Thursday 13 March 2025
Researchers have made a significant discovery about the way mammography machines process images, which could impact the accuracy of artificial intelligence (AI) used in breast cancer screening.
The study, which analyzed data from over 500 mammography systems across the UK, found that each system has its own unique settings for image processing. This means that images taken on different machines can look quite different, even if they’re supposed to be displaying the same information.
One of the key findings was that each machine has multiple software versions and processing settings, which can affect how images are displayed. For example, some systems may have more aggressive noise reduction algorithms, while others may have different contrast levels. These variations can make it difficult for AI algorithms to accurately interpret the images.
The researchers also found that physical differences in the machines themselves can impact image quality. For instance, some mammography units use linear grids to reduce scatter radiation, while others use 2D anti-scatter grids. These differences can affect how much radiation is absorbed by the detector and ultimately impact the quality of the image.
The study’s findings have significant implications for the use of AI in breast cancer screening. Currently, AI algorithms are trained on images taken from a specific machine or group of machines. However, if the settings on these machines differ significantly from those used in clinical practice, it could lead to inaccurate diagnoses.
To address this issue, researchers suggest that AI algorithms should be re-trained on a larger dataset that includes images from multiple machines and processing settings. This would help to ensure that the AI is accurate and reliable across different clinical environments.
The study’s authors also recommend that mammography units take steps to standardize their image processing settings and physical configurations. This could involve implementing quality control measures to ensure that all machines are set up consistently, or developing new software algorithms that can adapt to different machine settings.
Overall, the study highlights the importance of considering the complexities of mammography imaging when developing AI-powered breast cancer screening tools. By taking a more nuanced approach to image processing and machine configuration, researchers hope to improve the accuracy and reliability of these life-saving technologies.
Cite this article: “Mammography Machine Variations Complicate AI Accuracy in Breast Cancer Screening”, The Science Archive, 2025.
Mammography, Artificial Intelligence, Breast Cancer Screening, Image Processing, Machine Learning, Radiation, Detector, Scatter Radiation, Anti-Scatter Grids, Linear Grids
