Radiation Dose Reduction in Digital Mammography by Deep-Learning Algorithm Image Reconstruction: A Preliminary Study

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence

Background:

• Low-dose mammography is crucial for breast cancer screening.
• Image quality and lesion detectability can be compromised at reduced radiation doses.
• Advanced image processing techniques are needed to enhance low-dose mammograms.

Purpose of the Study:

• To develop and evaluate a denoising convolutional neural network (CNN) for breast cancer diagnosis.
• To assess the efficacy of the CNN in improving image quality and lesion detection in low-dose mammography.
• To investigate the potential for radiation dose reduction in mammography through AI-based image processing.

Main Methods:

• A prospective study involving 6 breast radiologists evaluating low-dose mammograms.
• Radiologists assessed lesion detection and image quality before and after applying a denoising CNN.
• Comparison of low-dose, 40% reconstructed full-dose, and 100% full-dose images for diagnostic performance.

Main Results:

• Denoising processing improved lesion perception on low-dose images compared to mastectomy specimens.
• Full-dose images received higher ratings for resolution and diagnostic quality of calcifications, masses, and distortions than 40% reconstructed images.
• 40% reconstructed images demonstrated comparable overall quality, lesion visibility, and contrast to 100% full-dose images, indicating no significant differences.

Conclusions:

• Denoising and image reconstruction techniques using CNNs can significantly reduce radiation dose in mammography.
• AI-powered image processing holds promise for effective breast cancer diagnosis with lower radiation exposure.
• This approach can maintain diagnostic accuracy while minimizing patient risk.