The way in which embedded integration is realized determines the success of AI-based systems in biomedical engineering applications — whether for diagnoses, defibrillators, or live images from minimally invasive surgery. Preconfigured embedded modules help developers find specific solutions for medical applications faster. Congatec has developed such a platform based on NXP i.MX 8 Plus processors and equipped it with an integrated neural processing unit (NPU) that allows it to contribute to the patient care process like a human would.

Defibrillator Gets the Heart Rate Back in Rhythm

At 6:43 pm on June 12, 2021, the soccer world held its breath: During the game between Denmark and Finland, Danish player Christian Eriksen’s heart stopped beating. He collapsed. Luckily, the doctors were close by and intervened quickly to save his life. One of the things they used was a defibrillator, which today can also be operated by people other than medical professionals. Thanks to modern computer technology, so-called AEDs (automated external defibrillators) can be programmed to deliver life-saving electric shocks and thus restore normal heart rhythms completely automatically. All that needs to be done to save a life is to attach two adhesive electrodes between the right collarbone and the left armpit — the rest is controlled and computed by integrated processors such as those in the i.MX 8 application processor platform from NXP.

Biomedical Engineering Applications Using NXP Processors

The defibrillator is just one example of a medical use for NXP technology in e-health solutions, which range from patient admission through obtaining of findings and diagnosis to surgery and bedside care. To boost the efficiency of medical applications with artificial intelligence, NXP equipped its new i.MX 8M Plus processor with an NPU for deep learning and machine learning.

With AI-based computer technology, AEDs become lifesavers that can restore patients’ normal heart rhythms completely automatically. (Credit: Alessandro Melis/Dreamstime.com)

Inspired by the architecture of the neural network of the brain, neuromorphic processors are optimized for pattern recognition, so they require much less power in watts for performing such tasks than classic CPUs do. At the same time, such low-power NPUs are capable of several tera operations per second (TOPS) and thus present new possibilities to developers of smart e-health solutions.

Around-the-Clock Digital Healthcare Assistant for Enhanced Patient Well-Being

The NPU-accelerated NXP i.MX 8M Plus processor can reach 2.3 TOPS with its AI engine and also offers four Arm Cortex-A53 cores and one Arm Cortex-M7 controller for conventional computing tasks. It also has an image signal processor (ISP) for parallel real-time processing of high-resolution images and videos. With this, it can preprocess and prepare imaging data during image acquisition for faster supply of clearer results by the NPU. This is advantageous in all imaging procedures as well as in minimally invasive surgery, where precise visual information is needed in real time.

In anesthesia machines, too, i.MX processors with AI accelerators can help OP teams precisely administer and control anesthetics to patients. A monitoring device controls the flow of compressed gases for anesthesia machines before they are passed through a vaporizer and the resulting mixture is expired via the patient’s respiratory system.

The Congatec starter kit for AI vision integrates the individual components, with a Basler dart camera acting as the camera, a SMARC 2.1 carrier board with 2x MIPI CSI as the communication interface, and a SMARC 2.1 module as the processor. (Credit: Congatec)

AI in Radiology and at the Bedside

With AI in radiology, it is possible to analyze information from x-ray and ultrasound machines and use it when obtaining findings. For example, a doctor examining a lung with a CT scan can look at the entire thorax and thus also be able to detect calcification of the coronary arteries. This additional information is helpful when it comes to comprehensively planning and initiating the treatment for a patient. An AI system compares the patient values with reference values stored in the software. Any deviations or abnormalities are marked in an image so that the doctor can examine the corresponding areas more closely.

At the bedside, a processor with an NPU is a digital partner that intensifies the patient care. AI in electrically operated beds can learn which positions a patient prefers and what kind of support the patient needs when sitting or getting up. AI applications also monitor vital signs such as blood pressure, temperature, or ECG data and control infusion pumps. This enables controlled delivery of fluids such as blood, medications, and nutrients to patients via the circulatory system. In addition, AI solutions can support medical staff in communicating with patients — especially when the patient is no longer able to speak clearly due to a medical condition.

