Powering Precision Medicine with Artificial Intelligence
Mar 06, 2018
Mar 06, 2018
Precision medicine is one of the most exciting and encouraging advances in healthcare today. It is moving us from one-size-fits-all healthcare to personalized, data-driven treatment that enables more efficient spending and improved patient outcomes. As defined by the National Institute of Health (NIH), precision medicine is an emerging approach for disease treatment and prevention that takes into account individual variability in genes, environment, and lifestyle. Doctors and researchers can use precision medicine to more accurately predict which treatment and prevention strategies will work best for a particular patient.
In other words, precision medicine offers a path to helping people recover from illness faster – and stay healthy longer. However, there are barriers to the widespread adoption of precision medicine. These include an ever-increasing amount of data, a shortage of specialists, and the exorbitantly high costs of drug development.
Consider these research-based findings:
As artificial intelligence (AI) enters the precision medicine picture, it can help organizations capitalize on precision medicine in multiple ways. In terms of data challenges, AI leverages deep learning approaches to overcome the obstacles inherent in large data sets and unstructured data. In clinical settings, AI functions as an assistant that helps clinicians work more efficiently and make more accurate diagnoses, which helps improve the productivity of healthcare workers. And at a broader level, AI helps companies accelerate drug development to cut costs and achieve faster time to medicine while reducing errors in the system.
To make this story more tangible, let’s consider a few examples of use cases for AI in precision medicine.
Problem: China has too few expert radiologists to serve a population of 1.3 billion people. To exacerbate the problem, the nation’s 80,000 radiologists spend most of their time looking at normal images, leading to delayed diagnosis for the abnormal cases.
Solution: An AI-based solution running on Intel® Xeon® Processors has been commercially deployed in multiple hospitals, including one of the top ten facilities in China, and has already detected abnormal thyroid nodules in over 5,000 patients. It is estimated that at full deployment this solution could shift 70% of the initial screening workload to less experienced clinicians. This allowed the experienced clinicians to focus on the more challenging cases, increasing their job satisfaction. The average accuracy was about 75% for clinicians while the latest AI program approached 85% accuracy.
Image source: Intel
Problem: 2D and 3D medical datasets are large. With conventional systems, clinicians often need to break the datasets down into patches and then work with small batch sizes—even though they achieve better accuracy when they use whole images.
Solution: With AI and the greater memory available on Intel® Xeon® Processor-based systems vs. GPUs, researchers can work with whole images and use deep learning to train models to detect brain tumors in hours rather than days. Learn more here.
Brain tumor segmentation using UNet
Image source: Multimodal Brain Tumor Segmentation Challenge
Problem: In the drug discovery process, researchers must look at millions of images for hundreds of thousands of compounds. Adding to the challenge, high-content cellular microscopy data is large in size and it is generally cost-prohibitive to manually label all images.
Solution: AI reduces the time required for HCS analysis by using multiscale convolutional neural networks. With AI, the network discovers features by itself, without the need for specialists’ time.
Problem: Genomic data is useful for predicting disease risk compared to clinical data alone. Moreover, it enables us to develop fine-grain categories of disease—to personalize the diagnosis.
Solution: Intel developed a deep learning model that was able to achieve 85% accuracy on disease risk prediction. It was able to correctly identify 23 subjects at risk for cardiovascular disease that were not detected by traditional statistical methods.
Genomic Data Used for Disease Risk Prediction
Image source: Intel; based on work in progress with The Scripps Research Institute.
Intel is working at multiple levels to help organizations use AI to advance the cause of precision medicine. Intel covers the full stack of AI solutions—including algorithms, frameworks, libraries, and processor technology.
Intel and partners will showcase many of these tools and technologies for AI in precision medicine this week at the Health Care Information and Management Systems Society (HIMSS) conference in Las Vegas. These include the Intel® Deep Learning Deployment Toolkit, which delivers optimized inferencing on Intel® architecture without the need for specialized hardware such as GPUs to keep deployment costs lower. This toolkit helps bring the power of AI to clinical diagnostic scanning and other healthcare workflows.
If you’re attending HIMSS 2018, please stop by the Intel booth for a firsthand look at some of the AI tools and technologies that are revolutionizing precision medicine. Or you can learn more at ai.intel.com/technology.
 https://www.emc.com/analyst-report/digital-universe-healthcare-vertical-report-ar.pdf; https://www.cdc.gov/mmwr/preview/mmwrhtml/su6103a6.htm
 https://www.rcr.ac.uk/system/files/publication/field_publication_files/cr_workforce_census_2016_report_0.pdf; https://pdfs.semanticscholar.org/029d/1c7b66191a212033b33194868d7128dde72e.pdf
Get the latest from Intel AI