Synthesis of Mammogram From Digital Breast Tomosynthesis Using Deep Convolutional Neural Network With Gradient Guided...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Synthetic digital mammography (SDM) aims to reduce radiation dose by generating 2D images from digital breast tomosynthesis (DBT) volumes.
• Existing methods often alter image appearance compared to full-field digital mammography (FFDM) due to post-processing techniques.
• A learning-based approach is needed to accurately model the transformation from DBT to FFDM for SDM generation.

Purpose of the Study:

• To develop a deep convolutional neural network (DCNN) for generating SDM from DBT volumes.
• To improve the perceptual quality and preserve subtle mammographic features in SDM images.
• To investigate the effectiveness of different objective functions in enhancing SDM image quality.

Main Methods:

• A deep convolutional neural network (DCNN) was proposed to learn the transformation from DBT to FFDM.
• Gradient guided conditional generative adversarial networks (GGGAN) objective function was designed to preserve microcalcifications (MCs).
• Perceptual loss was incorporated to enhance the overall perceptual quality of the generated SDM.

Main Results:

• The DCNN demonstrated progressive performance improvements with different objective functions.
• Evaluation using image quality criteria showed enhanced preservation of masses and microcalcifications.
• The developed methodology effectively generated SDM with improved fidelity to FFDM.

Conclusions:

• The proposed DCNN with GGGAN and perceptual loss effectively generates synthetic digital mammography.
• The approach successfully preserves critical mammographic features, enhancing diagnostic potential.
• This methodology offers a promising direction for improving image quality in medical imaging generation tasks.