Seminal report focuses on hope, hype, promise, and peril of AI use in medical arena

A Regenstrief Institute research scientist and vice president is a key contributor to a groundbreaking new publication exploring opportunities, issues and concerns related to artificial intelligence and its role to improve human health.

Eneida Mendonca, M.D., PhD, an expert in natural language processing, machine learning, predictive analytics and AI adoption, is a co-author of "Artificial Intelligence in Health Care: The Hope, The Hype, The Promise, and The Peril," a National Academy of Medicine (NAM) Special Publication. Other authors are from Harvard University, the Mayo Clinic, Johns Hopkins, Stanford University, Columbia University, Vanderbilt University and the Gates Foundation.

Dr. Mendonca is a co-lead in the seminal report that considers the potential tradeoffs and unintended consequences of AI (chapter 4) and co-author on the chapter that explores deploying AI in clinical settings (chapter 6). Each chapter includes recommendations.

Regenstrief Institute President and Chief Executive Officer Peter Embí, M.D., who participated in November in a high profile NAM conference on artificial intelligence and health care, said, "The potential for AI is enormous and we are optimistic about AI's promise to bring improvements in health, but we must also be cautious, thoughtful and act wisely. As Dr. Mendonca and co-authors write in the NAM report, 'though there is much upside in the potential for the use of AI in medicine, like all technologies, implementation does not come without certain risks.' "

To mitigate these risks, Dr. Mendonca and her NAM report co-authors call for transparency in the collection and use of data that drive AI solutions as well as in the design of the complex computational processes that make AI possible. For example, the report suggests consideration for establishing review bodies, to oversee AI in medicine.

The authors also underscore workforce issues, noting that advanced technologies will almost certainly change roles. "Instead of trying to replace medical workers, the coming era of AI automation can instead be directed toward enabling a broader reach of the workforce to do more good for more people, given a constrained set of scarce resources," they write. Turning to security issues, the report encourages development of AI systems resistant to misuse by bad actors. {module INSIDE STORY}

In a section titled "Net Gains, Unequal Pains" the authors caution that "if high tech healthcare is only available and used by those already plugged in socioeconomically, such advances may inadvertently reinforce health care disparity."

Turning to AI deployment, Dr. Mendonca and co-authors believe that AI will play a major role in the image interpretation processes of radiology, ophthalmology, dermatology and pathology as well as in the signal processing used in ECG, audiology and EEG tests. They also note AI's potential to "assist with prioritization of clinical resources and management of volume and intensity of patient contacts, as well as targeting services to patients most likely to benefit." They see great opportunity for AI in areas outside the point of care including population health as well as managing administrative tasks.

"We need to focus on clinical safety and carefully monitor uses and outcomes after implementation as we integrate AI within our electronic medical record systems," said Dr. Mendonca. "As we wrote in the National Academy report, 'Virtually none of the more than 320,000 health apps currently available and which have been downloaded nearly 4 billion times, has actually been shown to improve health.' "

"While being beneficial, AI has the potential to create unintended consequences so must be subject to regulation and be ethically implemented. A regulatory framework would be better established proactively, rather than in response to specific issues," said Dr. Mendonca. "Health systems must take steps to ensure the technology is enhancing care for all patients. System leaders must make efforts to avoid introducing social bias into the use of AI applications, which includes demanding transparency in the data collection and algorithm evaluation process. General IT governance structures must be adapted to manage AI and, if possible, the technology should be used in the context of a learning health system so its impact can be constantly evaluated and adjusted to maximize benefit."

"We are in the early developmental stages with many hurdles to overcome, but AI clearly holds immense potential in the healthcare arena," said Dr. Embí. "Leveraging AI to focus on clinical safety and effectiveness as well as stakeholder and user engagement are of paramount importance as are continual monitoring and evaluation. The bottom line is humans need to be intelligent about artificial intelligence."

NAM noted in a press statement, "AI has the potential to revolutionize health care. However, as we move into a future supported by technology together, we must ensure high data quality standards, that equity and inclusivity are always prioritized, that transparency is use-case-specific, that new technologies are supported by appropriate and adequate education and training, and that all technologies are appropriately regulated and supported by specific and tailored legislation."

The NAM special publication on AI is viewed as a reference document for all stakeholders involved in AI, health care or the intersection of the two. It prioritizes caution in implementation of this technology, prioritization of human connections between clinicians and patients, and an unwavering focus on equity and inclusion. NAM, established in 1970 as the Institute of Medicine, is an independent organization of eminent professionals from diverse fields including health and medicine; the natural, social, and behavioral sciences; and beyond. The new publication and associated resources can be downloaded at http://www.nam.edu/AIPub.

