Video Denoising

Video denoising refers to the process of removing noise from sequences of image frames (videos) to enhance visual clarity and data quality. In medical imaging, especially in fields like light scattering imaging (LSI) or intraoperative video monitoring, denoising is critical for accurate interpretation and analysis.

• Improve image quality: Reduce random noise while preserving structural details and motion consistency.
• Enhance signal detection: Boost signal-to-noise ratio (SNR) and contrast-to-noise ratio (CNR) for better visualization of tissues, tumors, or instruments.
• Facilitate automated analysis: Clean data improves the performance of AI and machine learning algorithms.

• Traditional Filtering:
  • Gaussian blur, median filters, and temporal averaging
  • Fast but often results in loss of detail and motion blur

• Model-Based Methods:
  • Total variation minimization, block-matching (e.g., BM3D)
  • Require manual tuning and are computationally intensive

• Deep Learning-Based Methods:
  • FastDVDNet, DNNs, autoencoders, or GANs
  • Exploit temporal information and spatial correlations
  • Can be trained in a self-supervised (unsupervised) manner, avoiding the need for clean ground-truth videos
  • More robust to complex, dynamic noise

• Intraoperative Light Scattering Imaging (LSI):
  • Enhances visibility of brain structures, tumor margins, or blood flow dynamics
  • Reduces visual interference caused by tissue movement, blood, or lighting fluctuations

• Neuroendoscopy and Microscope Recordings:
  • Improves clarity in endoscopic video streams
  • Enables high-quality recordings for surgical planning, teaching, or AI training

• Post-processing of Surgical Videos:
  • Denoised videos can be used for case documentation, outcome analysis, or dataset generation

• Preserving fine details: Especially important for small vessels or tumor boundaries
• Real-time processing: Denoising must be fast enough for intraoperative use
• Generalizability: Algorithms must adapt to different lighting conditions, tissue types, and imaging setups

• Lin et al. (2025)

¹⁾ proposed an unsupervised adaptive denoising framework using FastDVDNet and noise distribution maps to enhance LSI videos in applications such as nanoparticle analysis and single-cell imaging:

• Significant improvements in SNR and CNR
• Enhanced reliability in particle sizing and cell classification

Video denoising is a vital step in modern neurosurgical imaging workflows. By improving visual and analytical quality, it supports safer surgeries, better diagnostics, and the integration of advanced AI tools.