Computer Vision News - June 2022

44 Artificial Intelligence for Medical Imaging Intelligence applications can be added to any breast screening modality: Mammography: during this exam, the breast is pressed between 2 surfaces, and a 2D X-ray image is acquired. Alternatively, tomosynthesis acquires a 3D tomography scan of the breast using similar technology. Applying a lower radiation dose will reduce image quality. However, neural networks can be trained to reconstruct high- quality images from low-quality ones . This technique is in use in CT scans, and can further reduce radiation exposure in mammography and tomosynthesis. Oncethe image(or tomography) isacquired, the radiologist is required to review it and search for suspicious lesions/sites in the breast for further examination. Both classic computer vision and more advanced deep learning methods can be used to detect and segment suspicious landmarks. MRI: a breast MRI exam is also used for Breast cancer is the 2nd most common cancer in women (after skin cancer). Risk of breast cancer increases with age, and is also affected by family history and genetic factors. Like most cancers, early detection increases survival rates dramatically . Therefore, breast screening for women is very common, and multi- imaging modalities are utilized. Current recommendations state that women over 40 should have a yearly mammography examination. Women at higher risk (family history of breast cancer, carriers of BRCA gene, etc.) undergo additional MRI and/or ultrasound exams. It has been shown that a significant portion of radiologist workload originates in breast exams, so any means to assist radiologists in reading these exams can help reduce physician fatigue and error rates. Additionally, some modalities like mammograms use harmful radiation for imaging, so any reduction of radiation dose is beneficial for the patient. Artificial ARTIFICIAL INTELLIGENCE IN BREAST CANCER SCREENING Breast Mammography - suspicious lesion is circled