Leaders in healthcare are increasingly turning to AI to improve diagnosis and screening processes. The technology is now capable of delivering real-time personalised patient care and is bringing the promise of precision medicine ever closer to reality.

The Stack recently spoke to Bhushan Desam, global director of AI business at Lenovo DCG, and Dario Garcia-Gasulla, senior researcher of the High Performance Artificial Intelligence research group at Barcelona Supercomputing Center, to investigate AI’s potential impact in healthcare.

Listing a wide range of potential use cases, Desam suggests that AI will soon help doctors; prescribe drugs and dosages that are more likely to work for a specific patient, match patients to clinical trials, prioritise care in emergency rooms, and even provide care to underserved populations where there are significant shortages of qualified healthcare providers.

In some countries, he notes, medical malpractice accounts for up to 2.4% of total healthcare costs, increasing expense by billions.

In addition to improving quality, AI also has the potential to significantly reduce error rates and malpractice cases by physicians, therefore reducing overall costs.

According to Garcia-Gasulla, machine learning models can process huge amounts of data using high compute resources and help to facilitate this work for humans. For example, through the analysis of data, healthcare leaders can detect patterns and make decisions to improve healthcare.

Applying AI during routine exams can help detect symptoms of visual impairments much earlier and with higher accuracy

“Currently, the most successful integrations are through intelligent decision support systems. These can be used as tools by healthcare leaders to make it easier to identify the easiest cases and detect potential mistakes, all of which speeds the process of screening and diagnosis,” he explains.

AI developments in eye disease and beyond

As a specific area of development, Desam speaks of the significant effort that is going into analysing and understanding retinal images using AI, particularly with deep learning techniques.

While image-based diagnosis has been around for many years, deep learning is improving the accuracy with which images are interpreted, forging a path to the eventual widespread use of the technology in routine clinical care.

In ophthalmology, he believes that applying AI during routine exams can help detect symptoms of visual impairments much earlier and with higher accuracy, and will also help to provide quicker and more reliable care to patients.

Advances in AI and its eventual adoption in the clinic will revolutionise the entire healthcare system

Garcia-Gasulla adds that, in recent years, AI has progressed significantly in the field of image processing:

“Most image related tasks can be tackled with AI models with a high degree of reliability. The main requirement in this process is having access to large data sets. In the case of ophthalmology, fundus retina images are very cheap to obtain, and the procedure is non-intrusive, as anyone with a smartphone could obtain these images.

“This is a perfect case for an application, in which everyone could periodically take a picture of their eye to find early signs of disease, enabling earlier detection of pathologies, and more timely treatment of conditions.”

Beyond this field, the experts argue that advances in AI and its eventual adoption in the clinic will revolutionise the entire healthcare system, including for drug manufacturers, insurance companies, and healthcare organisations, such as how AI can identify better drug formulations and select the right patients for drug trials, helping accelerate drug discovery for diseases, such as cancer.

Healthcare is a highly regulated industry, so there are a lot of restrictions that prevent healthcare providers from using patient data for AI discovery

AI could also help process electronic health records (HER), which contain a vast repository of disease and treatment data that could be mined to increase medical insight and improve patient outcomes.

Transfer of learning and other challenges

One of the challenges of AI is increasing the versatility and generality of its methods. Garcia-Gasulla explains that AI solutions work best when there is little variance, as these AI models strongly specialise in a certain task.

However, it is desirable to have AI models that can learn from various datasets, even when these data are not directly related to a particular problem.

“Although there is some progress in this direction, looking ahead, the field of transfer learning will be very promising. Transfer learning explores how to reuse what is learnt by an AI model to solve different sets of tasks,” he says.

AI technologies also require vast volumes of patient data to help develop algorithms, adds Desam. In many countries, healthcare is a highly regulated industry, so there are a lot of restrictions that prevent healthcare providers from using patient data for AI discovery and model development. In addition, these data must be of high-quality to get the most value out of it.

He continues: “Another major challenge is the transparency of AI algorithms, especially those based on deep learning where it is difficult to explain how the algorithm arrived at that prediction. Traditional statistical modeling, which is more widely known to researchers and clinicians is inherently more transparent.

Therefore, in the upcoming years, it will be important for healthcare providers to be open to using these unfamiliar AI tools given the incredible potential benefit they can deliver from bench to clinic.”

Future of disease diagnosis and screening

Looking to the future, Desam concludes that with advances in Internet of Things (IoT) and wearables, patients could choose to be monitored continuously as they go about their day to improve diagnosis and eventually get real-time feedback of medication or behavioural changes to improve care management.

“Since data privacy is a significant concern in healthcare, we could also see smart contracts through blockchain technology become a method of securely transferring patient data, and sharing that data with relevant parties. While blockchain technology is still in its infancy, its impact will grow within healthcare in the coming years.”