Technology Trends and Standards in the Medical Devices Industry - MATLAB
Video Player is loading.
Current Time 0:00
Duration 41:37
Loaded: 0.09%
Stream Type LIVE
Remaining Time 41:37
 
1x
  • Chapters
  • descriptions off, selected
  • captions off, selected
      Video length is 41:37

      Technology Trends and Standards in the Medical Devices Industry

      Overview

      Are you interested in the newest developments and technology trends in the medical devices industry? Do you want to see examples of the next-generation medical devices that companies are creating to advance healthcare? Then watch this webinar to hear more about these important topics and apply them in practice. 

      The topics will cover the current regulatory standards and technology trends in the medical devices industry such as MDR and FDA regulations, Software-as-a-Medical Device (SaMD), Digital Health, Telehealth, In Silico Medicine, Digital Twins and AI-based medical devices. Furthermore, case studies will be shown where MATLAB and Simulink have been used by engineers and researchers to create these applications.

      Highlights

      • Insights into the latest technology trends in the medical devices industry
      • Overview of MDR and FDA regulations for AI and medical software
      • Case studies of next-generation medical devices from the industry

      About the Presenter

      Dr. Visa Suomi is the Medical Devices Industry Manager at MathWorks. He has over 10 years of international experience in the life sciences and healthcare sector, with an interdisciplinary background from the medical technology industry, academia, and clinical research. He holds a doctoral degree (DPhil in Healthcare Innovation) from the University of Oxford, UK, with the focus on translating academic and clinical research into commercial applications.

      Recorded: 26 Jul 2022

      Hello, everyone. My name is Visa Suomi. I'm the Medical Device Industry Manager at MathWorks. I work at MathWorks with many medical device customers throughout Europe. And in this presentation today, I'm going to talk about technology trends and standards in the medical device industry. And show some use case of how industry has used MathWorks solutions to develop these kind of applications.

      So first, preview of MathWorks. The MathWorks is a global software company. We have around 5 million users worldwide. We have around 5,000 staff in 34 offices around the world. We work in close partnership with many medical industry companies worldwide and in all the other industries. And we are a private company. So we focus on the long-term customer success.

      So at MathWorks we have two main products. The first one is MATLAB. MATLAB is this text-based graphical user interface, where you can create syntax language for developing algorithms and AI models for biomedical data analysis applications.

      The other product that they have is Simulink, which you can use to simulate complex medical devices, including their sensors and software and all the control algorithms. And that helps you to create this consistent level representation of your medical device. Additionally, we have over 100 add-on products that can be used for specialized R&D tasks, such as machine learning, signal processing, image processing, and many other applications.

      So the agenda for today's presentation is shown here. I first will talk about technology trends in the medical device industry and about the applications that we have seen our customers developing in this area. After that, I would talk about the regulatory standards and especially how the regulators see these health applications and other type of AI-based medical devices developing in the future.

      After that, I will talk about workflows and applications and how you can use better solutions to develop these kind of applications. And finally, I will give you some further reading and materials that you can have a look on your own time. And maybe do some exercises to better understand how they developed these applications.

      So here are shown five, the most prominent technology trends in the medical device industry. And these are the applications that we have seen our customers developing in their R&D. So the big underlying trend is digital health. And digital health basically means digitalization in health care. So you have more digital data available or stored in a usable format. And based on that data, you can use it to develop different applications. For example, telehealth services are used to remotely monitor patients and also treat patients remotely.

      Digital twins are virtual replicas of medical devices or patients. And AI can be used to predict the future condition of these patients or medical devices based on machine learning. In silico medicine means basically using computational modeling and simulation to virtually represent a patient or a medical device and make, for example, predictive treatment decisions.

      AI-based medical devices are applications where AI is used for one specific task to replace image processing or signal processing applications. And finally, cloud services are also becoming more common. Companies are building their services and software under both the cloud. And also, more and more patient data is stored on the cloud.

