Harnessing artificial intelligence in medicine
September 25, 2019
Artificial intelligence in medicine
How UT Southwestern doctors and scientists are harnessing AI.
With artificial intelligence (AI), researchers are using the power of computers to make inroads in diagnosing and treating diseases.
“AI is going to transform health care. Nothing is comparable,” says Steve Jiang, Ph.D., Professor of Radiation Oncology and Director of UT Southwestern’s Medical Artificial Intelligence and Automation Laboratory (MAIA). “Almost everything we do in health care will be impacted by artificial intelligence – to improve the efficiency and quality of the work. AI helps humans do a better job, faster.”
UT Southwestern experts share a few ways they see AI bringing changes to health care.
1. To predict whether cancer will progress
Gaudenz Danuser, Ph.D., Professor and Chairman of the Lyda Hill Department of Bioinformatics at UT Southwestern, says, “It’s becoming very clear that artificial forms of intelligence can in many ways massively outperform human intelligence. But AI is going to augment – not replace – human decisions.”
Dr. Danuser’s laboratory is using AI to predict whether stage 3 melanoma is likely to advance to stage 4. For the research, a computer monitors cells under a microscope for four hours, and an AI system then extracts 50 numbers for each cell that best represent the cell behavior.
By collecting and mapping many sets of 50 numbers from many different cells, AI can predict which cancers will progress. Dr. Danuser’s lab is working to reverse engineer what these numbers signify.
“We have no understanding as humans what these 50 numbers mean,” he says. “The AI system is extracting information from data that we as human beings would never think of. That is, for me, the most exciting aspect.”
2. To prioritize workflow for radiologists
Artificial intelligence is also making solid inroads in medical imaging, since imaging analysis is an area where technology excels.
Thomas O’Neill, M.D., Assistant Professor of Radiology, explains how three factors combine to make AI valuable in medical imaging: machines and algorithms that are good at computer vision, a large number of scans, and metadata labels attached to the images for training AI algorithms.
AI algorithms can screen images immediately after the scan is finished and use that information to help optimize workflow for the radiologist.
As part of a research project that’s now used clinically, an algorithm can detect and flag a possible intercranial hemorrhage on a head CT. “The computer-aided detection tool notifies the radiologist and prioritizes those scans, so we look at those studies first,” Dr. O’Neill says.
Early data shows that turnaround time for these patients is improving. “You still need a radiologist to make the final interpretation, but it’s definitely a workflow assistance tool,” he says.
Similar algorithms are helping radiologists detect cervical spine fractures and pulmonary embolisms, and also predict which tumor nodules are likely to become cancer. “AI tools are good at these specialized tasks,” says Dr. O’Neill.
In the future, Dr. O’Neill expects to see AI used for automated tumor segmentation, identifying fractures on plain films, and spotting lung nodules on chest X-rays to help detect cancer early.
3. To integrate and mine knowledge platforms to make new discoveries
The Kidney Cancer Program at Simmons Cancer Center, led by Director James Brugarolas, M.D., Ph.D., is using information technology to automatically extract information from the electronic medical record (EMR) and integrate this data with genomics and drug responses in avatars. “A lot can be learned from 3,000 patient records, over 1,500 whole exomes, and more than 100 patient avatars,” says Payal Kapur, M.D., the Director of GU Pathology and of the Histology Core of the Kidney Cancer Specialized Program of Research Excellence.
Together with doctors Satwik Rajaram, Ph.D., Assistant Professor of Bioinformatics, and Ivan Pedrosa, M.D., Ph.D., Professor of Radiology, her team is also deploying AI to analyze kidney cancer samples and vertically integrate multiple platforms, including imaging, pathology, and molecular data.
“Almost everything we do in health care will be impacted by artificial intelligence – to improve the efficiency and quality of the work. AI helps humans do a better job, faster.”
4. To access information in electronic medical records
Information technology is making inroads in EMRs. Shaalan Beg, M.D., Associate Professor of Medical Oncology and Medical Director of the Clinical Research Office at the Simmons Cancer Center, explains how.
He says some valuable data in EMRs – like the details the doctor enters about your symptoms and the reason for your visit – can’t easily be compared among different doctors, patients, or health care systems.
Coding or structuring this data consistently would make the data more searchable. So, the American Society of Clinical Oncology is launching an initiative to define which variables need to be entered in structured fields.
“That way, you can write whatever you need to, but a few of those elements need to be entered in a certain way and consistently across different centers,” Dr. Beg says.
That might sound like a simple solution, but it has powerful ramifications. Once the data from medical records is entered consistently, researchers can create algorithms to search for similarities between symptoms, diagnoses, or drug interactions.
“We can use that information to make clinical observations and decisions,” Dr. Beg says.
5. To build smarter hospitals
Dr. Jiang and his team of MAIA medical physicists are busy developing intelligent medical devices and computer algorithms for Simmons Cancer Center clinicians to improve treatment and patient safety. Among its many projects, the group has been working to incorporate artificial intelligence into the fabric of a medical clinic.
The team’s Real Time Location System (RTLS) uses sensors based on Bluetooth technology to track the location of patients, clinical staff, and equipment in order to improve workflow and patient safety.
“We’ve been working on this for a couple of years,” Dr. Jiang says. “The first step is to give a patient a wristband with a sensor in it to wear during their clinic visit. We can then track the cancer patient – just like a GPS. We can also verify the patient’s identity using the sensor.”
If a patient has been waiting in an exam room for, say, 15 minutes, the system can send a text or other reminder to staff. If the patient is still waiting 15 minutes or a half-hour later, the system can send a stronger reminder.
“We think future hospitals should be really smart – with sensors in the building, on the walls, in the ceiling, on the patient, and on the equipment collecting data,” Dr. Jiang says. “AI can then analyze that data to create the best workflow and patient monitoring.”
Better planning through automation
Treatment planning for cancer radiotherapy, where an optimal treatment strategy is designed for each individual patient and executed for the whole treatment course, is similar to the design of a blueprint for building construction. If a treatment plan is poorly designed, the desired treatment outcome cannot be achieved, no matter how well other components of radiation therapy are performed.
In the current clinical workflow, a treatment planner works toward a plan, and multiple rounds of consultation between the planner and physician are often needed. Consequently, planning time can be up to a week for complex cases, and plan quality can vary significantly.
Researchers at the Medical Artificial Intelligence and Automation Laboratory (MAIA) are working to combat this by revolutionizing treatment planning with the use of AI technologies. Treatment planning consists of two major aspects: commonality and individuality. By exploiting the commonality through deep supervised learning, physicians can develop a treatment plan as good as those for previously treated similar patients, and individuality can be actualized by learning physicians’ special considerations for a particular patient using deep reinforcement learning. Combined, these elements can consistently and efficiently produce high-
quality treatment plans.
“AI is changing the world and also changing health care,” says Steve Jiang, Ph.D., Professor of Radiation Oncology and Director of the MAIA Lab. “That’s what we’re working on. We’re trying to use AI to solve important clinical problems.”