The benefits of machine learning have been abundantly clear for decades, but perhaps nowhere more so than in the field of medicine. With physics-based simulations, 3D modeling, and machine learning methods, medical professionals and researchers have made significant strides in numerous specialties. While there has been much publicity and hype surrounding the use of machine learning, artificial intelligence, and virtual reality in a healthcare setting, the benefits of such technologies in many cases make the hype well-earned.
Since the advent of computer technology, the medical community has been devising ways to improve best practices for surgery, imaging, and overall patient care through its use. Machine learning dates back to as early as the 1940s, when a team that included neuroscientist Warren McCulloch conceived a mathematical model of neural networks. Since that breakthrough, ever-evolving technology has allowed the medical community to take machine learning farther than those early innovators likely ever dreamed possible.
Analytics in personalized medicine
Machine learning, at its core, is the use of computer systems that are able to learn and adapt to situations presented to them. Through the use of algorithms and statistical models, machines can analyze patterns within data. The application of machine learning in a medical setting is varied. Scientists and medical personnel have even used machine learning to model disease trajectory and treatment.
When dealing with the level of data found within medical processes, using machine learning to process such data can be invaluable. Training machine learning algorithms with multi-modal datasets improves the accuracy of predictions that clinicians and researchers can obtain for real patients. This information derived from machine learning techniques can inform both additional medical advancements and education.
By leveraging the plethora of available patient data, we are seeing increased use of machine learning and other computational techniques to stratify treatment options for diseases such as cancer. These methods are providing a virtual testbed to assess treatment plans or interventions in a personalized way. The long-term goal is to provide safer treatments and improved outcomes for patients. With data scientists working closely with medical professionals, we are already seeing the potential impact of computational tools.
Patient recruiting for research
Clinical trials are an invaluable step for research and advancements in medicine, but recruiting patients for these trials can be difficult. Machine learning techniques have been developed that assess and screen patients’ medical record data to highlight patients who may or may not be good candidates for trials. Efficiently identifying candidates can streamline the process of enrolling patients.
In addition, machine learning can be used to process the large datasets produced by clinical trials, thus reducing some of the assessment burdens by identifying complex patterns that may be difficult to identify manually, and helping improve researchers’ ability to predict outcomes. Over the years, machine learning has dramatically evolved to include processing abilities for increasingly more extensive datasets.
Many clinicians remain hopeful that, as technology progresses, machine learning will be able to help with ever-increasing clinical trial sizes.
Machine learning for diagnosis
One way machine learning is applied in the medical field is for diagnostics. Medical professionals are trained to review the vital signs, imaging records, and medical history of patients in order to give a thorough diagnosis and treatment plan. Machine learning is offering a promising paradigm shift in how clinicians diagnose disease.
Due to the dramatic increase in available clinical data from imaging to electronic health records to continuous monitoring information from wearable devices, machine learning provides a way to augment conventional diagnostic approaches through quick analysis of large amounts of multi-model data. For patients facing life-threatening or limiting conditions, these advances provide more personalized and efficient analysis.
Cardiac disease is one area in which machine learning is already aiding diagnostics. There has been rapid growth in the use of artificial intelligence technology for identifying atrial fibrillation or cardiac arrhythmia. Recent work has demonstrated the use of machine learning in predicting heart failure in asymptomatic patients. In our work, we focus on the role that understanding a patient’s blood flow can have in classifying the severity of coronary lesions.
In recent years, the use of in silico models has proven useful in reducing the number of invasive measurements to assess a coronary lesion. However, these simulations typically require large computing resources to model even one heartbeat. Machine learning not only provides methods to reduce the time-per-simulation, but we have also shown that machine learning techniques can be used to reduce the number of patient-specific blood flow simulations needed to assess potential physiological states.
By combining physics-based models with machine learning, only a limited number of states (e.g. rest or mild exercise) need to be modeled in order to accurately predict flow metrics under other conditions (e.g. running at altitude). These efforts allow clinicians to assess how a patient will function under a wider number of states than what can be directly measured in the clinic.
Traditionally, gathering this blood flow data would be expensive, time-consuming, and potentially invasive for the patient. With machine learning, the data-gathering process can be significantly streamlined, leading to both faster diagnosis and more informed treatment plans.
Future for smarter diagnosis
Many of those involved in research and development are hoping to bridge the gap between machine learning theory and application. Machine learning has come a long way since its earliest application in the early 20th century, and its potential in personalized healthcare and early detection of disease has been well demonstrated.
During the pandemic, machine learning was utilized to determine likely outcomes of infections, disease spread, and projections for variants. The work of machine learning in diagnostics, prognosis prediction, and clinical trials signal a bright future ahead for its further application in the medical field.
There are still challenges ahead in adopting machine learning for personalized medicine. Some examples include questions of fairness and potential bias when the training data is incomplete or demographically skewed in its collection as well as the need for transparency. The ‘black box’ nature of some machine learning models is concerning when used for clinical decision support.
In many ways, researchers have only just scratched the surface of what is possible with machine learning in the field of medicine. The amount of data being collected in medical communities is staggering, and medical experts from different specialties bring different types of data to the table – as well as different perspectives – on how to best analyze such data. Machine learning not only improves outcomes, but encourages collaboration in the medical world. Such collaboration can lead to more innovation and positive applications of technology.
Machine learning is improving the speed of data processing, thereby improving the speed of diagnosis and care given to patients. The impact of machine learning is already abundantly clear in research and in the patient-provider relationship.
Dr. Amanda E. Randles, Ph.D., is the Alfred Winborne Mordecai and Victoria Stover Mordecai Assistant Professor of Biomedical Sciences at Duke University and a fellow for the National Academy of Inventors (NAI), a member-based organization of patent-holding inventors at academic institutions and nonprofit research organizations across the U.S. and the globe.
