Artificial intelligence is transforming the healthcare sector. A research team at the Stevens Institute of Technology designed and built an AI-powered tool that can detect early signs of Alzheimer's based on patients' speech patterns with an accuracy of 95%. 

To achieve similar results, medical organizations are looking to deploy AI solutions. This article explores artificial intelligence use cases in healthcare and explains how this technology can support doctors in their daily tasks and improve patient experience. We also warn of three barriers to AI in healthcare that can hinder adoption. 

Top five AI use cases in healthcare

Improving diagnostics accuracy

One of the main artificial intelligence applications in healthcare is radiology. This technology can support doctors in detecting and classifying different abnormalities.

Spotting hidden fractures

AI can detect fractures that are invisible to the human eye. For example, Imagen's OsteoDetect solution was approved by the FDA for detecting distal radius fractures in wrist scans after it displayed solid performance on 1,000 wrist images.

Classifying brain tumors

AI can not only produce accurate results but also reduce tumor classification time from 40 minutes required for the traditional intraoperative pathology analysis to merely three minutes. For instance, AI is used to spot and classify brain tumors located in the posterior fossa part of the brain in children. This type of tumor is among the leading causes of child death.

Transforming patient experience

Studies show that around 89% of people like to google their symptoms before scheduling an appointment with a doctor. Such research can generate scary results. Instead of turning to Google, one can contact an AI-powered virtual assistant that will help cope with the symptoms, monitor the patient's health parameters, and support them in managing chronic conditions.

Molly is one example of such a virtual assistant. It remotely supports patients with common medical conditions, addresses their concerns, and helps schedule an appointment with a physician of choice. 

Automating tedious repetitive tasks

Recent research reveals that physicians spend around 16 minutes per patient to fill in their electronic health records (EHRs). Technologies, such as AI-powered robotic process automation (RPA) in healthcare, can take this burden off the doctors' shoulders and free their time for more quality tasks.

These tools can transcribe recordings taken during patient examination, help search EHRs using voice commands, facilitate appointment scheduling, and even support insurance claim processing. With the help of RPA, Care1st Health Plan Arizona managed to bring down claim processing time from 20 seconds to a mere three.

Facilitating clinical trials

Clinical trials of prospective drugs last nine years on average and cost around $1.3 billion. And still, around 90% of the trialed drugs fail. This makes clinical trials an exciting application of AI in healthcare.

Designing clinical trials

This process is challenging as pharmaceutical companies need to look through vast amounts of unstructured data that is hard to analyze. AI can systematically aggregate and process this data to extract patterns. One example is that it can derive relevant regulatory protocols and ensure compliance, which, in turn, results in fewer protocol amendments over the course of the trial. Research shows that one significant amendment can make the trial run three months longer than initially planned and can cost up to $500,000 depending on the trial's phase.

Recruiting participants

AI can help find participants that match the trial's selection criteria. The technology analyzes data from EHRs and medical images to find participants with suitable characteristics. Ohio-based startup Deep Lens uses its extensive database of cancer studies to match patients to trials. In addition to benefiting pharmaceutical companies, it supports newly diagnosed patients by assigning them to trials fast.

Monitoring adherence

Adherence verification conventionally relies on patients' memory, which is prone to error. Artificial intelligence can record and monitor participants' actions. One such solution is presented by AiCure, a prominent AI company. It prompts patients to take a video of themselves swallowing a pill, and the AI component can verify that this is the right patient and the prescribed pill.

Catching on prescription errors

Up to 7,000 people perish annually in the US due to prescription errors. One AI application in healthcare is that it can analyze historic EHR data of a particular patient, compare to their new prescriptions, and highlight any inconsistencies. Brigham and Women's Hospital used such a system to identify erroneous prescriptions. This helped them save $1.3 million in healthcare costs in one year.

Limitations of AI in healthcare

Despite the exciting applications of artificial intelligence in healthcare, the technology has its limitations that hinder adoption. Here are the most prominent ones.

Bias

There are several types of bias that can be ingrained in AI algorithms or acquired as machine learning models continue to learn. At the moment, there are no regulations to govern algorithm development. So, vendors are expected to produce fair tools, but there are no consequences if the condition of fairness is not satisfied. It is best to test the model on a representative dataset before putting it to use, and conduct regular independent audits after deployment.

Difficulties associated with training data

Higher volumes of training data lead to more accurate algorithms. Unfortunately, obtaining medical data for training is a challenge due to privacy concerns. And the readily available datasets, such as LUNA dataset that contains CT images, are rather small. Additionally, labeling such training data is mostly manual, time-consuming, and tailored to a specific disease, and hence, can't be reused.

Lack of explainability

Many of AI-powered solutions operate as black-box models, meaning that it's not clear how they arrived at a particular decision. In the healthcare sector, this is barely acceptable, as doctors need to be able to explain to patients why they recommend one treatment over another, or why they believe a patient has a tumor. Medical facilities can opt for artificial intelligence healthcare companies that offer explainable AI solutions where algorithms back every decision with a justification.

Future trends of AI in healthcare

The global market of artificial intelligence in healthcare was valued at $6.9 billion in 2021 and it's expected to skyrocket to $67.4 billion by 2027 growing at a CAGR of 46.2%.

With all the exciting use cases of AI in healthcare, and the market growth rate, we can see that this technology is here to stay. And for doctors, who jeopardize their hospital's experiments with AI out of fear of being replaced, take a look at Bernard Marr's interview with Tom Lawry, National Director of AI for Health and Life Sciences at Microsoft. "What artificial intelligence is good at is things like pattern recognition," Tom said. "It's great at sifting through massive amounts of data to find something that humans either aren't capable of finding or would take years humans to find. On the other hand, humans are great at wisdom, common sense, empathy, and creativity."