      So the first digital trend is telehealth, and it's also called telemedicine in some contexts. And it basically enables remote health services. By some estimates, telehealth utilization is now 38 times higher than before the pandemic. And some of the applications in this telehealth area include, for example, remote patient clinics and contact. So a patient and clinician can have interaction and conversations remotely rather than going to the clinic.

      Clinicians and surgeons can also do remotely medical treatments with a patient without having to actually be physically present at the site. And thirdly, patients can also be monitored remotely, say, using wearable devices or home monitoring solutions. And then, the clinician can interpret that data when it's stored on the cloud and see what is the condition of the patient.

      So here is a real-life example of a telehealth application. So one of our customers, Corindus, is a surgical robotics company. And they had developed this surgical robotic system in order to treat patients remotely in cardiac surgeries. So the way they developed this application is that they used Simulink, the first simulator robotic control systems, and tests how the robotic performs in the simulation

      Additionally, they had a remote link between the surgical robot and the clinician who was operating the robot with the control panel. And they had this real-time visual video link between the site where the patient was and the physician who was operating a robot, so that the physician or the surgeon could see in real time how the operation was going. And by doing so, they were able to perform the very world's first in-human telerobotic PCI surgery in India, where they used this robot remotely at the site, which are several tens of kilometers away from the actual site where the physician was operating the robot.

      So the second trend I want to talk about are digital twins. So digital twins is basically a virtual representation of a physical object or a system across its lifecycle. So what it means is that this virtual object collects data from different sources to have the dynamic history of that patient or maybe a medical device across its life. And based on this data, it's going to be used to develop different type of machine learning models for diagnosis or prediction.

      Some applications for this type of representation are, for example, personalized medicine. So in the case of a patient, you could see how the patient responds to different medical treatments or new developed medical devices. All the applications in terms of medical devices. You could have a medical device, and then you could predict what is the near-future condition of that medical device, and when is it going to fall.

      The last example are clinical workflows. For example, if you have virtual representation of a hospital, you could have different clinical units and help patients move around those units. And then you could use this virtual representation to optimize the workflows in the hospital setting.

      So let's have a bit closer look of the first example. This is a digital twin of a human. So when you have this virtual representation of a human, you could collect, for example, genetic data of that patient, the database. You could conduct laboratory studies, maybe some plot studies, and so forth. And the test results will be collected in this database.

      You could do some imaging studies, maybe MRI scans, or CT scans and collective imaging data. You could have wearable devices that collect biomedical signals, say, the heart rate and the activity data. And you could also have behavioral data collected for that patient, and also the social determinants. And when you collect all that data into one virtual representation of the patient, then you can make predictions, and also use personalized medicine to find out what would be the best treatment for a patient for a specific condition. And also what is the expected outcome from that treatment.

      Another example of a digital twin could be a medical device. So in this case, you can, again, collect all the data from the operational functions of the medical device in one single database. Say, the physical properties, the algorithms that the device is operating with, the maintenance history, some sensor data, some performance metrics on the operation, and also environmental factors, such as the humidity and other conditions in that environment.

      And once you have all this data collected in the database, and it gets continuously updated, you can use machine learning models to make predictions about the future condition of that medical device. And especially in safety critical applications, such as ICU units, you could then predict if the device is going to fail and if maintenance should be alarmed.

      A third example of a digital twin would be a medical facility. For example, a hospital. So in this case, you could collect data, such as the building layout; the equipment resources, set imaging devices, and everything in the radiology unit; the personnel resources; all the processes and workflows that happen in that environment; some usage statistics, such as how much the scanners are being used at what time of the day, how many patients are there in the waiting rooms, and so forth; and all the operational data of the hospital in one representation. By using this data, you can then optimize the usage of these devices and make sure that everything within the facility are used efficiently.

      So here is one practical example of a digital twin. So in this case, this is a hydraulic pump. It could be any pump, say, an infusion pump or even an insulin pump or something that is used in a medical facility. And in this case, this is simulating the model of the pump, so it has all the mechanics, all the algorithms, all the fluid workflows-- sorry, all the fluid workflows-- built into this one representation of a pump. By running this pump virtually, you can generate data for different sensors.

