Neuroscience
Computational spatial analysis of peripheral nervous system anatomical data
The ultimate objective of this research is to improve nerve electrostimulation and neuromodulation strategies for the treatment of patients by advancing the understanding of peripheral nerve neuroanatomy through the development and application of mathematical and computer science tools.
From the statistics and computer science perspective, these projects belong to the field of spatial point pattern analysis. Generally, sets of real or abstract objects represented by spatial locations and spatial, morphological, statistical, or categorical qualities comprise spatial point patterns. The analysis of such data is applicable to several scientific fields, such as ecology, epidemiology, geoscience, biology, astronomy, econometrics, and criminal justice. Using multiple spatial statistical functions and generative models to characterize the pattern, identifying aspects of the underlying generating process, and developing models to mimic comparable realizations are all components of the analysis of spatial point patterns. In our research, we employ spatial point pattern analysis to answer multiple puzzling questions related to neuroscience and neuroanatomy. The innovative aspect of the methodological development is the merging of spatial quantification with techniques from various other areas of computer science, such as machine learning, graph theory, optimization, and image processing.
Our research utilized two types of data: (1) transmission electron microscopy images of the vagus nerve cross-sections from rats and humans and (2) confocal microscopy images of the enteric nervous system in the colon from mice and humans. These data types require sophisticated image preprocessing to extract spatial locations and morphometric characteristics of the entities of interest.
Organization of unmyelinated axons
Briefly, using a custom-built, high-throughput deep learning model, the cross-sections of the vagus and pelvic nerves are processed to segment the unmyelinated axons. To generate spatial point patterns, the positions of segmented axons are approximated using their centroids. Their spatial configurations are characterized by first- and second-order spatial statistics. Sinkhorn distance, a metric developed from the solution of the optimal transportation problem, is used to quantify the similarities between these spatial statistics. The following spatial analysis demonstrates that the organization of the unmyelinated axons in the vagus and pelvic nerve cross-sections is non-random (but rather inhomogeneous and anisotropic), which can contribute to the development of spatially selective stimulation of peripheral nerves as treatment regimens with fewer side effects. Future goals of this work include classifying nerve cross-sections based on their spatial features for samples varying in age, sex, health conditions, etc., and developing spatial models for nerve electrostimulation research.
This research has two components:
- Automated segmentation of unmyelinated fibers, and other hard-to-segment structures in PNA (collaboration with Terry Powley, Leif Havton, and Murat Dundar)
- Quantitative descriptions of continuous spatial variation and associated spatial autocorrelation, co-occurrence, and colocalization (in collaboration with Terry Powley group)
- Point-pattern analysis and quantitative description of relation/correlation between spatially arranged classes of structures (collaboration with Leif Havton group and Warren Grill group)
Organization of the interganglionic connections in the enteric nervous system
Images of the enteric nervous system (ENS) of mice and humans are used to construct a network model of the ENS’s architecture. The model identifies the spatial and topological characteristics of the ganglia (clusters of neurons and glial cells), the inter-ganglionic connections, and the neuronal structure inside the ganglia by combining spatial point pattern analysis with graph generation. The spatial patterns are generated using a hybrid hardcore-Strauss process, while the spatially-embedded network is constructed using a planar random graph creation process. We cannot yet conclude that the parameters of the current model are ideal for all ENS samples belonging to the same age group or having comparable pathologies due to a lack of biological data and diversity in sample size. However, our research suggests that the expressiveness of the generative model will be sufficient to represent the variability of such ENS data. We hope that an increased understanding of the ENS connectome will facilitate the use of neuromodulation in the treatment of bowel motility disorders and clarify anatomic diagnostic criteria.
- Spatial analysis of network-like structures and modeling the organization and architecture of the enteric nervous system (collaboration with Marthe Howard group and Bob Heuckeroth lab, computational work with Alex Pothen).
This research has been generously supported by the NIH SPARC initiative.
Page updated on 2022-12-07 17:20:17 -0500