You are here

Peering into the hardware that makes our minds tick

  • Neuron reconstructed

Neuron reconstructed

An entrepreneurship opportunity

One of the hallmarks of the human body is our nervous system, which helps us sense our surroundings and carry out complicated activities. The nervous system is made of millions of neurons. Studying them can provide insight into normal bodily function, disorders and a holy grail of research: how the human brain functions.

No two neurons are exactly alike. Reconstructing this natural complexity to study them better is quite a task. Scientists from IISc have devised a framework to reconstruct 3-dimensional neuronal structures with minimal manual effort and minimum processing time. The authors have said that their algorithm could potentially be made commercial, and they would be more than happy to collaborate with young interested entrepreneurs.

The 3-D visualization of neurons involves arduous manual calibration, which is very slow. Besides, the resulting visualizations are not very precise, considering how fine the hair like fibres on neurons are. The new framework – a multi-stage process with feedback – devised by IISc is a much better alternative. Where commercial softwares take 4-6 hours or more to generate the 3-D boundary with significant manual effort, this framework takes about 10 minutes on an 8 GB 1.86 GHz quad-core workstation and that too with minimal manual effort.

This framework begins by detecting 2-D outlines of different layers using image processing. The splitting or segmentation of this layers itself is quite flexible. It is automatic and is also supplemented by the user. The next step is stitching these layers together. This is done after the boundary pixels for each 2-D layer has been determined, which again provides for user flexibility. The actual reconstruction in 3 dimensions is another algorithm that uses geometric properties of adjacent points to recreate a volume. The biggest issue with this reconstruction process are disconnected locations in between. This is solved by a correction mode that joins the gaps. The smallest 3-D unit is called a voxel – which is a portmanteau of “volume” and “pixel” The final stage of error analysis is done by the process backward with the 3-D model, to check if the original mesh results.

“The software for this framework was devised from scratch”, said Dr. Vijay Natarajan, one of the co-authors of the study. “It is an ‘in-house’ software based on OpenGL. We are open to sharing this algorithm with collaborators and researchers, so in that sense it is ‘public’”.

Dr. Sikdar, a co-author on the study, said “Theoretically it is possible to do 3-D reconstruction of other objects, but we need to have access to 2-D slices or contours of considerable resolution. The process itself is very challenging”.

“The 2-D images are obtained by inserting a fluorescent dye and scanning different slices. If there are branches or kinks, the dye might not stain the fine structures uniformly resulting in discontinuities. The other problem is with non-specific staining, where objects that aren’t of concern get dyed by the stain as well. The automatic procedure helps filter this noise”, he said.

The collaboration was spontaneous and independent, said the authors, but the Centre for Mathematical Biology at IISc did play a role. They are also the funding agency of the project under the Department of Science and Technology.

With the advent of 3-D printing, there is potential to print these neurons as well for education purposes. “I know for sure that models of protein molecules have been printed, and I have printed one myself! With the advent of commercial, compact 3-D colour printers, we may be able to print these neurons and use them to study their tree-like structure”, said Dr. Natarajan.

3-D neuron reconstruction has been exploited commercially by a Swiss company called BitPlane. However, this algorithm’s sophistication is comparable. The authors opined that their algorithm could potentially be made commercial, and they would be more than happy to collaborate with young interested entrepreneurs.

About the authors:

Vijay Natarajan is an Associate Professor with the Department of Computer Science and Automation, and Supercomputer Education and Research Centreat the Indian Institute of Science. Kanuj Kumar was a student in his lab. Prof Sikdar and Kalyan Srinivas are with the Molecular Biophysics Unit, IISc.

http://www.csa.iisc.ernet.in/~vijayn/ | Tel: +91 80 22932909; Email: vijayn@csa.iisc.ernet.in