A new study uses machine learning on ultrasound images of thyroid nodules to predict the risk of malignancy

Thyroid nodules are small lumps that form within the thyroid gland and are quite common in the general population, with a prevalence as high as 67%. The great majority of thyroid nodules are not cancerous and cause no symptoms. However, there are currently limited guidelines on what to do with a nodule when the risk of cancer is uncertain. A new study from The Sidney Kimmel Cancer Center - Jefferson Health investigates whether a non-invasive method of ultrasound imaging, combined with a Google-platform machine-learning algorithm, could be used as a rapid and inexpensive first screen for thyroid cancer.

"Currently, ultrasounds can tell us if a nodule looks suspicious, and then the decision is made whether to do a needle biopsy or not," says Elizabeth Cottril, MD, an otolaryngologist at Thomas Jefferson University, and clinical leader of the study. "But fine-needle biopsies only act as a peephole, they don't tell us the whole picture. As a result, some biopsies return inconclusive results for whether or not the nodule may be malignant, or cancerous, in other words." {module In-article} 

If examining the cells of a needle biopsy proves inconclusive, the sample can be further tested via molecular diagnostics to determine the risk of malignancy. These tests look for the presence of certain mutations or molecular markers that are associated with malignant thyroid cancers. When nodules test positive for high-risk markers or mutations, the thyroid may be surgically removed. However, the standards for when to use molecular testing are still in development, and the test is not yet offered in all practice settings, especially at smaller community hospitals.

In order to improve the predictive power of the first-line diagnostic, the ultrasound, Jefferson researchers looked into machine learning or artificial intelligence models developed by Google. These applications are being used in other spaces: retail giants like Urban Outfitters use machine learning to help classify their many products, making it easier for the consumer to find an item they're interested in. Disney uses it to annotate their products based on specific characters or movies. In this case, the researchers applied a machine-learning algorithm to ultrasound images of patients' thyroid nodules to see if it could pick out distinguishing patterns. The study was published in JAMA-Oto on October 24th.

"The goal of our study was to see whether automated machine learning could use image-processing technology to predict the genetic risk of thyroid nodules, compared to molecular testing," says Kelly Daniels, a fourth-year medical student at Jefferson and first author of the study.

The researchers trained the algorithm on images from 121 patients who underwent ultrasound-guided fine needle-biopsy with subsequent molecular testing. From 134 total lesions, 43 nodules were classified as high risk and 91 were classified as low risk, based on a panel of genes used in the molecular testing. A preliminary set of images with known risk classifications was used to train the model or algorithm. From this bank of labeled images, the algorithm utilized machine-learning technology to pick out patterns associated with high and low-risk nodules, respectively. It used these patterns to form its own set of internal parameters that could be used to sort future sets of images; it essentially "trained" itself on this new task. Then the investigators tested the trained model on a different set of unlabeled images to see how closely it could classify high and low genetic risk nodules, compared to molecular test results.

"Machine learning is a low-cost and efficient tool that could help physicians arrive at a quicker decision as to how to approach an indeterminate nodule," says John Eisenbrey, PhD, associate professor of radiology and lead author of the study. "No one has used machine learning in the field of genetic risk stratification of thyroid nodule on ultrasound."

The researchers found that their algorithm performed with 97% specificity and 90% predictive positive value, meaning that 97% of patients who truly have benign nodules will have their ultrasound read as "benign" by the algorithm, and 90% of malignant or "positive" nodules are truly positive as classified by the algorithm . The high specificity is indicative of a low rate of false positives; this means that if the algorithm reads a nodule as "malignant" it is very likely to truly be malignant. The overall accuracy of the algorithm was 77.4%.

"This was such an important collaboration of surgeons and radiologists, and there's already interest from other institutions to pool our resources. The more data we feed the algorithm, the stronger and more predictive we'd expect it to become," says Dr. Cottril.

"There are so many potential applications of machine learning," says Dr. Eisenbrey. "In the future, we'd like to make use of feature extraction, which will help us identify anatomically relevant features of high-risk nodules."

Though preliminary, the study suggests that automated machine learning shows promise as an additional diagnostic tool that could improve the efficiency of thyroid cancer diagnoses. Once it becomes more robust, the approach could give doctors and patients more information in order to decide if thyroid lobe removal is necessary.