What is artificial intelligence?
Artificial intelligence (AI) generally refers to algorithms that are able to mimic human intelligence such as learning and problem solving. There are four different levels of AI, ranging from simplest to most advanced including reactive machines, limited memory, theory of mind and self-awareness. Although, the latter two are only theoretical at this point, the first two are widely used nowadays. Reactive machines are trained to do a particular set of tasks with some input information, such as recommendation engines and search algorithms. Limited memory machines are able to store input data and data collected during any actions and decisions they make to then analyze all the stored data and improve the outcome over time. This is the level where machine learning (ML) and deep learning (DL) really begin. Some of the examples of limited memory AI applications include self-driving vehicles and chat-bots.
In the past few decades, AI has influenced nearly all aspects of our lives. ML is a branch of AI aimed at training algorithms to perform various tasks such as making predictions, natural language processing (NLP), or recommending movies and products. The ML models leverage the computer’s ability to find hidden patterns in a large amount of data. The diversity and complexity of the problems solved by ML algorithms make it a promising tool to deal with challenges in the healthcare system.
How is AI used in healthcare?
In medicine, AI tools are used to search through medical data in order to uncover trends and insights that can help with test results interpretation and health outcomes improvement. Thanks to the recent advances in computer science and reduced computational time and costs, AI is quickly becoming an integral part of modern healthcare. ML as a field of AI can transform the patients’ care and the field of medicine. It helps healthcare providers and doctors to predict, diagnose, and treat health conditions with a higher standard of care. ML and DL algorithm have applications in different field of medicine such as radiology, pathology, dermatology, cardiology, and health systems. The consensus is that taking advantage of ML algorithms along with the best clinicians’ expertise will provide a standard of care that outperforms what either one can do alone.
ML in Medical Diagnosis
The application of ML can reduce the frequency of misdiagnosed incidents and thus improve public health and save patients’ lives while it is lowering the overall cost of care. As an example of the application of ML in medical diagnosis, we can refer to the classification algorithms to detect the existence of abnormalities as well as different types of disease based on different modalities of medical images as their inputs. As another group of diagnosis algorithm, we can refer to the segmentation models to determine the areas on the anatomical body map which are mostly affected by the disease.
ML in Predictive Medicine
The application of ML in the field of physiological monitoring has recently gained traction. The algorithm opens a window that urges the clinicians to look deeper into patients’ clinical signs. These algorithms can be used to detect any abnormalities in physiological signals, patients’ electric health records and their health status. Some examples of the application of these algorithms can be detecting arterial fibrillation using information from electrocardiograms as well as predicting hypotensive events in ICU and OR settings based on physiological signals.
ML in Medical Treatments
ML algorithms are also useful to recommend treatments more suited to individual patients based on their demographic information, their medical conditions and other randomized trial data to find the best possible treatment option. Fore example, a reinforcement learning algorithm shown to be more effective than human clinicians in recommending the use of vasopressors, medications or fluid interventions for sepsis patients.
AI in Medical Devices
AI has also an important role in devolving medical devices and addressing different aspects of robotic surgery. Further ML and DL can be useful in improving patients’ safety while using medical devices. As an example of this application, we can refer to a recent work on predicting specific absorption rate in MRI imaging which can help prevent safety issues.
Application of AI in radiology
Neuroscience relies on high resolution imaging to study the brain function and to determine causes of a disease. Ultra-high field (UHF) magnetic resonance imaging (MRI) has become the key research tool for the study of human brain. Recent FDA approval of some UHF 7T MRI systems created a pathway from pure research applications to clinical use of these systems. The advantages of UFH MRI for neuroimaging are clear and enormous, however a key limitation to the wider clinical use of this systems is the safety concern related to the nonuniform deposition of radiofrequency power associated with such UHFs. Specific absorption rate (SAR) is the quantitative measure of MRI safety and special effort must be taken to minimize this value. Current technology can only measure the overall average, or global, SAR. Local SAR variations are very difficult to predict due to anatomical and positional variations between patients. Many institutions use a conservative estimate of the peak local SAR, thereby limiting the imaging performance achievable by UHF MRI, in particular, the image resolution and/or scan time. The advent of AI and ML techniques allowed for creation of safety prediction tools that could be integrated in an MRI exam to facilitate safe generation of ultra-high resolution images .
- FDA clears first 7T magnetic resonance imaging device. https://www.fda.gov/news-events/press-announcements/fda-clears-first-7t-magnetic-resonance-imaging-device
- S. Gokyar, F. J. L. Robb, W. Kainz, A. Chaudhari and S. A. Winkler, “MRSaiFE: An AI-Based Approach Towards the Real-Time Prediction of Specific Absorption Rate,” in IEEE Access, vol. 9, pp. 140824-140834, 2021, doi: 10.1109/ACCESS.2021.3118290.