Healthcare Gives AI a Wellness Check

Is AI ready for the tough demands of healthcare? Berlin Institute of Health, University Hospital, and Penn Medicine are early AI adopters, adding AI into their workforce to speed diagnosis, improve workflows, and deliver better patient outcomes. These organizations and others are testing AI limits and opening doors for innovations.

Image credit: GE Healthcare

Recently Dave Ryan, General Manager of Global Health and Life Science at Intel joined us on an IoT Integrator Wire podcast to discuss the impact of different technologies on healthcare and the response to Covid-19. Our conversation spanned multiple health and technology topics with pointed attention on AI. Specifically, has AI proven itself to be ready for healthcare?

For the past few years, healthcare has looked at AI as a next-generation technology, even after having proven itself in retail and financial services. Ryan counters that AI is ready, and Covid-19 is definitely fast tracking AI to help with vaccine research, point-of-care diagnosis, and patient workflow during surges. In fact, Ryan says that since the virus outbreak, medical researchers and healthcare providers are using or planning to expedite AI solutions at much higher rates than ever before. A recent report from ABI Research supports Ryan’s assertion, showing increased AI adoption in healthcare, with AI spending in healthcare and pharmaceutical industries expected to increase from $463 million in 2019 to more than $2 billion during the next five years.

On the flip side of this positive AI forecast, another group of scientists remain skeptical that AI is ready for life-and-death consequences of healthcare. Their argument is that the data sets are too small and the models’ results are not accurate enough to meet the higher demands of medicine and patient care. When asked about these concerns, Ryan notes that “the concern on the size of the data sets is valid.” He adds, “One of the interesting approaches we’re seeing is federated learning, which preserves data privacy.”

AI Takes on Cancer, Brain Tumors, and Collapsed Lungs

In the face of the pandemic and potential future surges, medical facilities are betting that AI will help more than hurt. The extreme need for added data intelligence at hospitals, research facilities, and drug manufacturers is exactly the open window that will attract a larger set of healthcare supporters to AI. Examples are coming in from around the world with compelling AI proof points.

At the research facilities of Berlin Institute of Health (BIH) and the Charité – Universitätsmedizin Berlin, scientists are developing a digital imaging analysis system that uses AI to evaluate tissue sections on thousands of microscopic images. In early efforts, the scientists have trained the software to reliably identify lung, breast, and colon cancer, immune cells in tumor tissue, and various tumor markers. Ultimately, the software will help trained medical professionals detect cancer, analyze infections, and check for degenerative connective tissue and autoimmune disease.

GE Health has a similar objective behind its GE Critical Care Suite, an imaging device with embedded AI for case prioritization and quality control. The x-ray imaging devices, which were the first of its kind to be approved by the FDA, helps prioritize critical cases, such as pneumothorax, or collapsed lung. University Hospitals Cleveland Medical Center is the first facility to adopt this technology, and it is now detecting about 7 to 15 collapsed lungs per day within the hospital. Healthcare workers rely on the solution to scan diagnostic images and detect large and small pneumothoraces more easily.

In another AI effort, Penn Medicine is bringing in AI to help detect brain tumors. What’s unique in this example is that researchers are addressing the smaller data sets and model accuracy by applying federated learning. This technique trains an AI algorithm across decentralized devices or sensors holding data samples, without exchanging those samples and protecting patient data.

Federated learning is appealing to healthcare because researchers from multiple organizations can access larger data samples while preserving patient privacy. At Penn Medicine, 29 research institutions will collaborate using the federated-learning approach to develop models for identifying brain tumors. The research will be trained on the largest brain tumor dataset to date. Early findings show that the models are more accurate than traditionally trained models.

Intel: Helping to Make AI Real

Essential to these AI solutions is massive compute power, and Intel silicon, software, and toolkits are delivering the processing power capable of speeding solution development. GE Healthcare is using Intel® processor-based power for its Critical Care Suite at University Health. The development team also leveraged the Intel® Distribution of OpenVINO™ toolkit to improve the models’ performance.

At Penn Medicine, researchers are using Intel technology to deploy the federated-learning approach for its dozens of research institutes to collaborate on deep learning models. In a separate healthcare project that is more data analytics and HPC focused, Berlin Institute of Health is also leaning on Intel for the design of a highly optimized HPC architecture to sequence single cell RNA for Covid-19 patients.