The AI landscape is changing fast, and industries are racing to be first in line to new insights, reduced costs, and creating better lives. While embracing any new technology industry-wide is a daunting task, the medical field is one area where we are seeing potentially life-saving advancements already.
We’re beginning to see changes in subtle but meaningful ways. Most healthcare providers are already using background speech recognition software in some form.1 In some places, they’ve lost the need for keyboards altogether, which allows for a more human-oriented relationship with patients.2 Similarly, Google’s London-based company DeepMind is saving nurses up to two hours a day—extra time that can now be spent with patients.3 But these are small examples of what a technology like Deep Learning can do to advance the medical field.
What is Deep Learning, exactly? Broadly speaking, it is a very sophisticated pattern recognition method that gets better the more data you give it. In very simple terms, it includes training, which involves feeding existing data into a framework4 (e.g., a data automation tool like TensorFlow); and inference, which is where the newly trained model creates new data (i.e., learns). Siri, Cortana, and Alexa are all examples of this technology already being applied in your everyday life.
However, one major limitation in our ability to apply this method is collecting the information that can be trained. While the medical industry is slowly but surely digitizing its data, there is no single unified database to store that information that is as broad and deep as, say, Google’s open datasets. But the more we do things like upgrade our infrastructures and incorporate wearable devices that collect high-frequency biometric data (think fitbits, but more advanced),5 the more useful information there will be. The more data we have, the more it can be collected, analyzed, learned from, and the more accurate predictions we will be able to make that change lives for the better.
Earlier this month, Intel (in collaboration with Siemens Healthineers) announced a new AI-based model for real-time cardiovascular disease diagnosis. This will be made possible by Intel’s® new 2nd generation Xeon® Scalable processors. Combined with Intel’s® Deep Learning Boost and the new Vector Neural Network Instructions (VNNI), which are designed for targeted workloads like image processing, the new technology provides up to 30x faster inference performance. This is a great example of a real-world benefit of AI happening in the very near future. And it’s just the beginning.
Looking forward, consider what two generations ahead could have at their disposal. They’d have an ever-growing trove of personal biometric and genomic data that’s being compiled and analyzed continuously, plus the lifelong data of their parents, plus the vast ocean of anonymized data of the general population to compare it to. We could use neural networks to make the wealth of medical literature interpretable. We could change clinical workflows to be more patient-focused, drastically improve accuracy and diagnosis, revolutionize preventative treatments, and find localized patterns of disease (e.g., cancer clusters) with unprecedented speed and precision. We’re already seeing personalized cancer treatments6, but beyond that we could provide a lifetime of truly individualized medicine. And that future is getting closer, in large part, thanks to Intel.
As for what can be done to improve your workflow today, BOXX offers a number of solutions. Our new line of data science workstations—powered by Intel’s® new 2nd generation Xeon® Scalable processors and NVIDIA’s® Quadro® RTX™ GPUs—are perfectly suited for the unique demands of deep learning development. And for the largest workloads, the NVIDIA® DGX™ systems, built on NVIDIA’s revolutionary Volta™ GPU platform, are unmatched, purpose-built supercomputers designed to accelerate the most demanding AI projects.
4 Frameworks aren’t strictly necessary, but they make things much, much easier.
5 Some smartwatches are beginning to collect more sophisticated biometric data.