The University of California, San Francisco is developing and training an artificial intelligence model that could help clinicians diagnose tears in knee cartilage.
As many athletes and active people have learned, a tear in the knee cartilage, or the meniscus, can lead to long-term health and lifestyle consequences, from debilitating osteoarthritis to limits on physical activity. One of the keys to mitigating these consequences is to identify and treat tears in the meniscus early on, before the condition brings larger health issues.
While this goal is pretty simple, the path forward is rather complicated. To diagnose a torn meniscus, clinicians need to review and interpret hundreds of high-resolution 3D magnetic resonance imaging (MRI) slices showing a patient’s knee from different angles. Radiologists then assign a numerical score to indicate the presence of a tear and its severity. This labor-intensive, time-consuming process relies heavily on the skills and availability of clinical specialists, and the interpretation of the images themselves can be rather subjective.
At the University of California, San Francisco (UCSF) and its Center for Digital Health Innovation (CDHI), researchers are working to address these challenges by adding artificial intelligence to the diagnostic equation. In this initiative, explored in a recent Intel case study, the research team is working to develop and train a deep learning model that can examine MRI results, identify those that show signs of torn knee cartilage and, eventually, objectively classify meniscus tears. The ultimate goal is to develop an accurate, data-driven grading system of meniscus lesions, and one that can provide results to patients immediately after scanning.
The solution
To support this ambitious AI initiative, the UCSF CDHI research team used an open-source distributed deep-learning library, BigDL on Apache Spark, to develop algorithms and train models on a data analytics cluster built with leading-edge technologies. That cluster is based on Dell EMC™ PowerEdge™ servers, Intel® Xeon® Scalable processors and the Cloudera Distribution of Apache Hadoop for storing, processing and analyzing data. This approach allowed UCSF to train 3D models where the data resides, taking advantage of the larger-than-accelerator memory footprint. Other technologies in the solution include the Intel® Math Kernel Library (Intel® MKL) to accelerate math processing routines, the TensorFlow open source framework for deep learning and machine learning, and the TensorBoard suite of open source visualization tools.
The 3D convolutional neural network at the heart of this image-classification solution is using existing MRI images to train a model to recognize meniscus tears. The initial goal of this incremental process is to develop a model that can determine whether a patient’s cartilage is normal or torn, and to make this determination with a level of accuracy that meets or exceeds that of trained radiologists. This advance alone could help drive patient care forward by enabling radiologists to quickly identify the patients they need to focus on.
The bigger picture
This work that is under way at UCSF provides a glimpse into the future of healthcare. In this emerging era, AI and other data-driven technologies will help transform patient care and make the healthcare system more efficient. These new technologies will also help address a critical shortage of physicians. A 2018 study conducted for the Association of American Medical Colleges (AAMC) predicts that the United States will face a shortage of 42,600 to 121,300 physicians by 2030, and that these shortages will be particularly large in specialty-care fields.[1]
Technologies like AI will also help us contain the rising costs of caring for an aging and growing population. A study by a team of researchers from the consulting firm Accenture found that the use of 10 promising AI applications could create up to $150 billion in annual savings for U.S. healthcare by 2026.[2] Notably, this study ranked the use case of automated image diagnosis — like that being developed at UCSF — as one of 10 AI applications that could change healthcare.
To learn more
For a closer look at UCSF’s efforts to advance the use of AI in clinical medicine, read the Intel case study “Using Artificial Intelligences Solutions to Improve Patient Care.” And for a deeper dive into the broader topics explored here, read the Dell EMC ebook “Making digital transformation in healthcare a reality.”
[1] Association of American Medical Colleges (AAMC), “GME Funding and Its Role in Addressing the Physician Shortage,” May 29, 2018.
[2] Brian Kalis, Matt Collier, Richard Fu, “10 Promising AI Applications in Health Care,” Harvard Business Review, May 10, 2018.