Artificial Intelligence in Medicine
Introduction
A patient might visit a computer before consulting a doctor in the standard medical practice of the future, which could arrive sooner than expected. It seems feasible that the days of misdiagnosis and treating disease symptoms instead of their underlying cause could soon be a thing of the past due to artificial intelligence (AI) developments. Think about how many decades of blood pressure data are available or how much space you'd have to clear off your laptop to accommodate a complete 3D model of an organ. The increasing amount of data generated in clinics and kept in electronic medical records via common testing and medical imaging has made it possible for additional uses of artificial intelligence and performance-data-driven medicine. Both doctors and researchers now approach clinical problem-solving in various ways and will continue to do so due to these applications.
Virtual and physical artificial intelligence (AI) are the two categories in medical technology. Applications like electronic health record systems and neural network-based therapies are examples of the virtual part. The physical section focuses on elderly care, surgeries with robotic assistants, and intelligent prostheses for people with disabilities.
While certain algorithms may compete with and occasionally exceed doctors in several tasks, they must still be fully incorporated into routine medical practice. Why? Before these algorithms significantly impact medicine and improve medical interventions' efficiency, various regulations must be resolved.
What makes an Algorithm Intelligent?
AI algorithms must learn how to execute their tasks, like how doctors are trained via years of medical school, completing assignments and practical examinations, obtaining grades, and picking up from mistakes. AI algorithms can generally carry out activities that require human intelligence, such as pattern and speech recognition, decision-making, and image analysis. However, when providing an image to an algorithm, humans must directly instruct the computer what to look for in the image. In summary, AI algorithms are good for automating difficult jobs and sometimes exceed humans in the duties they are trained to execute.
Computer systems frequently first receive structured data, so each data point has a label or annotation that an algorithm can recognize. An efficient AI algorithm is then built using this data. The algorithm's performance is evaluated for accuracy once exposed to sufficient sets of data points and their labels, exactly like students take exams. These "exams" of algorithms typically entail the input of test data for which the programmers already know the responses, thus enabling them to judge the algorithm's capacity to identify the right response. The algorithm might be updated, given more data, or implemented based on the testing outcomes to assist the person who created it in making judgments.
Numerous different algorithms are capable of learning from data. Most AI-based medical applications need data as an input, often either numerical (like blood pressure or heart rate) or image-based (like MRI scans or Images of Biopsy Tissue Samples). The algorithms then generate A probability or categorization using the acquired data. In this case, the actionable result might be the probability of developing an arterial clot given data such as blood pressure and heart rate, or it might be the classification of a tissue sample from an image as cancerous or not. To determine an algorithm's capability and worth in the clinic, medical applications compare its efficiency on a diagnostic task to that of a doctor.
Recent Applications of AI in Medicine Based on Algorithms
Many clinical challenges are suitable for AI applications due to improved computer power and the enormous quantities of data generated by healthcare systems. The two latest applications of accurate, clinically applicable algorithms are described here. By simplifying diagnosis, these applications can be advantageous to patients and doctors.
The first of these algorithms is one of the various examples of existing algorithms that outperform doctors in categorization tasks. The researchers at Seoul National University Hospital and College of Medicine developed DLAD (Deep Learning Based Automatic Detection) in the autumn of 2018 to analyze chest radiographs and detect abnormal cell development, such as possible cancers. When the results were compared to the detecting abilities of other doctors, the system beat 17 of the 18 doctors on the same set of images.
In the fall of 2018, researchers from Google AI Healthcare created LYNA (Lymph Node Assistant), the second of those algorithms. This learning system analyzed histology slides and dyed tissue samples to identify lymph node biopsies with metastatic breast cancer tumors. It is significant to note that, although not the first significant application of AI to histological analysis, our system accurately detected problematic regions in the given biopsy samples. LYNA was tested on two datasets and was 99% accurate in identifying samples as malignant or non-cancerous. Additionally, LYNA reduced typical slide review time in half when used with doctors' regular evaluation of stained tissue samples.
Other imaging-based algorithms have recently demonstrated a similar capacity to enhance physician accuracy. In the near term, clinicians can utilize these algorithms to help them verify their diagnosis and evaluate patient data more quickly while maintaining accuracy. Long-term, though, government-approved algorithms can operate independently in the clinic, allowing clinicians to focus on problems computers can't resolve. By emphasizing important elements of images that need more investigation, algorithms like LYNA and DLAD help doctors classify healthy and sick samples.
