As data science and medicine come together, artificial intelligence (AI) is opening new possibilities in nearly every medical specialty. At Brigham and Women’s Hospital, a large academic medical center within an even larger medical ecosystem, initiatives underway reveal how close we are to adding AI to the physician’s toolkit to improve patient care.
A national leader in AI, Brigham clinicians and scientists are poised to bring the power of AI into the clinic. The breadth of research throughout this Harvard teaching hospital, plus collaborations with Harvard Medical School and other teaching hospitals nearby, create the ideal opportunity to aggregate the data needed to advance AI
And, unlike a technology company, clinical problem-solving is in the Brigham’s DNA.
“Here, we can take artificial intelligence and use all the talents, people and expertise in our environment to the best advantage,” said Giles W. Boland, MD, chair of Brigham’s Department of Radiology. “With so many physicians who are specialists in so many diseases and areas, we have the knowledge to ask: What are the right medical questions to go after? How do we design and build algorithms to answer the questions that are most important for patients?”
Institutional foresight and early investment in AI infrastructure at the Brigham included ways to bring technical experts and clinicians together. “By embedding computer scientists, computer engineers and data scientists in our clinical departments, we’re enabling the means to address big problems in medicine,” said Jeffrey Golden, MD, chair of the Department of Pathology. Even earlier, the Brigham invested in a Center for Clinical Data Science.
These factors lead to a large accumulation of well-documented patients and patient data to aggregate into learning algorithms – material that makes up what Boland calls “the new gold.”
Moving AI Toward the Clinic
Today, work in nearly every specialty at Brigham and Women’s Hospital is bringing AI closer to clinical use, as projects move forward through investigational stages. Some ongoing work includes:
- Harnessing machine learning to tackle difficile infection: Georg Gerber, MD, PhD, MPH, FASCP, chief of computational pathology in the Department of Pathology, designs novel Bayesian machine learning methods to understand how microbiota in the gut may enable or thwart C. difficile, the most common hospital-acquired infection in the United States. With Lynn Bry, MD, PhD, also in the Department of Pathology and the Director of the Massachusetts Host-Microbiome Center and Jessica Allegretti, MD, MPH, director of clinical trials in the hospital’s Crohn’s and Colitis Center, their research team has identified which microbes in the gut prevent and treat C.difficile in mice — work that they plan to move soon into human trials.
- Detecting early glioblastoma recurrence through a smart phone: Neurosurgeon Timothy Smith, MD, PhD, MPH is investigating whether smartphone data can reveal neurocognitive decline and other indications of glioblastoma recurrence. Voice recordings, GPS tracking, the time it takes to return a text may indicate difficulty in navigating daily life. Smith and colleagues are building and testing combinations of data streams to build a model of cognition that could inform a physician of status changes sooner than intermittent imaging and clinical exams, as well as be more accurate than patient self-assessment.
- Predicting Chronic Obstructive Pulmonary Disease (COPD) staging and outcomes using chest CT: Pulmonologist George R. Washko, Jr., MD, and Raul San Jose Estepar, PhD, directors of the Applied Chest Imaging Laboratory at the Brigham, used a montage of chest CT images to train a convolution neural network to detect COPD, stage its severity and predict clinical outcomes such as respiratory events and mortality. This technique reflects the ability of deep learning methods to diagnose and prognosticate acute events in COPD, a major driver for healthcare utilization, just from routine imaging examinations. Recent advances in image processing are now allowing the team to focus on integrating image analytics into clinical care and healthcare analytics.
Still Early Days for AI
Despite the progress, Boland said, “It’s still very early days for AI. The reality is that to build an AI algorithm that is accurate and safe for patients is hugely complex.”
Early wins for AI in medicine came in visual specialties — pathology, radiology, ophthalmology, dermatology. Now, as AI research moves into multisystem complex diseases, data from multiple sources (i.e., radiology, the electronic medical record, pathology) must be able to interact.
“Building relational databases that allow you to interrogate the data is something that we have invested in. That has permitted us to do things that cannot be easily done in most places. For example, we have created databases with digital pathology images, sequencing data from those tumors and all linked to the clinical data,” said Golden. “But it continues to be one of the biggest challenges in AI.”
Other hurdles to bringing AI to the clinic include regulatory uncertainties and the potential for racial and other biases rooted in the research populations used to train machine learning models. The greatest challenge may be allocating resources wisely to focus on the clinical scenarios of greatest impact.
But with enhanced computer power that has become available in the past few years, AI applications for patient care are coming closer to the clinic.
As for the perennial question of whether physicians could be supplanted by AI, Brigham’s leaders express consensus:
“The future is obviously unlimited, and massive attention and focus and research is being pulled into this,” said Boland. “Because the goal is to add value to overall patient care, it is inevitable that artificial intelligence programs will be built to aid physicians, but not to replace them, to make the right decisions at the right time.
Regarding her own project for automated infusion during C-sections, Dr. Kovacheva emphasizes that AI can be a tool for strengthening physician-patient interactions.
“I don’t think artificial intelligence will create an autonomous doctor anytime in the near future,” said Dr. Kovacheva. “But we will be able to automate those mundane tasks that take up some of the physician’s mental energy and allow us to focus on what’s really important — our patient.”