Current status: Under review.
Abstract. AI in healthcare has been recognized by both – academia and industry in revolutionizing how healthcare services will be offered by healthcare service providers and perceived all stakeholders. This article reviews artificial intelligence (AI) techniques currently used for healthcare and their future impacts, and focusing on applications in robotics, precision medicine, medical image analysis. A comprehensive literature review of papers published over the five years prior to 2020 was performed. To narrow this study, the use of AI applications in the field of oncology was selected, because common problems that occur in cancer prevention, diagnosis, treatment, and rehabilitation are being solved with the help of state-of-the-art AI methods. This study provides a summary of the undergoing AI-related research in healthcare, and discusses challenges and open research questions for further research.
Summary and conclusions
AI in health can be introduced as a set of multiple technologies enabling machines to sense, understand, act, react, and study so they can perform clinical and administrative healthcare functions and truly augment human activity.
Much of the recent progress in AI has relied on data-driven techniques related to deep learning and artificial neural networks. Given sufficiently large labeled training data sets and computation resources, these approaches are achieving results that outperform human readers. As a result, big strides have been made in areas such as computer vision, speech recognition, and language translation with regards to “narrow AI” tasks.
However, a wider set of AI capabilities is required to progress it towards solving real-world challenges. In practice, AI systems need to learn effectively and efficiently without large amounts of data. They need to be robust, fair and explainable. They need to integrate knowledge and reasoning together with learning to improve performance and enable sophisticated capabilities.
Robotics have taken amazing leaps in improving the healthcare services in a variety of medical sectors including oncology and surgical interventions. In addition, robots are now replacing human assistants as they have learned to be more social and reliable. However, there are still challenges to address to enable the use of medical robots in diagnosis and interventions.
The studies related to the applications of AI in precision medicine and oncology have reemphasized the importance of pathogenic germline mutations in cancers, even among patients lacking a notable family history of cancer. There have been notable cases of patients with a change in diagnosis or risk stratification due to genomic aberrations discovered on molecular testing.
In medical image analysis, the lack of data is a big problem. There is generally a lack of publicly available data, and high quality labelled data is even scarcer. Most of the datasets presented in the research papers involve a small number of patients. Therefore, the model may not be generalizable. The CapsNets look promising for the medical imaging community from this perspective as they show better performance comparing to CNNs utilizing much smaller datasets.
In reality, AI can never be a substitute for a human doctor. However, humans are fallible, and so is AI. Mistakes in healthcare are often costly, financially, physically, and emotionally. Therefore, AI could provide doctors more time and energy for the human aspects of health care. Keeping all these challenges in mind, we still believe that AI will allow us to unlock the value in data in entirely new, profound ways. In a few years, AI systems may be involved in every workflow in healthcare, from population health management to digital interfaces that are capable of answering specific patient questions in real-time basis.