Artificial intelligence in health care – how Apple Watch becomes a portable ECG. Interview with a Sparkbit expert.
The development of new technologies affects almost every sector of our activity. Among those areas, one raises special interest – it’s obviously the medical sector. How does the use of artificial intelligence algorithms and big data help in taking care of our health and how do such companies as Apple find their place in all that? You can read all about this in our latest interview with Sparkbit CTO, Jędrek Fulara.
- At the end of last year, Apple commenced in cooperation with the Stanford University the Heart Study project. What does it entail and what can be its consequences?
Apple Heart Study is one of the most impressive examples of practical use of artificial intelligence mechanisms in medicine. This is the first completely open clinical test which is devoted to detecting arrhythmia. Practically every user of an Apple Watch can participate in the program. The purpose of the study is to assess the effectiveness of arrhythmia detection in the overall population by automatic monitoring of the heart rate using the sensor installed in the watch. What differentiates this study is the fact that it’s completely non-invasive, does not require additional specialist equipment, such as an ECG device.
The process of testing is extremely simple – it’s enough for the participant to wear the Apple Watch for the appropriate amount of time. A specially developed application, on the basis of collected information, detects all anomalies appearing in the data and, as a result, identifies the probability of having a heart defect.
The traditional way to assess the heart rate is to make an ECG record based on measurement of the difference in electric potentials read from the chest and limbs. In the case of Apple Watch, to calculate the heart rate, the sensor installed in the watch uses green LEDs flashing hundreds of times per second, while photosensitive photodiodes detect the volume of blood flowing through the wrist. The unique optical structure of the sensor collects signals from four various points on the wrist and, after combining with complex software algorithms, Apple Watch separates heart rates from other unnecessary data (noise). Additionally, the heartbeat sensor may also use infrared light. On the other hand, green LEDs are used by the device to measure the heart rate during training and to calculate the heart rate variability (HRV).
Of course, the assessment of the heart rate wave is characterized by smaller diagnostic value than ECG records, however, taking into account the commonness of the use of Apple Watches and the ease in monitoring using this device, Apple pulse meter allows for collecting a greater quantity of data, and thus facilitates the work of algorithms detecting arrhythmia.
It’s for sure one of the most popular examples of the use of big data in the health sector – though certainly not the only one. The use of machine learning mechanisms, large data packages and the Internet of (Medical) Things (IoMT) in health applications is becoming a more and more common phenomenon. Thanks to these tools, we can encourage consumers to proactively manage their lifestyles, we can give them greater control over their health and, as a result, improve their well-being. Additionally, artificial intelligence increases the capacity of health care employees to better understand the daily patterns and needs of people they take care of, thus allowing them to make more apt diagnoses.
- Is it the only example of the use of artificial intelligence and big data in health care?
Recently, the research branch of Google launched the Google Deepmind Health project tasked with collecting medical documentation to ensure better and faster health service provision. Already at the present stage of implementation, Google Deepmind is able to process hundreds of thousands of medical information within a few minutes. Another example of the involvement of Google in the use of artificial intelligence in medicine is the NHS Foundation Trust project created in cooperation with Moorfields Eye Hospital, aimed at streamlining eyesight defect treatment. In this case, analysis of collected data and creation of a machine learning algorithm are still in the initial phase.
Furthermore, Verily, a part of the holding established by Google, is working on its own initiative of genetic data collection. Its goal is to use some of the algorithms that power the well-known Google search button to analyze what makes people healthy. This also includes experiments with disease monitoring technologies, including a digital contact lens, which can detect the level of blood sugar.
IMB is also involved in projects associated with health. IBM Watson – in cooperation with Cleveland Clinic Lerner College of Medicine at the Case Western Reserve University – launched an initiative called WatsonPaths. The system is composed of two cognitive data processing technologies, which may be used by the artificial intelligence algorithm created by IBM – Watson. It’s supposed to help doctors make more conscious and right decisions faster, as well as obtain new information from electronic medical documentation.
Another example is the system created by CareScore. Using an algorithm based on artificial intelligence mechanisms, as well as combinations of clinical, laboratory, demographic, and behavioral data, the CareScore product is able to specify the probability of a given patient being admitted to a hospital. Thanks to the system, health care institutions can improve the quality of offered services, while the patients can obtain a clearer image of their health condition.
Of course, those are only the several best-known examples of the use of new technologies in medicine. However, the AI market is full of promising novices who want to change the health care for the better.
- What is the biggest challenge Big Data has to face today with regard to health care?
Although the use of artificial intelligence mechanisms and Big Data to improve health care has a huge potential, we can’t forget that they also carry many risks. Medical care and the patient’s data are issues that require the highest level of accuracy, reliability and privacy protection. On the one hand, accuracy and a certain faultlessness of artificial intelligence systems can help maintain the top standard of medical care, but on the other hand, we should remember that technologies are still in the initial phase of development. While AI systems may be trained with regard to complex datasets, in the clinical conditions we can encounter situations and scenarios for which the algorithm has not been trained, which may result in its smaller accuracy and consequently threaten the patient’s safety.
Another issue lays in the availability of information. Contemporary artificial intelligence is hungry for data. For speech recognition on an Android phone to be accurate, Google trains a deep neural network on a material consisting of approximately 10,000 hours. In order to recognize images, ImageNet uses more than 1,034,908 images bearing a manually written annotation. These annotations, known as labels, are necessary for techniques such as deep learning to work. A similar situation occurs with systems used in medicine.
Currently, data used for creating systems based on AIs originate mostly from medical tests, characterized by a certain degree of selectivity. As a result, there is a threat that technologies based on machine learning will deepen the existing inequalities. A considerable part of this risk results from the certain existing distortion of the currently available medical data. Information originating from randomized control tests are often very biased, filled with stereotypes. The highly selective character of system tests – on the basis of which machine learning mechanisms are created – unfavorably affects, among others, women, the elderly and people with rare diseases. Just imagine a situation in which the algorithm, taught mainly with focus on data concerning white men, tries to write a prescription and specify medication dosing for a pregnant woman. There may be many similar problems related to lack of an appropriate research sample.