Artificial intelligence in healthcare

AI is being studied within the radiology field to detect and diagnose diseases within patients through Computerized Tomography (CT) and Magnetic Resonance (MR) Imaging.[25] The focus on Artificial Intelligence in radiology has rapidly increased in recent years according to the Radiology Society of North America, where they have seen growth from 0 to 3, 17, and overall 10% of total publications from 2015-2018 respectively.[25] A study at Stanford created an algorithm that could detect pneumonia in patients with a better average F1 metric (a statistical metric based on accuracy and recall), than radiologists involved in the trial. Through imaging in oncology, AI has been able to serve well for detecting abnormalities and monitoring change over time; two key factors in oncological health.[26] Many companies and vendor neutral systems such as icometrix, QUIBIM, Robovision, and UMC Utrecht’s IMAGRT have become available to provide a trainable machine learning platform to detect a wide range of diseases. The Radiological Society of North America has implemented presentations on AI in imaging during its annual conference.[25] Many professionals are optimistic about the future of AI processing in radiology, as it will cut down on needed interaction time and allow doctors to see more patients.[26] Although not always as good as a trained eye at deciphering malicious or benign growths, the history of medical imaging shows a trend toward rapid advancement in both capability and reliability of new systems.[26] The emergence of AI technology in radiology is perceived as a threat by some specialists, as it can improve by certain statistical metrics in isolated cases, where specialists cannot.

Recent advances have suggested the use of AI to describe and evaluate the outcome of maxillo-facial surgery or the assessment of cleft palate therapy in regard to facial attractiveness or age appearance.[27][28]

In 2018, a paper published in the journal Annals of Oncology mentioned that skin cancer could be detected more accurately by an artificial intelligence system (which used a deep learning convolutional neural network) than by dermatologists. On average, the human dermatologists accurately detected 86.6% of skin cancers from the images, compared to 95% for the CNN machine.[29]

In January 2020 researchers demonstrate an AI system, based on a Google DeepMind algorithm, that is capable of surpassing human experts in breast cancer detection.[30][31]

In July 2020 it was reported that an AI algorithm by the University of Pittsburgh achieves the highest accuracy to date in identifying prostate cancer, with 98% sensitivity and 97% specificity.[32][33]

In psychiatry, AI applications are still in a phase of proof-of-concept.[34] Areas where the evidence is widening quickly include chatbots, conversational agents that imitate human behaviour and which have been studied for anxiety and depression.[35]

Challenges include the fact that many applications in the field are developed and proposed by private corporations, such as the screening for suicidal ideation implemented by Facebook in 2017.[36] Such applications outside the healthcare system raise various professional, ethical and regulatory questions.[37]

Primary care has become one key development area for AI technologies.[38][39] AI in primary care has been used for supporting decision making, predictive modelling, and business analytics.[40] Despite the rapid advances in AI technologies, general practitioners' view on the role of AI in primary care is very limited–mainly focused on administrative and routine documentation tasks.[41][42]

An article by Jiang, et al. (2017) demonstrated that there are several types of AI techniques that have been used for a variety of different diseases, such as support vector machines, neural networks, and decision trees. Each of these techniques is described as having a “training goal” so “classifications agree with the outcomes as much as possible…”.

To demonstrate some specifics for disease diagnosis/classification there are two different techniques used in the classification of these diseases include using “Artificial Neural Networks (ANN) and Bayesian Networks (BN)”. It was found that ANN was better and could more accurately classify diabetes and CVD.

Through the use of Medical Learning Classifiers (MLC’s), Artificial Intelligence has been able to substantially aid doctors in patient diagnosis through the manipulation of mass Electronic Health Records (EHR’s).[43] Medical conditions have grown more complex, and with a vast history of electronic medical records building, the likelihood of case duplication is high.[43] Although someone today with a rare illness is less likely to be the only person to have suffered from any given disease, the inability to access cases from similarly symptomatic origins is a major roadblock for physicians.[43] The implementation of AI to not only help find similar cases and treatments, but also factor in chief symptoms and help the physicians ask the most appropriate questions helps the patient receive the most accurate diagnosis and treatment possible.[43]

An elderly man using a pulse oximeter to measure his blood oxygen levels

The increase of telemedicine, the treatment of patients remotely, has shown the rise of possible AI applications.[44] AI can assist in caring for patients remotely by monitoring their information through sensors.[45] A wearable device may allow for constant monitoring of a patient and the ability to notice changes that may be less distinguishable by humans. The information can be compared to other data that has already been collected using artificial intelligence algorithms that alert physicians if there are any issues to be aware of.[45]

Another application of artificial intelligence is in chat-bot therapy. Some researchers charge that the reliance on chat-bots for mental healthcare does not offer the reciprocity and accountability of care that should exist in the relationship between the consumer of mental healthcare and the care provider (be it a chat-bot or psychologist), though.[46]

Since the average age has risen due to a longer life expectancy, artificial intelligence could be useful in helping take care of older populations.[47] Tools such as environment and personal sensors can identify a person’s regular activities and alert a caretaker if a behavior or a measured vital is abnormal.[47] Although the technology is useful, there are also discussions about limitations of monitoring in order to respect a person’s privacy since there are technologies that are designed to map out home layouts and detect human interactions.[47]

Electronic health records (EHR) are crucial to the digitalization and information spread of the healthcare industry. Now that around 80% of medical practices use EHR, the next step is to use artificial intelligence to interpret the records and provide new information to physicians.[48] One application uses natural language processing (NLP) to make more succinct reports that limit the variation between medical terms by matching similar medical terms.[48] For example, the term heart attack and myocardial infarction mean the same things, but physicians may use one over the over based on personal preferences.[48] NLP algorithms consolidate these differences so that larger datasets can be analyzed.[48] Another use of NLP identifies phrases that are redundant due to repetition in a physician’s notes and keeps the relevant information to make it easier to read.[48]

