One of the most challenging problems of the 21st century is the exploration of human and animal neural systems. This exploration, which started essentially in the 19th century, became only realizable in the last 50 years due to the development of the electron microscope and electrophysiology methods. Although electron microscope affords us with extreme resolution enough to inspect even the most minute structures at sub cellular level, limits us to the study of non-living tissue. More recent developments in imaging instrumentation during the last 12 years, made available a new generation of microscopes which essentially work like scanners, and exploit a natural property of tissue called fluorescence. With these new laser microscopes we can explore living tissue either in-vitro cultures or by putting sedated small live mammals under the microscope. Thus, a whole new generation of images becomes available to neuroscientists, to explore not only the anatomical structure of neuronal cells but also the function at the single cellular level or in small groups of neurons.

Our work in this project focused on the development of tools for the analysis of these kinds of images with special emphasis on the generation of digital images of groups of neurons, the identification of their dendritic arbors, axons, and somas. Moreover, we developed an almost fully automated method for the extraction and detection of sub cellular structures playing a key role in synaptic function, the dendritic spines. These   are small protrusions emanating from the surface of dendritic branches. They work as terminals at the receiving end of synapses. Dendritic spine populations change over time, and their density plays a key role in modern studies of aiming to establish their role in the regulation of synaptic function in a series of neurological conditions such as autism, bipolar disorder, chemical dependence, and schizophrenia. Spines are anatomically, sub cellular compartments of neural cells. They can be generated in a matter of a few minutes and disappear a few hours after their birth. However some of them become permanent anatomical features of dendritic branches and when this happens neurons become stable and believed to have achieved a learning milestone.

With our work, we can generate, both computationally and literally inexpensively, binary images of dendritic arbors and of their spines. We can create metadata of the graph structure of dendritic arbors and include the locations of spines in three-dimensional coordinates. Moreover, the algorithms we created in the course of this project paved the pathway for the development of new computational tools for measuring the interaction between axons and dendritic arbors of neighboring neural cells, and to assess how these can be modified by means of neurodegeneration reversion drugs. The advantage that image analysis tools like ours, is that they enable scientists to observe the longitudinal connection between synaptic network anatomy and function, and its modifications due to learning or pharmaceutical interventions.

 The development of these computational tools was a result of novel achievements in the domain of Computational Harmonic Analysis. In particular, we invented new geometric metrics which can distinguish more round or oval-shaped 3D structures from tubular ones, regardless of their position in the 3D space. We also studied the development of sensitive filters, which can identify the surface of neuronal cells independently of their positioning in the interior of a 3D image of a group of neurons. Our spine detection tool mimics the function of an omnidirectional second-order derivative and it, thus, becomes a sensitive filter to dendritic surface anomalies. Our algorithms developed hand-in hand with new mathematical theories which aim in establishing why these algorithms work. All of our findings have been made available to the scientific community, in the form of open source, free to use, software with basic documentation, with publications in peer-reviewed journals, and with presentations in key conferences. Last but not least, our work on the processing of weak images of cultured neurons, where the visibility of neuronal structure has been degraded, due to light attenuation, led to the development of a new mathematical theory for the modeling and the elimination of shadow and glare effects from natural images. This new approach to the visibility effects due to variations of illumination led us to the development of algorithms for visibility restoration in natural images will be commercialized by a startup company in which the University of Houston holds a significant interest. The first application domain of this commercial effort is the enhancement of marine video acquired in poor visibility conditions by submersible vehicles or divers.

Last, the project generated numerous educational opportunities for doctoral students and two postdoctoral fellows. Their core experience involved a mix of high class pure mathematics in the domain of analysis and core algorithmic development and experimental validation using 3-D images of neurons and neurites.  

Last Modified: 07/16/2017
Modified by: Emanuel I Papadakis