Severely premature infants can face a staggering list of potential health problems. Retinopathy of prematurity — abnormal retinal blood vessel development that can lead to blindness — is one of them.
It’s treatable while infants weigh less than two pounds, but doctors often struggle to detect the condition in time, says Santosh Rao, head of the artificial intelligence, machine learning and deep learning platform for NetApp.
“Traditional techniques have been unable to spot, let alone cure, infant blindness,” he says.
But that is changing: Researchers at the Athinoula A. Martinos Center for Biomedical Imaging in Boston have found applying deep learning techniques to the analysis of retinal images results in highly accurate diagnosis and severity scoring of ROP — results that can match or exceed clinician-based techniques.
It’s one of many ways artificial intelligence and deep learning are increasingly fueling life-changing interventions in a wide range of medical specialties and patient populations.
How AI Is Shaping Diagnosis and Care
Cancer care is among the areas set to benefit. With a radiologist shortage in some markets, AI- and DL-based diagnostics could provide a powerful augmentation, Rao says.
“Our healthcare customers are starting to see the ability to spot cancer in earlier stages — to be able to classify the tumors as either benign or malignant, and to be able to spot cancer nodules that are so small that actual manual inspection of the images will never find them,” he says.
Quicker diagnosis has substantial benefits. Diseases caught in earlier stages often can be more easily and successfully treated, often at a lower cost.
Automation also can aid individual treatments with lifesaving potential. “The ability to apply personalized medicine, specifically in the form of personalized gene treatments, is now emerging as an exciting area of research that may be an alternative to traditional treatments,” says Rao.
The benefits aren’t limited to those patients facing cancer. Researchers at HeartFlow, a digital health company focused on transforming how heart disease is detected and treated, have been able to use deep learning to construct 3D models of individual patients’ coronary arteries, reducing the need for invasive angiograms by 61 percent and cutting expenses by 26 percent.
Additionally, the technology could play a role in the development of new pharmaceutical treatments. Prescription drugs often take more than a decade and billions of dollars to develop.
Using the deep learning technique known as natural language processing, researchers can automate the process of surveying research literature to detect patterns pointing toward potential targets for drug development.
Ways to Incorporate AI and ML in Healthcare
Rao sees massive potential in vast pools of data that patients have entrusted their doctors to leverage for better care. He finds value in segmenting data sets to adapt treatments along a number of axes, including hereditary genetics, location, dietary habits, age groups and gender.
“There is a need to personalize and localize the use case applications,” he says. “The application of the technology will be decentralized and scaled across regions and populations.”
Still, there is room for caution.
“AI, as a technology, is an extremely powerful and autonomous assist,” says Rao, noting that privacy and ethical issues must remain at the forefront. “As with all things powerful and autonomous, it comes with the potential to be applied judiciously — for the greater good — or maliciously, in use cases that can cause harm.”
How can healthcare providers and researchers effectively integrate AI and deep learning into care delivery? The first step for providers, Rao says, is to identify the types of data they have or can collect: images, videos, audio and text are all options.
The second step is to consider the applicable techniques that involve AI and DL: detection, classification, segmentation, prediction or recommendation.
Four broad categories of use cases exist in medicine, Rao notes: imaging and diagnostics; patient care; back-end functions, such as pricing and risk management; and drug research and development — which is why relevant data, appropriate techniques and use case scenarios must align to fuel transformative discoveries.
“That is the journey that providers have to go through,” says Rao. “Only then comes the decision to procure vendors to deliver the full-stack solution that helps them through the lifecycle of implementation.”
Watch this:https://www.youtube.com/watch?v=Z5vxRC8dMvs