Blind microscopy image denoising with a deep residual and multiscale encoder/decoder network.

Area of Science:

• Medical Imaging
• Computer-Aided Diagnosis (CAD)
• Deep Learning

Background:

• Image quality is crucial for accurate medical diagnoses, especially in microscopy.
• Perturbations during image acquisition can degrade medical image quality.
• Deep learning models are increasingly used for image denoising in CAD systems.

Purpose of the Study:

• To propose an innovative, lightweight deep multiscale convolutional encoder-decoder neural network for medical image denoising.
• To enhance the accuracy and reliability of computer-aided diagnosis through improved image analysis.
• To develop a model capable of real-time processing for clinical applications.

Main Methods:

• An encoder-decoder neural network architecture was designed.
• Deterministic mapping was employed in the encoder to create a hidden representation.
• Residual learning strategies with skip connections were utilized to optimize training.
• The model reconstructs denoised images from latent representations.

Main Results:

• The proposed model achieved an average Peak Signal-to-Noise Ratio (PSNR) of 38.38.
• The model achieved an average Structural Similarity Index Measure (SSIM) of 0.98.
• Performance surpassed existing state-of-the-art models on a dataset of 57,458 images.

Conclusions:

• The developed encoder-decoder denoiser offers superior performance in medical image denoising.
• This technology can significantly improve the accuracy and reliability of medical interpretations and diagnoses.
• The model's lightweight design and real-time processing capabilities are beneficial for various medical fields, including microscopy and surgery.