AI’s future in Healthcare

Oct 06, 2019

The use of Artificial Intelligence (AI) has been on the rise for the past decade, due to its wide-ranging scope and potential impact on society. AI can help to revolutionize a plethora of deep-rooted industries, including healthcare. Healthcare policy has become a focal point of concern throughout the country and world. Unlocking AI’s potential within healthcare could bring a paradigm shift to the industry as we know it, sooner than later. Recent increases in the availability of healthcare data and accelerated development of data analytics methods have begun to enable AI applications to have successful integrations in healthcare. When paired with the correct clinical questions and data, AI techniques can bring forth clinically relevant information that would otherwise be hidden to the human eye, which can drive decision making in the healthcare space.

AI operates by using sophisticated algorithms to “learn” or obtain insights inside large volumes of data, in this case the extensive amount of healthcare data. These insights can be extremely helpful when assisting physicians and aiding decision making in practice. To ensure AI is as effective as possible, it is paramount that it is provided enough accurate data. This data is generated from clinical activities such as screening, diagnosis, or treatment assignment. Data is usually available in the form of demographics, medical notes, electronic recordings from medical devices, and imaging results. Physical examination notes and clinical laboratory results are also critical sources of data but are distinguished because they contain large portions of unstructured narrative text that must first be converted into machine-understandable electronic medical records (EMR).

AI devices use two major techniques to gain insights. The first category includes machine learning (ML) techniques. In medical applications, ML techniques attempt to cluster patients’ traits to infer the probability of disease outcomes. ML uses various algorithms including linear regression, logistic regression, naïves Bayes, decision trees, random forest, support vector machine (SVM) and neural networks--depending on the desired outcome and clinical data. The second category includes natural language processing (NLP) methods that work with unstructured data, such as clinical notes or medical journals. NLP is used to enrich the structured data sources, mentioned above, by extracting information from unstructured data sources.

Despite AI’s substantial growth in the healthcare space, currently the main areas of use are in a select few disease types: cancer, nervous system disease, and cardiovascular disease. All three of these diseases are leading causes of death, therefore any improvement in detection could have immense impacts throughout all of society. Early diagnosis for these diseases can potentially be achieved through improving procedures such as imaging, genetic, EP, and EMR, which, not coincidentally, all happen to be strengths of AI.

A quick example of AI’s “real world” application in healthcare is its usage in attempting to detect early diagnosis of strokes. Eighty-five percent of Thrombotic strokes are caused by a blood clot in the region of the brain called “cerebral infraction”. Because of the inability to diagnose this early, only a few patients can receive timely treatment. As a result, a movement detecting AI device was developed to aid in early stroke prediction. Two ML algorithms were implemented into the device to build the model for prediction. The data gathered from this machine was collected and further modeled by ML algorithms, such as SVM. The algorithm final output could then correctly classify 90.5% of at-risk patients correctly.

Although AI is attracting a significant amount of attention throughout the medical research community, real-world implementation into medical practice still faces some hurdles. Regulations could possibly hinder AI’s ability to reach its full potential in medical practice. Current regulations do not have standards to assess the safety and efficacy of AI. Regulations also do not address a possible ethical backlash of AI in healthcare, like that of which autonomous vehicles have faced. Another potential obstacle is assuring accurate and sufficient data exchange. For AI to work well, it needs to be continuously trained by clinical data. After an AI’s systems are deployed with initial training from historical data, a continual data supply becomes a critical factor for further deployments and improvements of the system. Unfortunately, the current healthcare environment does not provide any incentives for sharing data on the system, but a healthcare revolution could soon start to materialize, in which rewards are shifted from treatment volume to treatment outcome. Under this new environment all parties in the healthcare system would have a much greater incentive to compile and exchange information. Research suggests that harnessing the power of data and analytics will play a large role in determining its impact and ability to progress in the future.