Artificial intelligence meets the artificial pancreas

By Thomas Søderholm, Business Development Manager - Health, Nordic Semiconductor

0
245

According to a biological physicist, Philip C. Nelson, chemically, humans are not much different from a can of soup. From an engineering complexity standpoint, however, the body is rather more remarkable.

Take for example the pancreas. This organ, which primitive medicine assumed was little more than a shock absorber to protect the stomach from the spine, is in fact the body’s factory for a range of digestive enzymes as well as insulin and glucagon, the hormones that ensure both our brains and bodies have a steady and reliable supply of energy. It’s near enough as fundamental to life as a heart that beats, just ask any of the nine million or so people across the globe who suffer from Type 1 diabetes.

Type 1 diabetes is a chronic disease that renders the insulin-producing cells in the pancreas ineffective, and left unchecked can damage the heart, blood vessels, eyes, kidneys and nerves, and, if untreated will result in coma and death. A hundred years ago, before the disease was properly understood, people with Type 1 diabetes had a lifespan of little more than a year or two. With a deeper understanding of the disease and advances in medical technology, today’s prognosis is considerably brighter. But living with diabetes remains difficult and still requires the individual to measure and control their blood glucose level by taking insulin either by subcutaneous injection or through an insulin infusion pump.

This ritual is required often up to four or five times a day, every day, for the rest of the person’s life, and will be until a cure is found. In the meantime, at the intersection of medical and wireless innovation is the promise of a much easier life for sufferers in the form of the ‘artificial pancreas’.

The art in the artificial

Unlike prostheses that substitute for other parts of the body that are either ‘missing’ or defective—lost limbs, arteries, heart valves, eyes, or teeth for example—the artificial pancreas doesn’t physically replace the faulty organ, but rather mimics it from outside the body using technology.

Two devices make up the artificial pancreas’ ‘closed-loop system’: A continuous glucose monitor (CGM) and an insulin infusion pump. The infusion pump can administer insulin at any time and for any duration depending on the user’s blood glucose levels. Those glucose levels are determined by the CGM which measures blood glucose levels every few minutes using a sensor inserted under the skin that then wirelessly relays the data to the pump using Bluetooth LE wireless connectivity. The infusion pump uses a complex algorithm to calculate how much insulin is needed to maintain blood glucose levels within normal physiological thresholds. The pump delivers the insulin via a catheter over an extended period to prevent the medication from pooling under the skin.  

In part thanks to the convenience of wireless technology, today’s treatment has transformed the lives of hundreds of thousands of Type 1 diabetics. And the developments keep coming; for example, the latest infusion pumps now also use Bluetooth LE to communicate with smartphones enabling the mobile devices to perform long-term data analysis and offer healthcare recommendations. Cellular connectivity also allows continuous contact with remote physicians – who can check if the patient is appropriately managing their condition and remind them to do so if they aren’t.

Advanced intelligence

While the contemporary artificial pancreas is impressive, it falls far short of the performance of the real organ. Mother nature has had plenty of time to evolve the pancreas and has done an admirable job. Engineers must meet much shorter project deadlines but are working hard to close the gap.

The next step is to use machine learning (ML) to refine the closed-loop control of the artificial pancreas while also factoring in physiological parameters such as the patient’s overall health, core temperature, heart rate, stress levels and sleep patterns – all of which influence blood glucose levels. The availability of such data would allow reinforcement of ML algorithms to plan the insulin infusion regime with a much greater level of sophistication and accuracy than is currently possible, and without the need for the patient to ever have to adjust the pump manually.

Advanced software requires advanced hardware support. Devices such as the Arm Cortex-M4 and -M33 embedded processors on Nordic’s Systems-on-Chip (SoCs) and Systems-in-Package (SiPs) are more than capable of providing this support. Moreover, the Arm M-class processors are well suited to running TinyML, a scaled-down form of ML suitable for running on connected devices with optimised processor and memory resources such as infusion pumps.

And the cellular IoT capability introduced by the nRF9160 SiP introduces direct connectivity between the infusion pump and the Cloud without the requirement for a smartphone gateway – opening up a host of remote ML and monitoring possibilities.

While the mechanical and electronic elements of the artificial pancreas are mature, the key to future improvement will lie in further refinement of the software algorithms, and the ability of resource-constrained chips to support them. That future is much closer than you might think.