How AI Can Enhance Diagnostics and Treatment Plans

Artificial intelligence (AI) plays a role in almost every aspect of our lives. It works behind the scenes to recommend books to read, TV shows to watch, or to quickly find photos of our pets and children and is an integral part of the technology used to create autonomous features in today’s vehicles. But one of AI’s most important roles is certainly within the healthcare arena.

“AI tools can rapidly collect and synthesize relevant data and make clinical recommendations based on full available information with a methodological precision and thoroughness that humans cannot,” said Ashok Chennuru, global chief data and insights officer, Carelon Digital Platforms. “This level of granular information allows for personalized patient care planning.” And it has the potential to revolutionize healthcare.

In the same way that predictive mathematical models drive decision-making in the financial sector (which stocks to buy, how much to purchase, and when to sell), AI provides doctors with tools to guide their decision-making. This can happen in every stage of the treatment process, from risk assessment and diagnosis through disease management and prognosis.
 

AI and Diagnosis

With AI, scientists feed a computer algorithm thousands of data points or images, training it to differentiate between diseased and healthy patterns. The more data it receives, the more the algorithm learns and the better it performs — with specific processes in place to ensure that it can identify and mitigate adverse bias across the life cycle of the algorithm.  

If what you’re looking to identify is one of the nation’s top health concerns, this technology is a gamechanger. Both heart disease and cancer change how tissue behaves physically, and increasingly, researchers can chart these differences and train an algorithm to detect them. AI may even hold the key to determining whether someone is at risk of developing disease. Take breast cancer, for example. With AI, doctors can convert mammograms from an image that detects breast cancer into a risk assessment tool that provides people with actionable information about their personal breast health.

“It’s that component of machine learning that allows us to identify disease before it can be spotted by trained physicians,” said Dr. Mark Traill, director of Medical Imaging AI Projects at University of Michigan Health-West. “Already, we’re using AI risk algorithms to go deeper into a standard 3D mammogram and identify patterns that suggest a person is at risk of developing an aggressive breast cancer over the next 12 months. These algorithms are astonishingly accurate, often outperforming trained radiologists. Since the miss rate for breast cancer is up to 35% with a standard mammogram, AI becomes especially important both for catching cancers at the earliest stages and preventing false positives.”

The first Food and Drug Administration-cleared system for 3D mammography, ProFound AI™, not only accurately analyzes each tomosynthesis image, detecting both malignant soft tissue densities and calcifications, but it also cuts reading time for radiologists in half. And doctors are working feverishly to integrate AI technology into screening protocols for all types of cancer. The FDA continues to update its list of AI/machine learning-enabled medical devices.

Two examples where AI shows promise:

• Lung cancer: In a study of more than 42,000 low-dose computed tomography scans (LDCT), AI performed as well or better than six radiologists in its ability to detect lung cancer tumors.
  
• Pancreatic cancer: In a proof-of-concept study at Johns Hopkins’ Sidney Kimmel Comprehensive Cancer Center, a machine learning tool called CompCyst outperformed current clinical practice in identifying pancreatic cysts that showed no risk of cancer, those that harbored cancer, and those that required immediate surgery.

Cancer isn’t the only target in this space. Doctors are increasingly using AI to identify diseases sooner and improve outcomes. AI can detect brain changes indicative of Alzheimer’s disease and stroke, analyze the heart’s electrical activity to spot current problems and predict future events, and diagnose osteoporosis and assess fracture risk from hip X-rays.
 

AI for Disease Management

Smartphone technology, video capabilities, even sophisticated apps and monitoring devices are changing the way individuals manage their health. They’re also producing a treasure trove of real-world data for AI applications, when used responsibly and in a way that protects privacy. With the convergence of AI and the information, shared with consent from individuals, the hope is that doctors can deliver truly personalized, more advanced medicine.

“These tools even allow us to take into consideration patient preferences about things like medication delivery and side effects,” said Dr. Chethan Sarabu, director of clinical informatics at Sharecare and a physician at Stanford Medicine.

AI and deep learning models may be especially useful for chronic conditions like multiple sclerosis and epilepsy, where several different medications help patients control the disease. The question then becomes, which medication is best for a particular person? To find out, clinicians typically run people to determine which medication controls symptoms of the disease for them with the fewest side effects.

“The discovery process can take weeks, months, or even years,” Chennuru said. “During this time, the patient may be at risk of experiencing flares and negative reactions to the medications. There’s also no way to know whether another therapy might work better unless the patient tries them all. Artificial intelligence can make this process more effective and focused by personalizing the treatment recommendations to the patient. That way, patients start treatment with the medication that is most likely to work rather than running down a list of meds.”

AI Adoption

AI adoption in healthcare has grown more slowly than other industries, partly because of all the complementary innovations that needed to happen first, as well as the need for data interoperability and representation. But Sarabu said the tide may be turning. Healthcare institutions across the country are starting to expand their clinical informatics capabilities to integrate and monitor the performance of AI tools.

Clinicians are already implementing diagnostic AI tools, “but we don’t know how they’re going to change healthcare delivery operations,” Sarabu said. The jury’s still out on how these new tools will be used safely in a way that protects privacy and mitigates adverse bias. Patients can consent to sharing their care data to help advance AI learning. Mitigating privacy, security, and adverse bias concerns are always critical.

“AI systems can assist with diagnosis and decisions about treatment plans, but the operative word here is assist,” Chennuru said. “AI can return insights and recommendations, but clinicians still review those insights and recommendations and draw their own conclusions. They consider context and display wisdom that applied AI systems do not.”

Experts project we’ll see the integration of more advanced, improved AI tools in healthcare delivery over the next several years as they continue to learn and test the technology. Medicine will eventually become so personalized that physicians will be able to pinpoint what caused the person’s ailment, predict the disease course, and determine how best to treat or even prevent it. “The technology is there,” Traill said. “Ten years from now, AI technology will be extremely precise, allowing physicians to screen for and manage disease very differently than we do today.” And where AI improves diagnosis and disease management, it can lead to better health outcomes for people.