      Say, in this case, you have a pressure sensor collecting all the data from the outlet of the pump. And then you can simulate different kind of conditions, say, if the flow is too high or maybe there are some vibrations, and how does that sensor data then look like. So by simulating this pump, you can simulate all of the sensor data that is collected. And based on that sensor data, you can then start developing algorithms to make predictions for predictive maintenance of this pump.

      So you are simulating a lot of different situations that the pump might be operating in. Then you use this data that has been simulated to train a machine learning model, and that machine learning model could have a classification about the future condition of the pump and if maintenance should be used. And once you have this machine learning model, you can then take the code from the machine learning model and deploy it actually on the pump on the site.

      So now, you have a real, running machine learning model running on this medical device. And its sensor data, based on the machine learning algorithm of what would be the future condition of the pump. That is called predicted maintenance.

      So the third trend I want to talk about is this in silico medicine. So in silico medicine basically uses computational models and simulations to replicate humans and medical devices in a virtual environment. So these virtual patients would then be used, say, in clinical trials, when you want to replace real humans in the testing of a medical device or maybe drugs. And you could potentially conduct clinical trials or testing at a lot larger scale than you would be doing with your patients.

      So in the virtual environment, you have unlimited possibilities to have different patient demographics, different conditions of the patients. And also, the number of the patients can be very large. So this also helps conducting these trials virtually so that there is no threat to the patient of this new type of therapy method that is being developed.

      And some of these applications in medical devices include, for example, during the R&D process, you could use these virtual patients to do design verification of your medical device without having actually a physical prototype of the medical device. You can also test the performance of that medical device with different patients and with different patient cohorts. You could draw on clinical evaluation with different patient populations and see how your treatment, say, the drug combination, or maybe the medical device, works within a larger setting of patients.

      And you can use these in silico models also to do treatment planning. One example of this is actually if you had a virtual representation of a patient for radiation therapy, and then you calculate and estimate how much radiation is being absorbed by different tissues in the way of that tumor. And you conduct the treatment. So that's a real life practical example of in silico medicine that is currently already happening in some hospitals. In silico medicine is this computational environment that you're going to see more in these different type of applications.

      So here is an example of a in silico model of a ventilator together with a patient model. So basically this, if you have never seen it before, it's Simulink Canvas. There you can simulate medical devices and their functions together with patient models and make this functional representation of a complete clinical environment together with the medical device.

      So basically, what you have here is you have the patient model. In this patient model, you have a passive patient with certain body temperature, the moisture, and muscle pressure. And that mimics the lungs of the patient that is in the ventilator. You also have the respiratory mask for inhaling and exhaling with some humility going out. You have the valves for inhale and exhale. Those valves are connected.

      For the inhale valve, you have humidifier and a heating wire on that the inspiratory tube. And for expiratory tube, you also have heat transfer between the tube and around the room temperature. And the algorithm of this medical device is here in the controller. On this controller you have the sensors coming from the patient, from the lungs, to have this closed-loop system. And this controller, then, controls the inhale and exhale valve that the ventilator is operating.

      It is also making sure that there are no dangerous conditions within the patient while the ventilator is operating. So within this in silico model, you can actually test your ventilator design with a real patient model. And even though this patient is a simplified passive patient, you could have a lot more functionality to the patient, say, in the form of an active patient. But just to kind of represent a simple example, in this example, you have the patient on the ventilator.

      So here is another example of an in silico model with artificial pancreas. So again, this is simulated in Simulink together with the patient model and all the algorithms within that medical device. So what you have here is the patient model. So in this patient model, you have the blood circulation with some glucose levels. And the glucose is administrated in the blood circulation with the meals that the patient is consuming.

