Visual and quantitative evaluation of microcalcifications in mammograms with deep learning-based super-resolution

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence

Background:

• Digital mammography is crucial for breast cancer screening.
• Microcalcifications are early indicators of breast cancer.
• Enhancing microcalcification visibility can improve diagnostic accuracy.

Purpose of the Study:

• To evaluate a deep learning-based super-resolution (SR) model for microcalcifications in digital mammography.
• To assess both visual and quantitative performance of the SR model.

Main Methods:

• A fast SR convolutional neural network was applied to mammograms from 93 patients with microcalcifications.
• Four breast radiologists performed visual evaluation using a 5-point scale.
• Quantitative evaluation used a perception-based image-quality evaluator (PIQE).

Main Results:

• Radiologists rated SR mammograms superior for microcalcification detection, diagnostic quality, contrast, and sharpness (p < .001).
• Noise scores were significantly lower in SR mammograms (p < .001).
• PIQE scores indicated significantly better image quality for SR mammograms (p < .001).

Conclusions:

• Deep learning-based SR models can enhance microcalcification visibility in mammography.
• Improved visibility aids in the detection and diagnosis of breast cancer.