DenoisEM

DenoisEM performs semi-automated denoising and deconvolution of electron microscopy (EM) images to reduce noise while preserving nanometer-resolution structural details in large 3D EM datasets.

Key Features:

• Algorithms: Implements non-local means, BLS-GSM (Blind Low-rank plus Sparse Gaussian Mixture), and Tikhonov deconvolution for image denoising and deblurring.
• GPU acceleration: Uses Quasar for GPU-accelerated processing to speed computationally intensive denoising operations.
• Parallel computing: Employs parallelized computation to enable fast parameter tuning and high-throughput image processing.
• Large-scale 3D handling: Targets processing of large-scale 3D EM datasets to accommodate volumetric EM data sizes.
• Preservation of structure: Reduces noise while aiming to preserve critical ultrastructural details at nanometer resolution.
• Performance: Demonstrates an order-of-magnitude speed improvement and reports the capability to accelerate data-acquisition workflows by a factor of four.
• Downstream compatibility: Produces denoised images suitable for downstream (semi-)automated segmentation and quantitative analysis of ultrastructures.

Scientific Applications:

• Volume EM denoising: Enhances image quality of large 3D EM datasets for visualization and analysis of nanometer-scale structures.
• Segmentation support: Provides input images that facilitate (semi-)automated segmentation of cellular ultrastructures in volume EM.
• Ultrastructure analysis: Improves signal-to-noise for subsequent quantitative analysis of subcellular features in EM data.
• Acquisition workflow optimization: Enables accelerated imaging strategies by compensating for increased noise from faster data acquisition.

Methodology:

Applies non-local means, BLS-GSM (Blind Low-rank plus Sparse Gaussian Mixture), and Tikhonov deconvolution with GPU acceleration via Quasar and parallelized parameter tuning.