Traced Neuron

Title

Towards the Neurome: Segmentation of single neurons from 3D confocal microscopy

Members

Suvadip Mukherjee, Saurav Basu, Barry Condron and Scott T. Acton

Funded by

NSF

Summary | Publications | Downloadable data | Code


Research Summary

As we step into an age of emerging research in Bioimage Informatics, a relevent problem in this field is that of analyzing the neuronal structure of complex organisms. While the neurome of the C.elegans have been fully understood, a major research focus is to analyze the neural network of more complicated species. A major challenge in this field is to analyse 3-D microscopy images to represent the neuronal morphology mathematically. In this project we are interested in studying the fruit fly Drosophila. An adult Drosophila larva has around 20,000 neurons in its nervous system, and this animal has been considered in the biology community as a prototype species for modeling the brain of more sophisticated animals, such as rats or humans. Our research in this field focuses on segmenting the neurons of the fruit fly from a 3-D confocal microscopy image. Over the last few years, we have developed three novel neuron tracers-- Tree2Tree [2], Tree2Tree-2[6] and Tubularity Flow Field (TuFF)[1].

Tree2Tree is a graph theoretic neuron tracer, which uses sophisticated graph theory algorithms along with traditional image processing techniques to compute the neuron trace (centerline). Tree2Tree-2 improves upon its predecessor by incorporating a variational segmentation step as well as a post processing stage to link the disjoint neuronal segments.

Our recent method is called TuFF. TuFF uses the flexibility of level sets to handle the neuron topology and has the capability of detecting and connecting neuron segments even when the signal intensity is significantly low. Results on 2-D and 3-D datasets have shown promise for further application to high throughput neuron imaging.

Once the segmentation is performed, the centerline of the object is extracted and saved in a swc format, which can be read by the popular image analysis software like Vaa3D. We have used Vaa3D for visualization and comparison of the tracers.

n1_gt n2_gt n4_gt n4_gt n5_gt
n1__ours n2__ours n4__ours n4__ours n5__ours
Figure 1: Sample neuron tracing results, shown using the Vaa3D software. The top row shows manually segmented ground truth. Ours results are shown in magenta.

The following points summarize our contribution to the project:

  • TuFF, Tree2Tree and Tree2Tree are automated. Segmentation does not require user input seed points.
  • Our tracers are robust to noise and clutter in confocal microscopy.
  • TuFF is equipped to handle significant intensity variation in the images.
  • High accuracy in tracing is observed in terms of neuron morphology preservation and accurate centerline detection.
  • Superior tracing performance in comparison to the state of the art methods.
To know more about our algorithms, check out our publications.

Publications [Top]

  1. S. Mukherjee, B. Condron and S.T. Acton, "Tubularity Flow Field – A Technique For Automatic Neuron Segmentation," IEEE Transactions on Image Processing, vol.24, no.1, pp.374,389, Jan. 2015. [paper]
  2. S. Basu, et al. "Segmentation and tracing of single neurons from 3D confocal microscope images." IEEE Journal of Biomedical and Health Informatics, 17.2 (2013): 319-335. [paper]
  3. S. Mukherjee et al. "Neuron segmentation with level sets", ACSSC 2013:1078-1082. [paper]
  4. S. Mukherjee, B. Condron and S. T. Acton, "Chasing the neurome: Segmentation and comparison of neurons," EUSIPCO 2013: 1-4 [paper]
  5. S. Mukherjee and S. T. Acton, "Vector field convolution medialness applied to neuron tracing," ICIP 2013: 665-669 [paper]
  6. S. Mukherjee et al., "Tree2Tree2: Neuron tracing in 3D," ISBI 2013: 448-451. [paper]
  7. R. Sarkar, S. Mukherjee and S. T. Acton, "Shape descriptors based on compressed sensing with application to neuron matching", ACSSC 2013: 970-974
  8. S. Mukherjee et al. "A geometric-statistical approach toward neuron matching", ISBI 2012:772-775. [paper]

Downloadable data [Top]

In our paper, we have used the 3D neuron stacks which are imaged by Dr. Condron's lab at UVa. This dataset is called Condron dataset. You may download and use the images for research purpose only. Each folder consists of 3D confocal microscopy images of Drosophila, a manual segmentation file (gt.swc) and our tracing result (TuFF.swc or T2T2.swc). There are many softwares which provide beautiful 3D vizualization. I have found Vaa3D to be particularly useful, since the traces can be "dragged and dropped". The remaining images used in our paper was obtained from the Diadem challenge dataset. If you are using our data for academic purposes, please cite the following papers:
  • S. Mukherjee, B. Condron and S.T. Acton, "Tubularity Flow Field – A Technique For Automatic Neuron Segmentation," IEEE Transactions on Image Processing, vol.24, no.1, pp.374,389, Jan. 2015.
  • S. Mukherjee, S. Basu, B. Condron and S.T. Acton et al., "Tree2Tree2: Neuron tracing in 3D," IEEE 10th International Symposium on Biomedical Imaging (ISBI) vol., no., pp.448,451, 7-11 April 2013.
n1_gt n2_gt n4_gt n4_gt n5_gt
[data] [data] [data] [data] [data]
n1__ours n2__ours n4__ours n4__ours n5__ours
[data] [data] [data] [data] [data]
n1__ours n2__ours n4__ours n4__ours n5__ours
[data] [data] [data] [data] [data]
n1__ours n2__ours
[data] [data]
We will keep uploading more data from time to time.

Code [Top]

[GUI] [3D Matlab code]