AI/Watson

Advancing medical imaging research with deep learning

Share this post:

With the rise of connected healthcare, medical data is proliferating at an exponential rate. Sources such as electronic health record (EHR) systems, clinical trials, genetic sequencing, research studies and radiologic imaging all generate useful data, but the sheer volume has the potential to overwhelm physicians hoping to better serve their patients.

Recent gains in computing power enable us to process more data than ever before. Still, we need help making sense of it. That’s why medical researchers are using AI technologies such as deep learning to glean beneficial insights.

Our research at Washington University St. Louis and Vanderbilt University starts at this confluence of processing power, deep learning and the extreme volume of medical data.

We are applying deep learning to a predictive solution for sharpening the detail of images from magnetic resonance imaging (MRI) scans of the brain. The goal is to help physicians interpret the images to better diagnose tumors and assess responses to treatment.

Putting deep learning to work

If you’ve ever had an MRI, you know that scans can be noisy and slow, and you’re confined to a narrow, tube-like space that can be quite uncomfortable. Clinical engineers have developed techniques to reduce scanning time by producing undersampled, or incomplete, images. The problem is, such images can contain distortions that hinder accurate diagnoses.

Deep learning helps us reconstruct the raw images. Once we create the data model and populate it with thousands of known images, the system can learn what a normal brain looks like. Then, when we apply deep learning to the undersampled images, the solution fills in the gaps. Although our research concerns MRI scans of the brain, it is relevant to other specialties as well.

Such solutions will only be useful if they are user friendly to physicians who are expert in medicine but not computer science. After all, it is physicians with clinical expertise, rather than computer engineers, who best understand the connection between MRI images and the patient’s condition. We envision physicians accessing our solution from the cloud on their desktop PCs.

Powering a predictive solution

Advancing this vision requires a computing platform that is both powerful and easy to use. That’s where IBM plays a critical role. We use an IBM Power8 Server, the first generation of IBM servers designed to efficiently process big data in the cloud. This system forms the computational backbone of our research.

We also use IBM Spectrum Conductor Deep Learning Impact software, which made it easy for us to develop our first deep learning solution by breaking the design into discrete steps. The software also helps automate and accelerate system resource management, distributed processing and prototyping using our clinical data.

The IBM platform enables us to quickly test innovative ideas. And IBM’s timely technical support is essential to our work.

The solution we are developing can shorten physicians’ learning curve to utilize the most sophisticated analytical tools available. Patients everywhere will benefit from their ability to enhance image identification and improve diagnoses and treatments.

  

Watch Xiaoyu Jiang and Yong Wang discuss deep learning medical innovation:

Research Fellow, Vanderbilt University Institute of Imaging Science

Yong Wang

Assistant Professor of Gynecology and Obstetrics, Radiology and Biomedical Engineering at Washington University St. Louis

More AI/Watson stories

How AI helps Japan Airlines personalize the travel experience

For airlines, the sheer volume of flights and travelers can sometimes make it difficult to provide a personalized customer experience. When airports are busy and flights are full, passengers sometimes feel that the airline simply sees them as objects to be transported from point A to point B. In response, Japan Airlines decided to set […]

Continue reading

The problem of rural internet access: How AI is helping

A disconnected rural population Nineteen million people in the United States lack high-speed internet access. Many of these people live in rural areas, where fiber, coaxial cable, DSL or any other kind of broadband infrastructure is scarce or nonexistent. This is a huge issue for rural community administrators and school officials who have been struggling […]

Continue reading

How AI technology can help boost fashion forecasting

India is a country of fashion-loving millennials who love to wear the latest colors and styles. To capitalize on this strong market, however, fast fashion retailers such as my company, Bestseller India, have several challenges to overcome. The first is common to all retailers: how to offer consumers the right merchandise in the right store […]

Continue reading