Illustration of the MTLSegNet architecture with encoder, bottleneck, and task-specific decoders. The encoder unit extracts the low- and high-level imaging features from the input mammogram. The extracted features are then fed into the bottleneck unit. The bottleneck unit further enhances the field of view by extracting more complex and spatial information at different resolutions. The task-specific decoders segment the breast area and the dense tissues, simultaneously. The loss function (focal Tversky loss) is computed using the corresponding predicted and ground-truth segmentations of the breast area and the dense tissues. We modified the weight adaptive multi-task loss function for the segmentation task and computed the combined loss of the breast-area and dense-tissue segmentation tasks. The predicted segmentation outputs are overlapped with the original mammogram. The red contour line is the predicted segmented breast area, and the solid green pixels represent the predicted fibroglandular tissues within the breast area.
Breast density is an important factor in mammography because it can affect the sensitivity and specificity of the test. High breast density can make it harder to detect abnormalities on a mammogram, because the dense tissue can mask small tumors. In addition, false positive results are more common in women with dense breasts, because the dense tissue can be misinterpreted as a tumor.
There are several methods for estimating breast density from mammograms. One approach is to use a visual assessment, in which a radiologist looks at the mammogram and estimates the percentage of the breast that is composed of dense tissue. However, this method is subjective and can vary from one radiologist to another.
Another approach is to use automated methods to quantify the amount of dense tissue in the breast. These methods typically involve analyzing the mammogram to identify areas of dense tissue and calculating the percentage of the breast that is composed of dense tissue. The specific algorithms and techniques used to identify and quantify dense tissue can vary, but they may involve techniques such as image processing, machine learning, or pattern recognition.
Regardless of the method used, the goal of a breast density estimation model is to accurately and reliably quantify the amount of dense tissue in the breast, so that this information can be used to inform the interpretation of the mammogram and guide further testing or treatment if necessary.
Our AI team have recently developed a reliable and scalable deep learning model to estimate area-based breast density from screening mammograms.
Gudhe, N.R., Behravan, H., Sudah, M. et al. Area-based breast percentage density estimation in mammograms using weight-adaptive multitask learning. Sci Rep 12, 12060 (2022). https://doi.org/10.1038/s41598-022-16141-2
If you are interested in a project related to breast density estimation from mammograms, you can consider reaching out to us.
University of Eastern Finland, Institute of Clinical Medicine
Yliopistonranta 1 C, Canthia building, Kuopio, Finland
hamid.behravan[@]uef.fi
+358 50 5139836
Copyright 2023 – Hamid Behravan