      In this case, you can adjust the glucose levels, and carbohydrates go into the blood circulation from this meal. You also have insulin pump connected to the diabetic patient. That is administrating insulin to control those glucose levels. The glucose levels from this patient then go through this continuous glucose monitoring system. And that glucose monitoring system is sending those readings into the controller.

      And within this controller, in this case, you have physiologic algorithm. But this could be also, say, a machine learning model or maybe a single machine algorithm that has the intelligence of this medical device. And this controller then controls the insulin pump that is administrating the insulin to the diabetic patient to control those blood sugar levels. So again, a simple example of how you can use in silico medicine to test new designs for a medical device together with a patient model.

      The fourth big trend are these AI-based medical devices. So machine learning and deep learning, they are obviously big trends in many industries, and the same applies to medical devices in the health care industry as well. And these type of models are becoming more and more common also in medical devices, where machine learning and AI is replacing traditional image and signal processing tasks. Say for example, classification, detection, and prediction.

      And AI has many use cases in health care and medical use. So for example, you can detect anomalies, say, in lung scans or MRI images. You can classify diseases, say, different heart conditions, or maybe you can classify some mental conditions based on the activity data. You can segment medical images when you want to do, for example, treatment planning. You want to segment, say, the lung tumors only if you are planning for radiation therapy, or you can segment the brain tumors to plan for a surgery, and so forth.

      And you can also use machine learning for outcome prediction, as I mentioned in the case of digital twins. So basically, when you have all the data of a patient or a medical device, you can predict the outcome of a specific treatment or drug combination or a future consideration of that medical device with AI.

      So here is a practical example of machine learning implement in the medical device. So in this case, MathWorks worked together with a company called Battelle, where they wanted to build a brain-computer interface, the wrist or hand movement. The patient had had a diving accident, and he had lost his ability to move hands in the lower body and also in the hands.

      So this company basically developed an algorithm that was acquiring and interpreting EEG signals from the patient when he was thinking about moving his hands or lifting something up. And basically by collecting this EEG data, this company could develop a machine learning model that classifies these EEG signals in the hand movements. The patient was also wearing this sleeve that was used to stimulate muscles within the hand.

      And therefore, this is kind of like a closed-loop system whereby the EEG signals were acquired, interpreted. And then the actions were forwarded at sleep, and the patient could move his hands. And by having this real-time system running in MATLAB, the patient was able to move his hands, pick up items from the table, and for extra, do some complicated tasks, such as picking up a mark and pouring water in another cup or moving items around on a table. So example of where AI can have real life-changing benefit when it's implemented in the medical device.

      So here is another example where AI was used in the design of medical devices. In this case, in the design of intraocular lenses. So basically, these intraocular lenses are implemented in the patient's eyes to improve their vision. And it's sometimes very hard to estimate how to create that lens and what will be the outcome of the patient in their vision once that surgery has been performed and these lenses have been put in their place.

      So this company basically wanted to improve the accuracy that the patients see after that surgery and the way these lenses could be optimized for that specific patient. And in this case, they used MATLAB optimization with AI algorithms to use this radial basis function to optimize the characteristics of these intraocular lenses for the specific patient. These algorithms were then implemented into this optical biometer, which is basically acquiring image data of the patient's eye. And by doing so, they were able to improve the eyesight in the patients. In 90% of surgical cases, they were able to achieve very high accuracy after surgery.

      The last trend on this list are cloud services. So cloud infrastructure and services are obviously becoming more and more common in many industries, and the same applies to medical devices and health care data as well. So a lot of the data storage and processing are moved to the cloud rather than having local facilities for database processing. And there are some benefits in doing so because these cloud services enable scalable and accessible health services.

      And by having this data stored in a centralized location where it's easily accessible, enables different type of services being conducted in the health care. For example, healthcare IoT. So we have a lot of connected devices, say, wearables or remote patient monitoring devices that are connected to servers for data collection. It also enables companies to build medical software as a service. So maybe you have a digital pathology workflow built on the cloud only, or maybe some AI-based radiology applications that you want to run on the cloud for image analysis.

