We Stand with Ukraine
Home / Blog / 4 Data Types Commonly Used by AI / ML Algorithms in Medical Devices

4 Data Types Commonly Used by AI / ML Algorithms in Medical Devices

Paul Kovalenko Paul Kovalenko | March 9, 2022 | 7 min

Artificial intelligence and machine learning are revolutionizing every industry: cars drive themselves, lending firms are getting better at calculating possibilities of repayment, and education platforms can now automatically find the best learning paths for individual users.
However, healthcare is one of the top-5 industries where AI and ML make a real difference. From assessing public health to predicting cancer, technology is enhancing the work of doctors and helping them to provide more accurate treatment. In 2021, the market size of artificial intelligence in healthcare reached $6,9 billion, and it is projected to grow to $67.4 billion by 2027.
The use of machine learning and artificial intelligence in medical devices is drastically growing. Thus, we are exploring AI and medical devices as well as the data types that power the algorithms.

What Medical Devices Use Artificial Intelligence and Machine Learning Algorithms?

Devices That Detect Health Conditions

Medical devices use AI and ML algorithms mostly for diagnostics. For example, they can analyze a patient’s cardiogram, paired with information about age and gender, and diagnose heart diseases with a sensitivity of 82.5% and a specificity of 92.2% which outperforms physicians significantly. It also has the same use case in diagnosing cancer, brain diseases, diabetic retinopathy, liver diseases, and diabetic retinopathy.

Devices for In-Vitro Diagnostics

Another application of artificial intelligence and machine learning in medical devices is in-vitro diagnostic tools. Technology has learned to recognize different types of cells, count them, and analyze the results.

Tools for Public Health Analysis

Some AI medical devices are used to assess public health: they analyze the health histories of as many citizens as possible and make a conclusion based on it.

Biosensors That Predict the Probability of Diseases

Lastly, there are AI-based biosensors that predict trends and probability of disease, as well as notify its users of dangerous vital signals and give recommendations for health improvement.

Need help with implementing AI/ML in medical devices?
Reach out to Langate

How Are AI and ML Algorithms Improving Medical Devices?

By using AI and machine learning in medical devices, doctors can collect an enormous amount of data and analyze it. So basically, scientists give the machine thousands of lab tests or pictures of diseases, and artificial intelligence medical devices analyze, compare the tests to each other and pick up the trends that humans could never notice because they cannot contemplate so much data simultaneously. Technology sees the bigger picture.
So next time an individual comes for a consultation, the technology does not compare their health history but considers their profile and compares it to all the other profiles that they have seen before. And it has seen more than one physician ever could. As a result, a machine’s diagnosis is much more accurate and faster too.
When it comes to public health, AI in medical devices is likely to pick up current trends and signs of epidemic much earlier and notify the government about the dangerous crises in advance. This enables healthcare organizations to stop the disease before it gets too bad and save millions of lives.

What Data Types Can Be Analyzed by AI/ML Algorithms in Medical Devices?

To provide accurate diagnosis and trend predictions, scientists use different data types of artificial intelligence algorithms in medical devices:

Analysis of High-Resolution Vital Signals

Machine learning algorithms in medical devices can analyze such vital signs as single-lead ECG, SpO2, heart rate, temperature, and blood pressure. It can detect anomalies very well since it is mostly about mathematics.
There are two ways to teach artificial intelligence algorithms in medical devices: supervised and unsupervised.

  • The supervised method means that data is carefully labeled by technicians before it is given to the machine. It usually helps to facilitate more controlled and accurate learning, but it does require a lot of effort and extreme attention to detail.
  • The unsupervised method means giving the machine all types of data and letting it figure out the rest. The technology should label and classify the data itself. Sometimes it may not go the way you expect it to. However, it requires less time and effort on the human side.

Now, there are two public data sets available for supervised learning – the MIT-BIH database and PTB Database. However, they are not enough for building an AI/ML medical device because it needs data collected by the device as well. At this point, you can decide between supervised and unsupervised methods. The first one will require more manpower but the second one seems to be feasible as well. At this point, it is still early to say about the efficiency of both.

Need help with implementing AI/ML in medical devices?
We can help

Practical Application of High-Resolution Vital Signals in Medical Devices

Today, you can already find the practical application of this data type – a single-lead monitoring device Zio monitor. It is a heart monitor that can be used even at home because it is very user-friendly and does not disrupt a person’s life. To date, it has already analyzed 4 million patients and used a supervised method with annotation from real physicians.

Analysis of Speech

Analysis of speech means processing human speech. There are several challenges when it comes to this data type.

  • Firstly, it is difficult to separate speech from the background noise, identify who tells what when there are several people taking part in the conversation, and identify separate words. These issues can be eliminated by either making records in a quiet room and high-quality audio or using open-source solutions and public APIs. The most popular APIs are by IBM, Amazon, Microsoft, and Google. Yet, you have to choose the one with the most fitting performance characteristics for your project.
  • Secondly, it is about speech recognition, converting the speech from audio to text. This task is normally performed with the following methods: tf-idf, bag-of-words, or simple keyword search. You can also use established solutions like Microsoft’s Text Analytics for Health or Amazon’s Comprehend Medical.

Practical Application of Speech Analysis in Medical Devices

One of the best practical applications of this data type is project EmpowerMD from Microsoft Research. It listens to the conversations between the patient and the doctor and automatically generates a medical note. It helps doctors to pay attention to actual patients rather than to typing the information on their computers.

Analysis of Streaming Auscultation Audio

Streaming audio means analyzing sounds rather than speech, for example, heart or lung sounds. It uses the same techniques as vitals analysis, but you will have to use a different database – a respiratory database on Kaggle. You will still need to mine some data from your device though.

Practical Application of Streaming Auscultation Audio Analysis in Medical Devices

A successful example of this data type being used in medical devices is Strados Labs which is a system that remotely measures, records, and analyses lung sounds to provide early predictions of worsening diseases.

Analysis of Population Health Data

Analysis of population health data is about drawing insights from the surveillance data concerning existing disease trends. It is one of the most underdeveloped research areas because there is not enough data available: public datasets are often outdated, suffer from outages, or restrict programmatic access.

Practical Application of Population Health Data in Medical Devices

One of the projects that became successful is Berg. Their AI algorithm analyzed biological (lipid, metabolite, enzyme, and protein profiles) and outcome data and discovered the pathway from healthy to diseased cells. Using this methodology, they expect to learn how to fight cancer.

Our Expertise

Langate has been working with healthcare technologies for more than 15 years now. We know how to work around different healthcare data types and are excited to implement machine learning in medical devices.
If you have any idea of developing an AI/ML-powered response device, do not hesitate to reach out to us. We will create a detailed plan of which technologies and libraries to employ in your project, what data types to use, and how to successfully collect and analyze the data. Our highly experienced data scientists at Langate will help you turn the most ambitious idea into a fully functional project.

Conclusion

Artificial intelligence and machine learning have the potential to revolutionize healthcare by automating certain tasks, driving more accurate diagnoses, and predicting public health threats.
Today, healthcare mainly focuses on four main data types of machine learning algorithms in medical devices: vitals, speech, audio, and population health data. At the moment, they all have some challenges, but it seems like we are getting there.
Do you want to become a part of that change? We can help you – just fill out the form below and let’s talk about your ideas.

Latest insights

View more
We use cookies to ensure that we give you the best experience on our website. By continuing to browse this site you accept the use of cookies.
I Agree
Privacy Policy