Last week, researchers from the Bakar Computational Health
Sciences Institute of the University of California San
Francisco published a call to introduce deep learning healthcare.
Deep learning (also known as deep structured learning or
hierarchical learning) is part of a broader family of machine
learning methods based on learning data representations, as
opposed to task-specific algorithms. In healthcare, this means
gathering all the data recorded by doctors in patient Electronic
Health Records (EHRs) and forming a digital knowledge base that
includes various patient demographic, personal and clinical data,
as well as all the past diagnoses, clinical decisions and treatment
outcomes. This data lake would then be harmonized and
preprocessed before allowing artificial intelligent systems to one
day generate computer-aided diagnosis and treatment plans.
In fact, developing a smart EHR is nothing new as the idea to
utilize the millions of patient EHRs to create a Learning Health
System (LHS) came about almost a decade ago, whereby doctors
could mediate the learning of the system in order to inform the
day to day decision making of medical practice and influence
healthcare policy. Today, the call is for deep learning to allow
electronic medical systems to improve themselves without any
human (especially day to day doctor) input.
The future is now, since in the USA close to 98% of all hospital
systems now use an EHR and already perform complex tasks such
as Computer Provider Order Entry (CPOE) for nearly 80% of
medical orders that are captured electronically resulting in over 1.7
billion prescriptions packed per year.
Unfortunately, there has not been much progress yet in
healthcare beyond the important first step of data collection. In
fact, even this preliminary step needs to be further improved
(essentially digitized) as most EHRs require the doctor to spend
valuable time inputting the data. Eventually, EHRs could be
populated by patients themselves through the use of wearable
devices, implanted sensors or genetic samples. This way, doctors
can instead focus the vast majority of their time on the critical
clinical decision-making process of treating the patient.