      It also enables wearables or home monitoring devices that are collecting real-time data from patients. And then that data is fed into the cloud where doctors can interpret the data and make treatment decisions. And this is also part of the telehealth application that was mentioned earlier. So cloud Infrastructure is crucial for making these kind of telehealth services operational.

      So here is an example of a digital health application built on top of a cloud service. So a company could send their work together with MathWorks to develop these data-driven mental health therapy services by collecting data from wearable medical devices. So basically, the patient was wearing this bracelet activity tracker that was collecting a lot of sensor data on the movements of the patient.

      And by using MATLAB, they were building these AI and signal processing algorithms to interpret the data and get actionable insights from that data to estimate what was the mental health state of that patient, and were there any indications, for example, about depression. And this data was fed to the cloud, and the whole ecosystem was built on top of Amazon Web Services. They have this cloud Infrastructure that could be used to create this kind of new digital health application.

      So now that I have talked about the main trends in the medical devices and health care industry, let's look at next what the regulators say about these new digital health applications and what kind of standards already exist to develop these applications within the medical device industry.

      The first thing that I want to mention is this IEC 62304, which is basically a standard for how to develop software for medical devices, and how does the process look like with all the planning stages, the design, the verification and testing basis, and all the release of the software. And this standard is harmonized, which means that it has been recognized by international regulators, such as in European Union and in the US. This standard covers all the lifecycle processes that are needed in order to develop medical device software, all the way from planning and the software release.

      As you can see in this standard, the software has been categorized into three different safety classes, with Class A being the lowest risk class of medical device software, and Class C being the highest risk class of medical device software. And depending on the safety class of your medical device, there's a different amount of documentation that is required from the development process to show that you have completed the steps mentioned in the standard.

      If you are interested to learn more about this standard and how model-based design can be used to conform to the standard, we have a white paper available called "Developing . IEC 62304- Compliant Embedded Software for Medical Devices," which is available on our website, where we explain in detail on how you can develop medical software and be compliant with the standard.

      One other standard worth mentioning is this IEC 82304, which is kind of an extension of the IEC 62304, covering also standalone software applications for medical and health and wellness. So basically, this covers the development of something like this that had apps, something that you could use on your wearables or maybe your mobile applications when they have diagnostic purpose. And unlike the IEC 62304, which is more oriented toward embedded software development, this has completely the standalone software scope.

      Understand it's not exclusive to medical devices. It also applies to health and wellness software applications, which are not classified as medical devices but still have some kind of health care obligation. This standard is currently not harmonized. So it has not been recognized by the regulators worldwide, but it is still a reference when you are developing these type of applications.

      The next document on this list is not necessarily a standard, but it's more like an action plan on how FDA sees in regulating these AI-based medical device software application. So they have released this document on how AI and machine learning models should be implemented and tested and regulated in medical devices and also in the European Union. The new regulations, the medical device regulation, in vitro diagnostic regulation, they specifically now address all the digital health applications.

      And despite that these regulatory frameworks globally are still developing in terms of the applications, there are already several AI-based medical devices on the market that have received market approval from the FDA and also from the European regulators. These applications are especially wide scope in medical imaging and patient monitoring applications.

      Next in the agenda, let's move on to the workflows and applications and also how in practice you can use MATLAB and Simulink and some AI and digital health workflows to develop these type of medical devices and applications that I have showed earlier in the presentation. So we're building these new type of digital health applications. It is efficient to have end-to-end workflow within the signal environment, where you can access data. For example, you want to acquire data from sensors, maybe biomedical signals, you have access to files and databases, and you can acquire all that data, say, in real time or stored data in your software environment.

      You can then analyze this data, export, visualize it, do some processing. For example, importation or enhancement. And also label the data. In this part, you can also use domain-specific algorithms to do some fielding for ECG signals. The next part, you can then start developing these applications and the intelligence work. So you can start training AI models. You can start building algorithms, maybe signal processing, image processing, or control algorithms. And you can also incorporate modeling a simulation. So you can start more modeling the medical device or maybe the patient, and you want to see how your organs behave in a real clinical situation in the simulation environment.

