Deep Perceptual Enhancement for Medical Image Analysis

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Medical image acquisition hardware limitations often result in low-quality images (e.g., low contrast, poor brightness, noise).
• Perceptually degraded medical images complicate diagnosis and clinical decision-making for healthcare professionals.

Purpose of the Study:

• To propose an end-to-end deep learning strategy for enhancing low-quality medical images.
• To accelerate medical image analysis tasks through improved image quality.

Main Methods:

• A fully convolutional deep network incorporating residual blocks and a residual gating mechanism was developed.
• The network addresses perceptual enhancement, including contrast correction, luminance correction, and denoising.
• A multi-term objective function guides the network to produce perceptually plausible enhanced images.

Main Results:

• The proposed method demonstrated superior performance over existing techniques across various medical image modalities.
• Quantitative improvements were observed, with gains of 5.00–7.00 dB in peak signal-to-noise ratio (PSNR) and 4.00–6.00 in DeltaE metrics.
• Significant enhancement in medical image analysis task performance was achieved.

Conclusions:

• The developed deep learning method effectively enhances low-quality medical images, addressing multiple perceptual degradations.
• This approach offers a promising solution for improving diagnostic accuracy and efficiency in clinical settings.
• The method shows potential for real-world applications in medical image analysis.