The use of artificial intelligence (AI) in radiology is rapidly evolving, with many possible applications. AI has been applied to different breast imaging modalities, including mammography, ultrasound, and magnetic resonance imaging (MRI), in different clinical scenarios. Breast cancer screening is perhaps the best known and most researched use case. The application of AI tools in breast imaging has an unprecedented opportunity to better derive clinical value from imaging data and reshape the way we care for our patients.
Despite mammography is probably the imaging modality with more data available for breast cancer screening, radiomics and AI studies are rapidly evolving to improve the assessment of breast MRI in different applications, including breast cancer risk prediction, lesion detection and classification. The aim of this presentation is to review the current knowledge and future applications of AI on MRI for breast cancer screening and risk prediction.
Fully automated detection of breast cancer on screening MRI using CNN has been shown to be possible, not only for systematic diagnostic interpretation but also for identifying tumor-containing slices stored on picture archiving and communication systems.
Dalmis et al. (2018) were one of the first authors who used deep learning to develop a CADe system that exploits the spatial information obtained from the early-phase scans, which can be applied to screening programs where abbreviated MRI protocols are used.
Herent et al. (2019) developed a deep learning model that simultaneously learns to detect lesions and characterize them. The model was based on a single two-dimensional T1-weighted fat suppressed MR image obtained after intravenous administration of a gadolinium chelate selected by radiologists. Training set included 335 MR images and the algorithm performance was evaluated on an independent test set of 168 MR images, in which the model reached a weighted mean AUC of 0.816.
Adachi et al (2020) evaluate an AI system to detect and diagnose lesions of maximum intensity projection (MIP) in DCE-MRI (training set: 214; validation set: 72; and testing set: 85). The AI system showed better diagnostic performance compared to the human readers (0.925 x 0.884 ; p = 0.002), and because of the increased performance of human readers with the assistance of the AI system, the AUC of human readers was significantly higher with than without the AI system (0.899 x 0.884; p = 0.039).
Zhou et al. (2019) evaluated the efficacy of 3D deep convolutional neural network (CNN) for diagnosing breast cancer and localizing the lesions at dynamic contrast enhanced (DCE) MRI data in a weakly supervised manner. A total of 1537 cases were included (training set: 1073; validation set: 157; and testing set: 307). The final algorithm performance for breast cancer diagnosis showed an AUC of 0.859. The weakly supervised learning method showed promise for localizing lesions in volumetric radiology images with only image-level labels, enabling a fully automatic pipeline, from breast MRI preprocessing to malignancy likelihood prediction and cancer annotating.
Dalmis et al. (2019) investigated artificial intelligence (AI)–based classification of benign and malignant breast lesions imaged with a multiparametric breast magnetic resonance imaging (MRI) protocol with DCE-MRI, T2-weighted, and DWI. 576 lesions were included. The AUC was 0.811 when only DCE-MRI information was used. However, the final AI system that combined all imaging and patient information resulted in an AUC of 0.852, significantly higher than the DCE-MRI alone (P = 0.002).
Liu et al. (2022) evaluated a weakly supervised deep learning approach to breast MRI assessment without pixel level segmentation, which achieved an AUC of 0.92 (SD 0.03) in distinguishing malignant from benign images.
Eskreis-Winkler et al. (2021) used deep learning to identify tumor-containing axial slices on breast MRI images of 273 breast cancer patients. The accuracy of the deep learning system for tumor detection was 92.8%.
Hirsch et al. (2022) developed a deep network architecture that would achieve fully automated radiologist-level segmentation of cancers at breast MRI. More than 38000 breast MRI exams were evaluated, which consisted of 2555 malignant and 60108 benign breast scans. The performance of the network was equivalent to that of the radiologists.
Zhang et al. (2022) implemented a deep learning Mask Regional Convolutional Neural Network (R-CNN) to search the entire set of images and detect suspicious lesions. They included 241 patients acquired using nonfat-sat sequence for training, and 98 patients acquired using fat-sat sequence for testing. The accuracy in the test dataset was 0.75. The used of both precontrast and subtraction images minimize the false-positive results from normal background parenchymal.
Lastly, breast MRI has also been proposed as a tool for breast cancer risk prediction which is relevant, for example, to define screening schemes. Portnoi et al. (2019) developed an image-based DL model to predict the 5-year breast cancer risk on the basis of a single breast MRI from a screening examination and showed that this model improved individual risk discrimination when compared with a state-of-the-art risk assessment model.
In conclusion, preliminary data suggest that application of AI can help to improve the access and results of breast cancer screening and risk prediction using MRI.

Acknowledgements

No acknowledgement found.