A GPU-Accelerated Framework for Scalable Multidimensional Analysis of Ultra-High-Resolution Imaging
Poster No:
1571
Submission Type:
Abstract Submission
Authors:
Alejandro Salinas-Medina1, Andrija Štajduhar2, Paule Toussaint1, Xue Liu1, Alan Evans3
Institutions:
1McGill University, Montreal, QC, 2University of Zagreb, Zagreb, Croatia, Zagreb, CR, 3McGill University, Montreal, Quebec
First Author:
Co-Author(s):
Introduction:
The analysis of ultra-high-resolution imaging data, particularly in cellular-level neuroimaging, presents significant computational challenges due to the sheer size (terabytes) and complexity of the datasets. Current methods often struggle to process these images efficiently at micron-level resolution, hindering advanced preprocessing, feature extraction, and annotation workflows. This work introduces a novel GPU-accelerated computational framework designed for scalable multidimensional analysis of such data, addressing the limitations of existing solutions.
Methods:
Our framework integrates state-of-the-art techniques in GPU computing, parallel processing, and memory-optimized tiling (Figure 1) to enable efficient processing of terabyte-scale images. Key components include:
• A highly parallelized implementation of anisotropic diffusion for noise reduction and edge enhancement, optimized for GPU execution.
• GPU-accelerated centroid detection algorithms leveraging local maxima identification, significantly reducing processing time compared to CPU-based methods.
• Dynamic tiling strategies with overlapping boundaries to enable subregion processing, minimizing memory transfers and maximizing parallelization.
• Adaptive memory management to accommodate diverse imaging modalities and data resolutions.
The framework leverages NVIDIA's CUDA ecosystem and CuPy for GPU computation, facilitating seamless scalability across hardware configurations. Techniques such as overlap-aware tiling and label propagation further enhance the analysis and annotation of intricate spatial structures.
• A highly parallelized implementation of anisotropic diffusion for noise reduction and edge enhancement, optimized for GPU execution.
• GPU-accelerated centroid detection algorithms leveraging local maxima identification, significantly reducing processing time compared to CPU-based methods.
• Dynamic tiling strategies with overlapping boundaries to enable subregion processing, minimizing memory transfers and maximizing parallelization.
• Adaptive memory management to accommodate diverse imaging modalities and data resolutions.
The framework leverages NVIDIA's CUDA ecosystem and CuPy for GPU computation, facilitating seamless scalability across hardware configurations. Techniques such as overlap-aware tiling and label propagation further enhance the analysis and annotation of intricate spatial structures.
·Figure 1. Framework´s flow diagram
Results:
By utilizing GPU acceleration, our framework achieves significant reductions in execution time compared to existing CPU-based solutions. These improvements are particularly evident in key processing steps:
• Noise Reduction: Anisotropic diffusion implemented on the GPU reduced processing time by 10x compared to CPU implementations.
• Centroid Detection: Our GPU-accelerated algorithm using local maxima identification processed 200,000 neuron centroids in under 30 seconds, compared to over 5 minutes on a comparable CPU implementation.
• Dynamic Tiling: The use of overlap-aware tiling effectively reduced redundant computations, leading to a 30% improvement in overall pipeline efficiency.
These performance gains enable near real-time processing of complex imaging pipelines, which was previously infeasible. Figure 2 demonstrates the applicability of our solution to various cellular-resolution imaging tasks.
• Noise Reduction: Anisotropic diffusion implemented on the GPU reduced processing time by 10x compared to CPU implementations.
• Centroid Detection: Our GPU-accelerated algorithm using local maxima identification processed 200,000 neuron centroids in under 30 seconds, compared to over 5 minutes on a comparable CPU implementation.
• Dynamic Tiling: The use of overlap-aware tiling effectively reduced redundant computations, leading to a 30% improvement in overall pipeline efficiency.
These performance gains enable near real-time processing of complex imaging pipelines, which was previously infeasible. Figure 2 demonstrates the applicability of our solution to various cellular-resolution imaging tasks.
·Figure 2. High-Resolution Segmentation of a 1-Micron BigBrain Slice: Whole-Slice Cellular Mapping with Detailed Zoom-In on Segmentation Quality
Conclusions:
Our GPU-accelerated framework offers a scalable and hardware-efficient approach to high-dimensional data analysis in cellular-resolution imaging. This generalizable solution provides a critical toolset for accelerated discoveries in computational anatomy and large-scale image processing. The observed performance improvements demonstrate the potential of GPU computing to overcome computational bottlenecks in analyzing ultra-high-resolution imaging data, paving the way for more efficient and comprehensive analyses of complex biological structures.
Modeling and Analysis Methods:
Methods Development 1
Segmentation and Parcellation 2
Keywords:
Computational Neuroscience
Computing
Cortical Layers
Modeling
Neuron
Open Data
Open-Source Code
Open-Source Software
Segmentation
Statistical Methods
1|2Indicates the priority used for review
Abstract Information
By submitting your proposal, you grant permission for the Organization for Human Brain Mapping (OHBM) to distribute your work in any format, including video, audio print and electronic text through OHBM OnDemand, social media channels, the OHBM website, or other electronic publications and media.
The Open Science Special Interest Group (OSSIG) is introducing a reproducibility challenge for OHBM 2025. This new initiative aims to enhance the reproducibility of scientific results and foster collaborations between labs. Teams will consist of a “source” party and a “reproducing” party, and will be evaluated on the success of their replication, the openness of the source work, and additional deliverables. Click here for more information. Propose your OHBM abstract(s) as source work for future OHBM meetings by selecting one of the following options:
Please indicate below if your study was a "resting state" or "task-activation” study.
Healthy subjects only or patients (note that patient studies may also involve healthy subjects):
Was this research conducted in the United States?
Were any human subjects research approved by the relevant Institutional Review Board or ethics panel? NOTE: Any human subjects studies without IRB approval will be automatically rejected.
Were any animal research approved by the relevant IACUC or other animal research panel? NOTE: Any animal studies without IACUC approval will be automatically rejected.
Please indicate which methods were used in your research:
Provide references using APA citation style.
2. Amunts, K., et al. BigBrain: An Ultrahigh-Resolution 3D Human Brain Model. Science, 2013.
3. NVIDIA CUDA Toolkit. Available: https://developer.nvidia.com/cuda