31 A Comparison of Deep Learning Artificial Intelligence Methods with Radiologists in Breast Cancer Screening for Guiding Patient Management

Background Breast cancer is the most common type of cancer in women. Mammograms are the most used method of screening; however, this technique has limited sensitivity. According to the USPSTF, the screening age for breast cancer in the US has reduced by 10 years which will produce thousands more images, increasing the demand on radiologists. Aim AI algorithms aim to assist radiologists and detect cancers that would have otherwise been missed. The aim of the integration of AI into screening is to enable quicker breast cancer detection, leading to earlier treatment which may include surgery in some cases. Results Recent studies which compare the performance of Deep learning, Convolutional neural network (ResNET) models with radiologists in breast cancer screening, have shown an increased accuracy of malignancy detection when radiologists read images in combination with AI than radiologists reading the scans alone. Discussion Radiogenomics could be utilised to predict the likelihood of patients developing breast cancer, taking into consideration patient factors and image data, aiming towards precision medicine. However, there are some limitations with the use of AI in the screening programme, such as access to large quality data, data privacy, transparency and explainability. Conclusions AI has the potential to shift breast cancer detection to a more proactive than reactive approach, increasing the speed and accuracy of diagnosis and contributing to the establishment of a more patient-centred approach to treatment. However, more research is needed into this field and ethical considerations must be evaluated prior to implementing AI fully within the screening programme.