Simplified Human-Machine Interactions

Control of medical human-machine interfaces (medical HMIs) via touchscreen, voice, or augmented reality is also included here. This not only simplifies operation of patient monitoring systems during anesthesia or at the bedside but also, thanks to the integrated AI processor, relieves physicians and care staff through activation of the corresponding early warning systems. In addition, AI solutions are increasingly being used in so-called hospital admission machines for patient admissions to hospitals. They support staff in emergency admissions by providing recommendations for the treatment order based on vital signs data such as blood sugar or lung function in real time to avoid dangerous bottlenecks. Using AI to create a digital twin of a patient to test therapies virtually in advance is even being envisaged for the future. This would allow doctors to assess possible effects of medications or therapies and minimize possible risks.

The important thing in all these use cases is that the medical devices must have the computing power to record and evaluate information autonomously. This calls for AI-based edge systems that work in operating rooms, emergency admissions departments, or intensive care units independently of any cloud connection. Through this, data evaluation in the cloud is avoided and, as a result, latency is reduced, and devices remain operational regardless of the availability status of the network.

Compact Starter Kit for Patient Care

To accelerate the use of the i.MX 8M Plus processor in all these possible AI applications, a starter kit enables developers to implement edge biomedical engineering applications quickly and safely. Given the diversity of possible medical applications, it goes without saying that the starter kit has to allow embedded applications to be designed to fit perfectly. The kit is supplied with a credit card-sized SMARC 2.1 computer-on-module with an i.MX 8M Plus processor.

The integrated NPU accelerates the local execution of AI inference at the edge completely without a continuous cloud connection. For vision-based tasks, the module integrates hardware-accelerated video encoding and decoding in the particularly data-efficient H.265 video compression standard. Through this, even high-resolution video streams from the two integrated MIPI CSI interfaces can be transmitted directly over the network to other e-health solutions or to the hospital PACS (picture archiving and communication system).

Tailored Adaptation Via eIQ

On the software side, an “eIQ Machine Learning” platform — “eIQ” stands for edge intelligence — provides developers with access to a development environment for their AI-based systems. The platform combines different libraries and development tools and is tailored to microprocessors and microcontrollers from NXP.

Software-based inference engines that apply logical rules to existing data and findings in order to deduce new facts are included. The eIQ platform supports inference engines and libraries such as Arm Neural Network (NN) and the open-source-based TensorFlow Lite.

The Seeing AI Eye

The central processing unit for AI vision: The conga-SMX8-Plus SMARC 2.1 module brings neuromorphic intelligence to the edge at a thermal design power (TDP) of just 6 W. (Credit: Congatec)

The application-specific design is ultimately implemented via the 3.5-in. carrier board. It serves as the central interface for data communication and links all the required peripherals to the module. For example, it features two MIPI CSI-2.0 interfaces for easy connection of cameras without the need for additional converter modules. The starter kit comes with the Basler dart BCON MIPI camera, but any OEM camera can be connected as long as it is compliant with MIPI CSI and is supported by the Congatec kit.

On the software side, the Basler pylon Camera Software Suite even supplies a uniform SDK that is also suitable for interfaces other than MIPI CSI-2.0 and allows control of cameras with the USB 3 or GigE standard. Thanks to the integrated Basler software package, developers have immediate access to central AI-based vision functions such as triggering, which enables ultrafast provision of stills and highly differentiated camera configuration options. Moreover, the software offers easy access to customer-specific inference algorithms based on the ecosystems of Arm NN and TensorFlow Lite for full utilization of the performance capabilities of the NPU.

The preconfigured starter kit not only saves time in the development of medical vision systems but also provides the assurance that comes from using technologies that are established on the market as well as known standards. The combination of Arm Cortex-A processors and NPUs in medical devices allow them to make intelligent decisions by drawing the right conclusions from sensor data with little or no human intervention.

AI as an All-Around Helper in Medicine

AI is used in medical practices, hospitals, and care facilities to support the important tasks of the medical staff in various settings. Whether with or without visual image processing, individual e-health solutions can be developed for concrete applications with AI via deep learning and machine learning. Simple and fast application development plays a key role in making it possible to explore the innumerable options of artificial intelligence. In this case, pre-evaluated starter kits pay off quickly because they speed up the development work. In addition, they offer wide-ranging options for exhausting the computing power at the edge that processor technology can supply through intelligent NPUs.

This article was written by Martin Danzer, Product Management Director at Congatec, Deggendorf, Germany. For more information, visit here .