The benchmark comparisons on the NucMM dataset show that our proposed method significantly outperforms state-of-the-art nuclei segmentation approaches.

Dataset

NucMM dataset characteristics.We collected and fully annotated a neuronal nuclei segmentation dataset at the sub-cubic millimeter scale. The two volumes in the dataset cover two species and two imaging modalities.

Method

Hybrid-representation learning model. (a) Our U3D-BCD model learns a set of hybrid representations simultaneously, including foreground mask, instance contour, and signed distance transform map calculated from the segmentation. (b) The representations are combined in seeding and watershed transform to produce high-quality segmentation masks.

Citation

@inproceedings{lin2021nucmm,
  title={NucMM Dataset: 3D Neuronal Nuclei Instance Segmentation at Sub-Cubic Millimeter Scale},
  author={Lin, Zudi and Wei, Donglai and Petkova, Mariela D and Wu, Yuelong and Ahmed, Zergham and 
          Zou, Silin and Wendt, Nils and Boulanger-Weill, Jonathan and Wang, Xueying and 
          Dhanyasi, Nagaraju and others},
  booktitle={International Conference on Medical Image Computing and Computer-Assisted Intervention},
  pages={164--174},
  year={2021},
  organization={Springer}
}

Acknowledgement

This work has been partially supported by NSF award IIS-1835231 and NIH award U19NS104653. We thank Daniel Franco-Barranco for setting up the challenge using NucMM. M.D.P. would like to acknowledge the support of Howard Hughes Medical Institute International Predoctoral Student Research Fellowship. I.A-C would like to acknowledge the support of the Beca Leonardo a Investigadores y Creadores Culturales 2020 de la Fundación BBVA.