The Role of AI in Medicine
John Koon posted on May 27, 2019 |

AI working with doctors: doctors can “train” AI software to increase its accuracy. (Image courtesy of Pixabay.)

AI working with doctors: doctors can “train” AI software to increase its accuracy. (Image courtesy of Pixabay.)

Artificial intelligence (AI) is the enabling of machines to “think” like humans. The ability of AI to make data-driven decisions (pattern recognition) and to identify shared characteristics among data points are especially relevant to medicine (data mining). Spurred on by the explosive expansion of the Internet of Things (IoT) and the decreasing cost of cloud storage and computing, the AI health-care market is likely to exceed $34 billion by 2025, according to a report by Tractica. In this article, we will explore the ever-expanding application of AI in medical diagnosis, drug and device development, and operational improvement. 

AI in Diagnosis

Image Analysis. IBM’s Watson was the first AI platform to enter the field of medical research. Since then, Intel, Google and Microsoft have formed academic partnerships for using AI to improve the imaging diagnostics in cancer screening, diabetic retinopathy, heart diseases, epidemics, and neurological disorders. There are many startups in this space: DeepGestalt uses AI to identify Hajdu-Cheney syndrome, which causes facial abnormalities; NeuroView Diagnostics and Brainomix focus on stroke diagnosis; MaxQ AI integrates patient data and the interpretation of medical images such as CT scans to increase diagnostic accuracy; Aier Eye Hospital Group in Chinaworks with Medimaging Integrated Solutions to screen patients for age-related macular degeneration and diabetic retinopathy; and Zhejiang University is working with Zhejiang DE Image Solutions to test women for thyroid cancer.

Voice Recognition. Corti has developed a technology that helps emergency dispatchers determine cardiac arrest based on the caller’s tone, spoken words and background noises. The software is more accurate and faster in its decision-making than humans, with a comparable error rate. Corti has started pilot studies in several EU countries, the U.S. and Asia.

AI will help researchers accelerate their work by mining massive datasets for associations between data points.
AI will help researchers accelerate their work by mining massive datasets for associations between data points.
AI in Drug and Device R & D

Digital Twin. AI can work with IoT devices to create a virtual replicant (a digital twin) of an organ for researchers. For example, Reuters has reported the development of a digital twin of a human heart with electrical and physiological properties. Researchers can run simulations on the digital twin to determine if a patient suffering from congestive heart failure needs a pacemaker or an operation. This research model has become part of a six-year clinical trial.

Data Mining. AI’s prowess in data mining can help accelerate drug development. Ariana Pharma’s KEM® artificial intelligence technology has found that Alzheimer’s patients with a specific genetic biomarker respond positively to the drug ANAVEX2-73, thus enabling the drug’s development for this group of patients. In addition, Recursion Pharmaceuticals is using AI to develop personalized medicine for Neurofibromatosis type 2, a rare hereditary cancer.

Clinical Trials. Drug development is slow and expensive process because most clinical trials recruit patients manually, often fail to enroll enough patients, and lose as much as 30 percent of the patient data due to participant dropout rates.Challenges also include patients not following treatment protocols and the often problematic data collection process.

Amazon is working with the Fred Hutchinson Cancer Research Center and Roche Diagnostics to use AI to help identify patients for clinical trials. In addition, Deep6 AI, Deep Lens,, SubjectWell, and PatientWing are active in the space of patient enrollment, while Medaptive Health, Brite Health, and Clinical Trial Connect are working to reduce patient dropout rates. AI can also help monitor patients and encourage them to follow treatment protocols. Companies working in this space include TowerView Health, Pillsy, Wellth, AdhereTech, MedMinder, Medisafe, emocha Mobile Health, and AiCure.

Bionics. In addition to helping with drug development, AI can assist in the design of smarter prosthetics. For example, the mind-controlled arm developed by University of Pittsburgh scientists can sense and react to pressure, Newcastle University researchers have developed a hand with a camera sensor and software that can make data-driven decisions, and  Endolite Linx can sense the user’s body position and adjust the person’s legs accordingly. AI can also help prosthetics interpret human intentions. A new bionic hand developed by researchers at the Imperial College of London can sense and analyze electrical signals from the amputee’s stump and instruct the bionic hand to move accordingly. 

AI in Operational Improvement

AI is helping doctors and hospitals re-engineer processes, triage patients, improve quality, and engage employees. 

Process Re-engineering. Working with Siemens, the Medical University of South Carolina created a digital twin of a stroke patient and another twin of its stroke center to analyze the workflow. Subsequently, by process re-engineering, the hospital achieved faster and more accurate diagnoses and reduced door-to-treatment time fourfold, thus leading to better patient outcomes and a reduction in care costs. Siemens also worked with the Mater Private Hospital in Ireland to redesign the layout and infrastructure of the hospital’s radiology department. Also using AI, Hospital del Mar in Barcelona has renovated its space to improve patient and employee well-being. 

Chatbots with AI. Chatbots, AI software that conducts a conversation via audio or text, can be used to reduce physicians’ workloads. Your.MD, Sensely, Buoy Health, Infermedica, and PACT Care BV are all startups that are developing chatbots, such as PACT Care’s Florence, which triage patients and make care recommendations to physicians.

Quality Improvement. Many companies are working to transform unstructured patient data into actionable steps. Pieces Tech and Prognos leverage AI to interpret patient data and recommend individualized treatment plans. KenSci and CareSkore use patient records to predict a patient’s health risk factors. Roam Analytics is working to integrate electronic medical records with patients’ comments and doctors’ notes to make treatment recommendations. Finally, is using AI to help medical teams coordinate stroke care to reduce wasted time and get patients to the appropriate care points.

Employee Well-Being. The implementation of electronic health record (EHR) systems has caused fatigue and burnout among health-care providers. Hospitals are looking to use AI to increase the usability of the EHRs. For example, Johns Hopkins Medicine is collaborating with Nuance to increase the engagement of health-care providers by easing the documentation burden and making real-time recommendations for diagnoses.


While AI has been very successful in the games of chess and Go, the effectiveness of AI in medicine is less clear. There has been some pushback that AI platforms, such as IBM’s Watson, do not show a substantial advantage over manual diagnosis. And AI arguably works better with input from humans—for example, having a doctor label some images to “teach” AI software what to look for will increase the software’s accuracy significantly more than feeding unlabeled datasets to the software. Also, research by Corti indicates that the best outcomes occur when machines work together with dispatchers.

Skepticism aside, public institutions and startups alike are jumping onto the AI bandwagon. The UK government is developing an algorithm to identify patients at risk of developing cancer using the medical, lifestyle and genetic data of patients available in government and nonprofit databases. It is also investing in improving the early diagnosis of lung and breast cancer. Finally, it is opening five new AI medical centers to accelerate disease diagnosis.

There remain technical challenges to the broader adoption of AI in medicine. Currently, AI analytics is done in the central cloud. However, data transfer between the point of a decision and the cloud can causea significant delay, which is problematic for applications that require real-time decision-making. Edge computing, where data analysis occurs close to the IoT device and at the site of data collection, is a possible solution. Also, there are challenges in filtering the data collected by IoT devices, integrating the datasets to yield actionable analytics, and maintaining data privacy. As AI matures, the field of medicine is expected to benefit from the technology, but this will occur gradually.

Recommended For You