Artificial intelligence has attracted significant attention recently, and yet many of the most popular examples we’ve seen demonstrating its potential benefits have been esoteric proof-of-concepts, such as mastering chess or finding cat videos on the internet.
While these developments have helped pave the way for further breakthroughs, they’ve also left many people asking where the tangible benefits are and what the era of machine intelligence really means to the real world.
At last, we’re reaching the tipping point where machine intelligence efforts are beginning to move past these preliminary examples into life-changing breakthroughs that can solve heretofore unsolvable problems. Nowhere is this more evident than in healthcare.
Healthcare is one of the most data-rich industries in the world. Record-keeping is an integral practice, and one that’s been made infinitely more accessible as health systems around the world have moved to electronic records. Diagnostic images, X-rays, CT scans, and MRI results are being stored digitally. While all these efforts have been made to reduce cost and increase the ease and effectiveness of patient care, in the era of machine intelligence they now create deep data lakes for analysis. This enables new research and pattern-finding that vastly exceeds the capabilities of human beings.
Healthcare data alone isn’t driving new breakthroughs. In fact, much of this data has been digital for years. However, the algorithms that were used to analyze the data couldn’t be run fast enough to provide valuable information in a timely manner. Now that’s changing owing to an unlikely application of graphics technology.
Graphics Processing Units (GPU) have been traditionally used to render graphics and video. GPUs are used to power everything from TV screens to immersive gaming experiences. However, the healthcare industry is now harnessing the power of these GPUs in machine intelligence applications.
Recent advances in GPU technology have made parallel processing fast, inexpensive, and powerful. Coupled with the expanding open-source software platforms, compute performance can finally keep pace with the needs of highly demanding machine intelligence algorithms. The ability to decipher the mystifying amount of data will have a profound impact on our health and healthcare systems, including the prediction and treatment of diseases.
Enhancing preventative medicine
Machine intelligence platforms are just beginning to prove their value for enhancing preventative medicine and stopping disease before it starts, a vital component of any healthcare strategy. Seven out of 10 deaths among Americans each year are due to chronic diseases (such as cancer and heart disease), and almost one out of every two adults suffer from at least one chronic illness, many of which are preventable.
Researchers have recently created an artificially intelligent diagnosis algorithm by programming a GPU to act as a neural network. By applying “deep learning” using the GPU, the team trained the neural network to identify and differentiate between malignant and benign skin lesions. The study’s result showed the algorithm to be as reliable as a human dermatologist is at detecting skin cancer, albeit with the potential to provide diagnoses at much greater speed and at lower cost. With 5.4 million new cases of skin cancer in the US every year, early detection can have an enormous impact on outcomes.
Machine intelligence is also already being used to predict the future health of individuals and populations by analyzing clinical and non-clinical data to identify high-risk patients before disaster strikes. Greater adoption of these technologies can also lead to better patient outcomes and reduce waste in the healthcare system by addressing issues in over-treatment and care delivery.
As these services become more widely available in the future, it’s projected that deep learning will provide greater accuracy, faster analysis and ultimately lower healthcare costs.
Supporting research and discovery
Machine intelligence is also forging a new path where traditional approaches to research haven’t been successful.
Historically, the cost of developing new treatments for rare diseases has been gargantuan and often prohibitive. The number of patients living with a single rare disease is small which makes it difficult to find participants for expensive clinical trials and for pharmaceutical companies to recoup costs once a drug goes to market. Machine intelligence can lower the time-to-market and cost barriers for medical research, thereby spurring progress in cases where there are limited financial incentives for pharmaceutical companies.
Pioneering companies are already realizing the potential of this technology, combining biological science with deep learning to discover new treatments for rare genetic diseases typically without lengthy and expensive research into new medications. The data generated from these applications over time can also become a resource for additional software to mine, and help explain why certain drugs work or suggest the most promising avenues to explore.
Unlocking healthcare breakthroughs for all
It’s no accident that researchers have relied heavily on open source initiatives to support their discoveries, and the trend towards open ecosystems only continues to build momentum. Using free, widely available internet images and open data platforms, researchers can engage in industry-wide collaboration to improve patient outcomes through artificial intelligence.
Why open source? Open source platforms offer feature-rich software created by a community of purpose-driven developers. Software portability between hardware vendors avoids hardware lock-in while still extracting the optimal performance. Business, academic and government organizations shouldn’t be chained to a single-vendor solution.
Open platforms, like the Radeon Open Compute Platform (ROCm), are vital to improve the access to math libraries, and provide a rich foundation of modern programming languages, which can speed up the development of high-performance, energy-efficient heterogeneous computing systems. And flexible, programmable compute-oriented GPU hardware such as Radeon Instinct accelerators will provide more choice in a market that was once locked to limited vendors.
As we continue our journey in the machine intelligence era, new GPU accelerators combined with optimized, open source deep learning frameworks will help solve some of healthcare’s most pressing challenges. We’re just at the beginning of this fantastic voyage, but are already witnessing the promise of how machine intelligence can help form a healthier world than ever.
