Can a person’s DNA help guide how their disease is treated? Through the use of big data and machine learning in precision medicine, the answer just might be yes. Imagine a future of patient care with improved diagnoses, personalized treatments, and better outcomes.
This patient-driven approach to medicine is one of the defining features of precision medicine. It stands in contrast to the “one-size-fits-all” method, where patient care is developed for a defined “average person”1. While this latter approach provides a standard treatment plan for large populations, it falls short of finding the best option for people who are outside the norm. Filling this gap will take the support of organizations that use big data and deep learning to determine the right care for a specific patient.
The Hadley Lab is one such institution that is taking a computational approach to medicine – bringing data-intensive technology to a field that has historically depended on written records. Based out of a public research university and medical and biological research hub, the lab is the namesake of Dr. Dexter Hadley, MD, PhD, who serves as its Principal Investigator.
Deep Learning Helps Lead to Digital Health
Dexter and his team are helping lead the movement to digitize personal health. Digital health brings together the accessibility and convenience of mobile devices (such as wearables and health apps) with the insights of genomic sequencing. This wealth of information helps people to understand their medical profile and manage their personal health. At clinics, this medical data assists doctors in providing more precise and personalized care to their patients.
Wearables and mobile health apps help people track their heart rate, nutrition, sleeping patterns and other important health factors quickly and continuously.
Deep learning is taking genomics a step further. By analyzing patients’ medical data, deep learning has the ability to trace critical patterns of disease within populations – such as skin cancer. To this end, the Hadley Lab created Skin2, its digital health imaging app for researchers, physicians and patients to track moles and skin lesions.Using patient-provided data, Skin is a shared platform for medical researchers to create more accurate models to diagnose skin disease. The mobile app uses deep learning to help identify factors in moles and skin lesions that contribute to specific clinical and pathological skin diagnoses.
One of the major goals of this platform is to reduce the number of overdiagnosed patients – those who mistakenly undergo a biopsy when their skin disease is in fact non-cancerous. Dexter notes that getting to this point will take the collection of thousands of images of patients’ skin conditions to train deep learning algorithms. After all, medical researchers can only do so much with artificial data; they need actual data from patients. It’s a problem of data scalability, which might only be solved with a top-down approach that opens access to anonymized health data.
Challenges in Medical Data Accessibility
Since medicine is a highly regulated field, getting access to patients’ data is a challenge. In fact, Dexter sees it as the biggest challenge his lab faces. It’s a double-edged sword: more medical data is needed for deep learning to evaluate diseases and improve potential treatments, but legal barriers often restrict access to this data.
There’s also the issue of opportunity cost. Some medical institutions profit by collecting, organizing, and cleaning data to sell to third-parties. Accordingly, by donating or using this data for research, there could be a big hit to a current revenue stream.
Deep learning moves beyond genomic sequencing to give insights for doctors and patients to help prevent and manage disease.
In other cases, even when medical data is voluntarily supplied by patients, medical professionals can make mistakes in evaluation. Take breast cancer screenings, for example. For women who have a single mammogram, the chance of having a false positive is generally 7 – 12 percent3. After 10 mammograms, this number rises to nearly 50 – 60 percent4. With such a high error rate, there is an opportunity to use deep learning for visual classification of breast cancer – potentially reducing false positives, improving the accuracy of diagnoses, cutting down medical costs and saving time.
“Given enough data, convolutional neural networks – a type of structure in deep learning often used to analyze visual imagery – can find signals in mammograms that a human eye couldn’t identify,” states Hadley.
Bringing Algorithms and Oversight to Medical Schools
Hadley and his team have made use of an open genomics discovery web platform in their medical research. The digital tool, which integrates with other large databases to help find more optimal solutions to treat diseases, has a time-consuming piece: data labels. Known as the Search Tag Analyze Resource, or STARGEO, the federally-funded project makes use of medical students – many of whom have no programming experience – to manually label genomic signatures of disease. There is more to be done to democratize medical data, Hadley argues.
“There is an ethical responsibility to teach medical applications of artificial intelligence by building these into medical school curriculum,” Hadley asserts.
To train the next generation of medical doctors and researchers, Dexter makes the case that computational sciences could be worth rolling into medical school curriculum. That includes teaching topics such as the roles of artificial intelligence in medicine – with a board of medical experts providing proper regulatory oversight. In turn, these new physicians and medical researchers could be empowered to apply AI ethically to solve some of the biggest challenges in human health.
More on the Future of Medicine
- Find out how data is accelerating research and healthcare innovation
- Learn how Western Digital helps advance patient care and enable future discoveries
- What is precision medicine? https://ghr.nlm.nih.gov/primer/precisionmedicine/definition
- Skin. on the App Store. https://itunes.apple.com/US/app/id1319397602
- Mammogram Accuracy – Accuracy of Mammograms. https://ww5.komen.org/BreastCancer/AccuracyofMammograms.html
- Screening & Early Detection References. https://ww5.komen.org/BreastCancer/EarlyDetectionReferences.html