      Lastly, once you have developed your model, so you have a AI model that you want to deploy on a real medical device or maybe in the cloud, you can do that, say, on the desktop, if you have access to a desktop application. You can do that on the cloud, or you can also have embedded code for embedded devices for integrating into a medical device. And when you use MATLAB or Simulink, you can do all these different steps in your workflow within a single environment. And you don't need to jump between different tools and solutions to process your data and build your application.

      Here's an overview of the different applications that you can develop with MATLAB and Simulink. So for example, maybe you're developing a new type of image processing algorithm, or maybe a computer vision algorithm, and you want to have efficient methods for approving those digital health applications. So you can do that in MATLAB. For example, you can have say machine learning applications. Or if you are developing a surgical robot, you want to simulate how the surgical robotics hardware interacts with the software by controlling those movements in surgical robot. You can simulate that and have an impression on how well the robot performs.

      And obviously, a lot more applications you could develop. You could develop antennas or power electronics for your medical device, do deep learning for image analysis, and also having the final object code generated from your model. Say, if you want to deploy your applications in medical devices, you can share it C, C++, ACL code from your applications, and then having that code embedded in their system.

      And also, if you want to have some kind of cloud applications, you can, for example, compile MATLAB applications to be deployed in the cloud. Or you can have highly efficient, secure code generated from your application and then deploy that on servers. So just an overview of the capabilities, and depending on your need, you might want to have a look in more detail on what we have to offer in that specific area.

      Lastly, I want to show a bit of materials and trainings that you can take advantage of when you start to develop your digital health applications. With our website you have, for example, these Onramp courses that are topic specific, short, two to three hour courses that you can do hands-on exercises to learn basically specific applications, image processing, or deep learning, or maybe machine learning. And have some exercises to do on your own time. These are completely free. They are available online.

      So for example, there's basic training for MATLAB, basic training for a Simulink, and also state for very many application-specific training courses. They have all of that more on our website on that regard. We have also a lot of webinars recorded on our website. So you have webinars on different topics, such as model-based design for example. If you want to learn how to develop medical device software and how to test and verify the software and also be compliant with medical device regulations.

      And then, if you want to train maybe a bigger crew in your company, maybe you have more employees that want to attend the same training. We have also on-site training, self-paced training online. So you can go online and book a course that fits your needs. We have professional trainers who can also come on site and do that training for the group of people in your company.

      We have also this white paper that I mentioned earlier, which expands on how this model-based design in practice, and how you can integrate verification and validation in your software development workflow, and how you can test complex medical devices. So this white paper basically gives you tips and best practices for getting started with model-based design and also has some case studies on the industry on how different companies have used model-based design in developing their medical devices. So if you want to have this white paper, you can email us at medical@mathworks.com. And we are happy to send you a copy.

      We have also this free e-book available, which explains how you can use model-based design to develop these AI-based digital health applications. So say you are developing a remote patient monitoring device and want to deploy those algorithms on the cloud. So in this e-book, we'll give you an overview of how MATLAB and Simulink can be used in the application development. And again, if you email us at medical@mathworks.com, we are happy to send you a copy of this e-book.

      And if you have a project in mind that you would want to use MATLAB or Simulink for developing digital health applications or medical devices, we can provide you with trials or test licenses for different products you can use in the development. And you can also get evaluation support from our application engineering team in getting started in the project.

      We also have consulting services that can help you implementing parts of your project. And as I mentioned, we have training services that can come on site to train a particular group of people. Or you can also conduct these trainings online, and therefore, getting started faster with your project. And we have also technical support available 24/7 online and on the phone. And from them, you can get help and feedback on your technical challenges that you might face in the development process.

      I want to thank everyone for joining this webinar. And if you want to learn more details about what MathWorks can help you with in terms of medical devices development, have a look at our website, mathworks.com/medical, where you can find more information. Thank you very much.

      View more related videos