Potential challenges of incorporating artificial intelligence (AI) solutions into workflows of breast imaging departments

Background and aims: Breast cancer screening increases workload of radiologists. Recently,using AI solutions has been of great interest to radiology domain for decreasing this presser. Wereview the challenges that may slow or prevent AI adoption into workflows of breast imagingdepartments.Method: The reviewed articles in this study were collected from web of science, PubMed (MEDLINE)and GoogleScholar databases without time limit and based on the keywords “artificialintelligence”, “breast imaging” cancer, limitations, and challenges.Results: Some of the anticipated limitations for AI applications in breast imaging are as follow:The hospital environment of numerous integrated programs communicate poorly with PACS andother radiology information systems. The risk of innate latent bias is exist if the algorithms havebeen developed on datasets of certain populations. To maintain the trust of the public and avoidthe controversies such as inappropriate data sharing it is necessary to guide the development ofhuman-centric AI. The ability to track data back to the source via a flag-based system is necessary.However, such systems do not currently exist and would not be easy to integrate. With respect totechnical requirements, image analysis through big data needs the specific graphical processingunits. Also larger data storage capacity is desired. It is important to ensure that patients cannotbe re-identified via the possibilities of image reconstruction. Even If the better performance ofthe algorithms are verified across a range of imaging modalities, independent prospective benchmarkingagainst national criteria is needed. It is vital that commercial companies disclose thelimitations of their algorithms and training radiologists how to interpret. Advancing new technicalexpertise will require highly skilled staff who currently relocate to industry.Conclusion: AI’s solutions can utility in breast cancer diagnostic, but with possible algorithmicerror and without powerful evidence to support it, AI should not be relied on for physicians’decisions. There are several ethical, technical, legal and regulatory challenges facing to adoptionAI algorithms into workflows of breast imaging departments. For adopting AI into radiology,further research are require about economic implications, working according to the required performance,avoidance from the latent bias and also to provide the updated guidance for healthcareprofessionals to follow.

کلیدواژه ها:

artificial intelligence ، breast imaging ، cancer ، limitations ، challenges