BDNet: A Real-Time Biomedical Image Denoising Network with Gradient Information Enhancement Loss

Area of Science:

• Biomedical Imaging
• Computational Biology
• Medical Diagnostics

Background:

• Image noise significantly challenges accurate analysis in biomedical imaging.
• Effective noise reduction is critical for medical diagnostics and research.
• Existing methods often struggle with preserving fine structural details.

Purpose of the Study:

• To develop a real-time biomedical image denoising network (BDNet).
• To enhance gradient and high-frequency information while suppressing noise.
• To improve the accuracy and efficiency of biomedical image analysis.

Main Methods:

• A lightweight U-Net-inspired encoder-decoder architecture.
• Incorporation of a Convolutional Block Attention Module at the bottleneck.
• A novel gradient-based loss function combining Sobel, L1, L2, and LSSIM losses.

Main Results:

• BDNet achieved state-of-the-art performance on the Fluorescence Microscopy Denoising (FMD) dataset.
• Outperformed existing convolutional and Transformer-based models in PSNR, RMSE, SSIM, and LPIPS.
• Demonstrated superior denoising capability and real-time inference speed.

Conclusions:

• BDNet offers an effective and practical solution for biomedical image denoising.
• The proposed network significantly improves biomedical image quality, especially for fluorescence microscopy.
• BDNet facilitates more accurate medical diagnostics and research through enhanced image analysis.