Artificial intelligence, as a dividend of the fourth industrial revolution, will create better healthcare outcomes, says the new WEF’s report confirming that AI-driven health solutions have proven more efficient and have become even more effective.
From helping treat heart failure, to remote monitoring for patients during COVID-19 quarantine, there are numerous examples of real practices that are driving investment in digital and AI technologies within healthcare.
How is it possible to incorporate such technologies in a complex field that deals with patients and provide them care? One way is by using medical devices as a “second pair of hands” to the medical professionals.
Medical Devices in the world of AI
Medical devices with artificial intelligence hold the promise of revolutionizing the healthcare industry as they help in developing more accurate and effective diagnoses and patient’s treatments.
The techniques to incorporate AI in medical devices are based on deep learning techniques using neural networks.
According to the definition, neural networks refer to a set of algorithms that are modeled after the human brain and are designed to recognise patterns. When exposed to data through neural networks, the deep learning techniques of machines [medical devices] are able to mimic human learning patterns and change without being programmed.
In this article we are going to analyze AI software as a medical device and its applications. The term Software as a Medical Device is defined as "software intended to be used for one or more medical purposes that perform these purposes without being part of a hardware medical device." and can be used for:
- Prediction and identification of diseases
- Data classification and analysis for diseases
- Optimization of medical therapy and clinical outcomes
- Diagnostic support application
- Patient monitoring
- Automating the delivery of treatment
- Medical imaging
- Patient Monitoring System
A physician can attach a wireless device with sensors to a patient's wrist to monitor vital signs such as blood pressure. The device would use machine learning to detect significant changes in the patient's vital signs based on algorithms trained on large datasets of similar patient data. If an important change is detected, the data is transmitted to physician’s monitors or other devices to notify about the change.
Philips Healthcare’s solution is a patient monitoring system example, which uses AI to predict when a life-threatening crisis may occur in a patient for effective, early intervention.
In an independent study examining the use of the Philips’ solution in the waiting room of an emergency department, results showed that "178 patients (79 percent of all participants) were able to be discharged from the hospital after their visit to the waiting room" and did not require hospitalization. Nearly 70 percent of clinicians who used the system felt they could easily identify patients who needed immediate medical attention.
- Medical Imaging
The use of artificial intelligence in medical imaging can improve the speed and accuracy of radiological scans. Increased speed means less radiation exposure for patients, which can contribute to shorter treatment times and better clinical outcomes. For example, computerized tomography (CT), as a medical device that uses AI, combines images from multiple X-rays of the same object. Compared with a simple one-dimensional X-ray image, CT scans can produce cross-sectional images or slices of bones, blood vessels, and soft tissues inside the body. By incorporating AI into the process, algorithms are developed to reorganize small patterns of organ damage that a physician might miss by simply looking at a scan. By capturing these finer details, this technology has the potential to enable faster diagnoses and fewer errors. For example, GE healthcare in partnership with NVIDIA, are identifying liver and kidney lesions more quickly due to a new speedy CT system using AI computing platform.
- Cell Sorting and Recognizing Cell Types
A study published in Nature demonstrated the use of an AI and neural network in cell sorting. The results show that the AI system takes less than a few milliseconds to classify cells and communicate a decision to a cell sorter to separate individual target cells in real time. This study demonstrates the applicability of AI in classifying white blood cells and epithelial cancer cells with a sensitivity of 95.71% and a specificity of 95.74%, label-free.
Building trust with AI
Integrating AI into clinicians' daily work will not only significantly improve patients' treatments, but the overall healthcare as well. However, to gain the trust needed for broader adoption, AI in healthcare must follow the principles of building responsible AI, as algorithms can sometimes introduce bias into their analyses, which has harmful effects on real human beings, and transparency around technology's limitations.
Having understood the general introduction of AI in medical devices with this article and touching upon the process of building trust in AI systems used in medical devices, the next article will outline the path to establishing AI systems that comply with regulations, reduce risk and harms, and understand the human impact of the model.
With the rapid developments in machine learning algorithms and improvements in machine performances, the AI technology is expected to play an important role in performing healthcare practices, collecting, analyzing and monitoring medical data and patients' treatments and care overall. This will be done through the use of artificial intelligence in medical devices. According to research, “...The Artificial Intelligence Based Medical Device Market Ecosystem is Expected to Grow at a CAGR of 25.7% by Forecast Year 2027…”, it is clear that the adoption of AI enabled medical devices is one significant stepping stone into the future of AI in healthcare.
This article is part of the series covering the topic of AI and Medical devices. The next article will outline the path to establishing AI systems within medical devices that comply with regulation, reduce risk and harms, and understand the human impact of the model.