Background: Breast Magnetic Resonance Imaging (MRI) is used for screening in women who are at increased risk of breast cancer, preoperative assessment of disease burden, and assessment of response to neoadjuvant chemotherapy. MRI is the most sensitive imaging modality for detection of breast cancer, but background enhancement can limit lesion identification.
An objective model of emerging machine learning (ML) techniques can now aid radiologist subjective interpretation of images.
Learning Objectives: Artificial Intelligence (AI) allows identification of features that may be beyond the scope of human detection. Use of AI and ML can improve cancer detection, screening guidance and prevention strategies, classification of cancers, and decrease scan time and cost. In this abstract, we will discuss the following AI applications and their use in interpreting breast MRI
• Radiomics application to breast MRI • Breast anatomic segmentation • Texture analysis
Abstract Content/Results: The learning objectives will be achieved through a pictoral review of how each AI application applies to interpretation of Breast MRI. A breast MRI case will demonstrate application of each AI technique.
Breast cancer heterogeneity undergoes temporal change and can contribute to treatment failure and poor prognosis. Biopsy can be limited due to the invasive nature of procedures and sampling errors. Use of AI and ML techniques such as radiomics can characterize both temporal and spatial characteristics of tumor through extracted data evaluation.
Radiomics involves pooling large numbers of quantitative features extracted from medical images to create decision support models. Automatic segmentation and feature extraction allows for more complete analysis (increased detection and characterization of sub centimeter lesions). Diffusion-weighted imaging and dynamic contrast-enhanced images can be used to analyze functional features of breast masses.
Breast Anatomic Segmentation involves identifying the breast-air boundary, breast-chest wall boundary, separation of fibroglandular tissue from fat, and increased levels of background parenchymal enhancement (BPE). Evaluating the breast- chest wall boundary can be difficult due to bias field artifact.
Texture analysis in radiomics is mathematically extracted quantitative statistical features of an image, including: grayscale intensity of pixels, relationship between these pixels in the x and y direction (Laws energy), edge detection after filter application, and co-occurrence matrix.
Conclusion: The performance of AI continues to improve, and its applications continue to expand. With respect to breast MRI, AI applications of radiomics, anatomic breast segmentation, and texture analysis provide improved breast MRI interpretation and offer the possibility not only for higher sensitivity and specificity for cancer detection but also for expanded clinical applications.