Beyond making content edits to an EHR, there are AI algorithms that evaluate an individual patient’s record and predict a risk for a disease based on their previous information and family history.[49] One general algorithm is a rule-based system that makes decisions similarly to how humans use flow charts.[50] This system takes in large amounts of data and creates a set of rules that connect specific observations to concluded diagnoses.[50] Thus, the algorithm can take in a new patient’s data and try to predict the likeliness that they will have a certain condition or disease.[50] Since the algorithms can evaluate a patient’s information based on collective data, they can find any outstanding issues to bring to a physician’s attention and save time.[49] One study conducted by the Centerstone research institute found that predictive modeling of EHR data has achieved 70–72% accuracy in predicting individualized treatment response.[51] These methods are helpful due to the fact that the amount of online health records doubles every five years.[49] Physicians do not have the bandwidth to process all this data manually, and AI can leverage this data to assist physicians in treating their patients.[49]

Improvements in natural language processing led to the development of algorithms to identify drug-drug interactions in medical literature.[52][53][54][55] Drug-drug interactions pose a threat to those taking multiple medications simultaneously, and the danger increases with the number of medications being taken.[56] To address the difficulty of tracking all known or suspected drug-drug interactions, machine learning algorithms have been created to extract information on interacting drugs and their possible effects from medical literature. Efforts were consolidated in 2013 in the DDIExtraction Challenge, in which a team of researchers at Carlos III University assembled a corpus of literature on drug-drug interactions to form a standardized test for such algorithms.[57] Competitors were tested on their ability to accurately determine, from the text, which drugs were shown to interact and what the characteristics of their interactions were.[58] Researchers continue to use this corpus to standardize the measurement of the effectiveness of their algorithms.[52][53][55]

Other algorithms identify drug-drug interactions from patterns in user-generated content, especially electronic health records and/or adverse event reports.[53][54] Organizations such as the FDA Adverse Event Reporting System (FAERS) and the World Health Organization's VigiBase allow doctors to submit reports of possible negative reactions to medications. Deep learning algorithms have been developed to parse these reports and detect patterns that imply drug-drug interactions.[59]

DSP-1181, a molecule of the drug for OCD (obsessive-compulsive disorder) treatment, was invented by artificial intelligence through joint efforts of Exscientia (British start-up) and Sumitomo Dainippon Pharma (Japanese pharmaceutical firm). The drug development took a single year, while pharmaceutical companies usually spend about five years on similar projects. DSP-1181 was accepted for a human trial.[60]

In September 2019 Insilico Medicine reports the creation, via artificial intelligence, of six novel inhibitors of the DDR1 gene, a kinase target implicated in fibrosis and other diseases. The system, known as Generative Tensorial Reinforcement Learning (GENTRL), designed the new compounds in 21 days, with a lead candidate tested and showing positive results in mice.[61][62][63]

The same month Canadian company Deep Genomics announces that its AI-based drug discovery platform has identified a target and drug candidate for Wilson's disease. The candidate, DG12P1, is designed to correct the exon-skipping effect of Met645Arg, a genetic mutation affecting the ATP7B copper-binding protein.[64]

In order to effectively train Machine Learning and use AI in healthcare, massive amounts of data must be gathered. Acquiring this data, however, comes at the cost of patient privacy in most cases and is not well received publicly. For example, a survey conducted in the UK estimated that 63% of the population is uncomfortable with sharing their personal data in order to improve artificial intelligence technology.[72] The scarcity of real, accessible patient data is a hindrance that deters the progress of developing and deploying more artificial intelligence in healthcare.

According to a recent study, AI can replace up to 35% of jobs in the UK within the next 10 to 20 years.[73] However, of these jobs, it was concluded that AI has not eliminated any healthcare jobs so far. Though if AI were to automate healthcare related jobs, the jobs most susceptible to automation would be those dealing with digital information, radiology, and pathology, as opposed to those dealing with doctor to patient interaction.[73]

Automation can provide benefits alongside doctors as well. It is expected that doctors who take advantage of AI in healthcare will provide greater quality healthcare than doctors and medical establishments who do not.[74] AI will likely not completely replace healthcare workers but rather give them more time to attend to their patients. AI may avert healthcare worker burnout and cognitive overload

AI will ultimately help contribute to progression of societal goals which include better communication, improved quality of healthcare, and autonomy.[75]

Since AI makes decisions solely on the data it receives as input, it is important that this data represents accurate patient demographics. In a hospital setting, patients do not have full knowledge of how predictive algorithms are created or calibrated. Therefore, these medical establishments can unfairly code their algorithms to discriminate against minorities and prioritize profits rather than providing optimal care.[76]

There can also be unintended bias in these algorithms that can exacerbate social and healthcare inequities.[76]  Since AI’s decisions are a direct reflection of its input data, the data it receives must have accurate representation of patient demographics. White males are overly represented in medical data sets.[77] Therefore, having minimal patient data on minorities can lead to AI making more accurate predictions for majority populations, leading to unintended worse medical outcomes for minority populations.[78] Collecting data from minority communities can also lead to medical discrimination. For instance, HIV is a prevalent virus among minority communities and HIV status can be used to discriminate against patients.[77]  However, these biases are able to be eliminated through careful implementation and a methodical collection of representative data.