Projects
A Framework for the Segmentation of the Cerebral Cortex Laminar Structure
Student Names: Jiaxuan (Tony) Wang
Quick description: Understanding the brain’s anatomical structure and how it processes information are critical steps to exploring potential treatments for neurological disorders (i.e., Parkinson’s or Alzheimer’s disease). Axonal tracing studies help neuroanatomists understand how signals are transmitted across different brain cortex regions and Brodmann’s six cortical layers. The fluorescent images taken of the brain in anterograde tracing studies (which highlight only the axon fibres) cannot show the six cortical layers. One way to identify the layers is to overlay the fluorescent image to a Nissl-stain highlighted histological image (due to the cortical layer features being visible in the Nissl image).
Wang, J., Gong, R., Heidari, S., Rogers, M., Tani, T., Abe, H., Ichinohe, N., Woodward, A., Delmas, P. (2024). A Deep Learning-Based Pipeline for Segmenting the Cerebral Cortex Laminar Structure in Histology Images. To appear in Neuroinformatics.
Syndemic modeling of phenomena
Student Names: Cesar Victoria-Ramirez, Yadira Fleitas-Toranzo
Quick description: The syndemic analysis of a population specific condition (commonly a disease) requires the identification of all the possible factors that can have a negative relation to it. This type of analysis requires a transdisciplinary approach where different areas of knowledge identify the factors that could have a correlation with the condition.
The syndemic approach requires high quality data and sufficient individuals to identify possible groups that share negative factors that increase the risk of developing the condition.
We can understand conditions as any negative situation, when using the syndemic approach the objecting is to generate models that consider multiple factors to explain the condition, the models could be static, simple dynamics like compartmental models, physical base dynamic models, mathematical and probabilistic models or machine learning models. The objective is to obtain the best descriptor of the phenomena and BioComLab approach is to use computer sciences to generate tailor solutions to each problem.
Integrating AR/VR capabilities to the IVSLab portable multi-camera system
Student Names: Iyoto Kuchimaru
Quick description: The IVSlab multi-camera system has real-time GPU capabilities to process videos and display outputs of deep learning and 3d map creation currently displayed on a mobile phone screen. The student’s task will be to interface with a VR/AR system for 3D display and interaction purposes including hand gestures control. Expectations will be to demonstrate the capabilities of the AR/VR addition on a set of applications to be determined as the project progresses.
Enhancing self-driving: Speed bump and pothole detection and quantization
Student Names: Ruigeng Wang
Quick description: Providing a hybrid state-of-the-art 2D/3D computer vision and deep learning solution to both detect and quantify potholes and speed bumps in real-time (as per car hardware) for self-driving systems.
Here, we focus on self-driving cars (including but not limited to TESLA) using a vision-based camera system rather than LiDAR or other similar active sensing approaches. We also tested our science and technology solution against the current Tesla FSD2.0 and FSD3.0 (full self-driving) platforms. Our solution combines 2D deep learning detection of objects and geometrical features of the road to assess potholes and speed bump grades to trigger a change of direction decision (or not). We trained and tested both on the open-source pothole and speed bump datasets and collected datasets.
The results show satisfactory performance and robustness on mixed data from open source and collected from onboard and off-the-shelf cameras on New Zealand roads.
