The role of artificial intelligence in breast cancer screening as a supportive tool for radiologists

Abstract

Introduction
Breast cancer is the most prevalent cancer in Polish women and worldwide. Still remains the second cause of cancer deaths in females. There are several risk factors, both modifiable and non-modifiable. Despite the vast majority is an adenocarcinoma, multiple molecular variants are distinguished with different management and prognosis. Early diagnosis is crucial for cancer-related burden and mortality reduction. For this reason several countries have implemented breast cancer screening programme. This imaging has some limitations associated with interpretative difficulties, huge number of examinations and shortage of staff. Artificial intelligence has been proposed for breast screening, both in assistance to radiologists and independently in the future.
Purpose and methods
The aim of this review is to provide an overview on clinical use of AI in breast cancer screening with focusing on impact on results quality and workload. The evaluation has been focused on the degree of AI participation in diagnostic process and its impact on screening or human performance.
Results
Analysed studies presented various ways of AI application in breast screening. Authors generally focused on cancer detection rate, false positive or negative results, recall rate and the workload involvement. Some improvements, especially in cancer detection rate have been noticed with significant workload reduction. Difficulties, possible errors and people’s opinion were also highlighted.
Conclussion
AI algorithms find their potential application in breast cancer screening, mainly as a supportive tool for radiologists. However, in this moment, standalone AI interpretation does not seem to be accurate and safe enough, especially due to the lack of standarization and not fully convinced population. Future AI systems should take into account all relevant patients’ data for better assessment of the examination. More prospective studies are needed to improve a knowledge about AI possibilities.