After his uncle died in 2018 from terminal lung cancer, Pelu Tran determined that artificial intelligence could make a difference, potentially helping doctors detect cancer earlier when scan images and avoid delaying important treatment.
Tran, an entrepreneur who studies both medicine and engineering at Stanford University, discovered that there are eight different companies whose AI applications have been approved by the Food and Drug Administration. your uncle’s lung cancer if they were used.
His desire to prevent missed diagnoses in the future and protect patients from medical error informs the mission of the technology company he co-founded, Ferrum Health, to help systems health systems implementing AI.
“Most diagnostic decisions today are made without the help of any kind of artificial intelligence, and that’s something we realized needs to change,” Tran said.
AI is used more often in radiology than any other specialty, with uses ranging from appointment scheduling and grading exams to diagnostics using imaging technology. In some cases, AI can be used to improve image acquisition, allowing for faster MRI scans and clearer, higher quality images, saving money and making it easier for patients. core. Machine learning technology is often touted as a way to save time and improve diagnostic qualityalthough some argue that AI is not worth the investment amid concerns about the potential for inaccuracies and bias.
“The promise of AI and many of the things we have at our disposal is the ability to move from a primarily retrospective approach to safety to a prospective approach to safety where we can anticipate failure and harm in advance. when it happens, then take action and reduce those possibilities in the future,” said Dr. Kedar Mate, president and CEO of the Institute for Health Improvement.
There are 392 AI and machine learning devices used in radiology FDA approved as of 2022, followed by 57 cardiovascular devices and 15 hematology devices.
According to a 2021 poll conducted by the Data Science Institute of the American College of Radiologists, nearly a third of radiologists are using AI in imaging, and 20% of practices say they plan to invest in it. invest in AI tools over the next 5 years.
Interest in AI comes as the field of radiology faces labor problems. According to the Association of American Medical Colleges, there is expected to be a shortage of up to 42,000 specialists, including radiologists, by 2033. That opens the door for more AI usage. .
Breast cancer imaging is one area in radiology where innovations in AI will likely be deployed first, largely due to the Radiography Quality Standards Act. Passed in 1992, this act was created to ensure that women have access to high-quality, uniform mammography facilities for the early detection of breast cancer that are accredited, certified by FDA and regularly tested.
There is now a team working on AI’s ability to detect breast cancer earlier and perform accurate diagnostic interpretations using use of imaging results while screening an increasing number of CT patients.
While the effectiveness and accuracy of machine learning technologies vary by use case and manufacturer, numerous studies show that algorithms can perform similarly to radiologists.
According to a NYU Langone Health study of 288,767 ultrasound exams performed from 143,203 women treated at NYU Langone hospitals, an AI tool improved doctors’ ability to accurately identify breast cancer. radiologists by 37% and reduced by 27% the number of biopsies needed to confirm a suspected tumor. between 2012 and 2018.
A 2022 report by the Norwegian Cancer Registry found that an AI system was able to accurately predict high breast cancer risk for 87.6% of 752 breast cancer cases detected in a single year. screening process. The study gathered from nearly 123,000 examinations performed on more than 47,000 women at four BreastScreen Norway participating facilities.
Newer AI systems used for a variety of purposes have the potential to enhance older computed tomography systems by processing CT images faster with better resolution and reduced amount of time needed for CT scans, reducing the patient’s radiation exposure.
However, the technology has its downsides and is a way to work independently of doctors, Morris said.
“Some algorithms have a harder time with asymmetry,” she said. “They’re all pretty good at detecting calcifications and tumors, but with some of the more subtle signs of breast cancer, they may not have been trained enough on cases to be able to recognize it.”
Since Ferrum Health was founded in 2017, Tran says the company has helped about 40 AI providers integrate into healthcare organizations, working with diagnostic tools and imaging in radiology. , women’s health, cardiology, orthopedics and oncology.
One major application of AI algorithms, Tran said, is to collect data on medical error rates. Using AI results, Ferrum Health can compare patient safety across the organizations it works with, showing where health systems are at fault. For example, one AI study found that one hospital missed a fracture in one out of every 200 upper extremity X-rays, the highest miss rate of all imaging tests. its photo.
Mate says for AI to succeed in the diagnostic space, it needs access to a large library of digital content and patient data so that its algorithms can be trained quickly. That can be difficult to achieve given industry-wide concerns about patient data privacy and bias, which happen when datasets are used to structure algorithms that exclude a single multiplier. demographics and incorporate human prejudice. Trends in AI mean that results cannot be broadly generalized, and the technology can be inaccurate when trying to diagnose certain groups of patients.
There are also concerns about technological limitations.
Because the FDA has not cleared AI models based on continuous learning, the healthcare sectors cannot benefit from AI’s key advantage, which is the ability to update algorithms based on new information in the future. real-time to improve patient care, Dr. Matthew Lungren, practicing interventional radiologist at the University of California, San Francisco. Continuous machine learning automatically retrains AI models with new data on a regular basis, so it stays up to date in a rapidly changing environment.
AI systems are also often extremely difficult to access historical data sets of the healthcare industry, which cannot interact with each other. That shows the need for the industry’s digital transformation before any significant progress in machine learning can be achieved, says Lungren.
“Most people immediately think that AI is very useful in making diagnoses,” said Dr Vadim Spektor, assistant professor of radiology at Columbia University’s Vagelos College of Physicians and Surgeons. “And it was probably one of the earliest uses of AI, but it’s one of the least effective uses of AI to date.”
Because AI algorithms focus on only one goal, they tend to target only one specific aspect of a problem and are primarily based on physician-recorded lists of patient records and patient records. Medical problems are prone to human error. If the information the AI device uses to conduct the assessment is incorrect, Spektor said, the results could be the same.
Another downside of AI is cost. With the exception of AI tools that take breast cancer images, which are paid for by insurance companies, Spektor says, there is essentially no reimbursement for machine learning devices, meaning all costs are borne by the system. self-pay healthcare system.
Most AI tools are subscription-based, he said, and require hospitals to pay the provider per use for every study processed by the algorithm. The most costly aspects of AI are deploying and maintaining the technology, troubleshooting problems, continuously evaluating quality, adjusting algorithms, and replacing outdated equipment, all of which take time.
Depending on scale, AI could cost health systems between $300,000 and $400,000, according to estimates.
The understaffed IT team and limited resources, as well as administrative committees, are having a hard time figuring out how to move expensive AI solutions through the system, Tran said. for wider application.
In 2021, the global AI radiology market size reaches $55.7 million and is expected to grow to $517.8 million by 2030, according to Vision Research Reports, a market research firm provides analysis and forecasts on various industries.
“In the near future, the best use of AI is where it can obviously do things faster without significant investment and cost,” Spektor said. “We need to think about how to redesign the whole process so that we can take advantage of what AI can do and what humans can do.”