Limits of AI in Medicine
1. Needs Human Surveillance
Human surveillance remains necessary even though AI in medicine has advanced significantly. For example, surgical robots think logically rather than sympathetically, unlike human surgeons. Health professionals could catch up on crucial behavioral signals that can aid in diagnosing or avoiding medical problems. "AI has been present for some time and is still evolving. Yang says that healthcare professionals and tech experts interact more as this field develops. To be effectively utilized, AI needs human input and review.
The IT and medical sectors are collaborating more to advance AI as it grows. "Medical professionals have to finish years of education to practise their trades," Yang adds. Subject Matter Experts (SMEs) provide crucial knowledge to improve explainable AI (XAI), enrich the data, and give healthcare workers valuable insights.
2. May Overlook Social Variables
The needs of patients frequently go beyond their immediate medical conditions. When making recommendations for a patient, social, economic, and historical aspects may be relevant. As an illustration, an AI system can assign a patient to a certain care facility based on a particular diagnosis. However, this approach might need to consider the patient's economic constraints or other unique preferences. The use of AI systems also raises privacy concerns. When gathering and using data, companies like Amazon are unrestricted. Hospitals, on the other hand, might experience certain challenges when attempting to route data from, say, Apple mobile devices. These social and government constraints might make it tougher for AI to support medical treatments.
3. May Lead to UnEmployment
AI may help reduce costs and clinician stress but also eliminate some employment. If this aspect causes professionals who invest time and money in their healthcare education to be replaced, equity issues may occur.
According to 2018 World Economic Forum research, AI will lead to 58 million new jobs by 2022. However, the report states AI will displace or destroy 75 million jobs by 2025. Jobs that require repetitive tasks will become redundant as AI becomes more integrated across numerous sectors, which is the main cause of this loss of job customers.
Even if AI has the opportunity to advance many areas of healthcare and medicine, it's crucial to think about how using this technology would affect society.
4.InAccuracies are Still Possible
The diagnosis data that can be gained from millions of situations that have been classified is crucial to medical AI. An error could be possible in situations with limited information on specific illnesses, statistics, or environmental factors. This aspect becomes extremely important when a patient is given a specific medication.
Yang notes that there will always be some missing data, regardless of the system. Some information may need to be included in the case of prescriptions regarding specific populations and responses to medicines. This can make diagnosing and treating patients who fit into particular demographics difficult.
To fill in data gaps, AI keeps evolving and improving. It's crucial to remember that certain populations still need to be included in the current body of data in the field.
5. Susceptible to Security Risks
Since data networks are typically used in AI, security problems may arise with AI systems. To ensure that the technology is expandable as Offensive AI becomes increasingly common, greater cyber security will be required. Forrester Consulting estimates that 88% of security sector decision-makers are convinced that offensive AI represents a new threat.
Just as AI uses the information to make systems more effective and accurate, cyberattacks will employ AI to get smarter with each win and loss, making them difficult to identify and avoid. Stopping assaults will be considerably more difficult if dangerous threats can defeat security safeguards.
Benefits Of Artificial Intelligence in Healthcare
The beneficial effects of artificial intelligence in the medical field enable doctors to take advantage of ground-breaking opportunities to treat patients in more accurate, proactive, and effective ways.
1. Lower Overall Costs to Healthcare Providers
The medical industry is constantly pressured to reduce costs without compromising care quality. Using artificial intelligence (AI) is one-way healthcare professionals are overcoming this hurdle. AI can assist in automating repetitive processes, saving the staff's time for more significant ones.
Medical professionals may diagnose and treat patients faster with the help of AI. Hospitals can use AI to check X-rays for illness signals in this case. In turn, this can reduce the total cost for healthcare providers by reducing the number of unneeded tests and operations. AI can also design customized treatment programs for each patient. This ensures every patient receives the best care possible, improving health outcomes and lowering costs.
2. Artificial Intelligence to Assist in Surgery
Artificial intelligence has various benefits in surgery and healthcare, from assisting surgeons during surgeries to automating identifying a patient's anatomy. Various studies have demonstrated how useful AI can support surgeons during operations.
For example, with AI, surgeons can choose the best plan of action for each case by developing detailed 3D models of the anatomy of their patients. AI-assisted surgery has also been found to reduce recovery times and the number of complications. Additionally, AI can shorten the time required for surgical treatments while increasing their accuracy.
AI's position in surgery is expected to grow as its benefits for healthcare continue to emerge.
3. Improving Clinical Health Data Management
Creating and collecting more information is necessary as healthcare adopts digital health. Clinicians and other healthcare professionals already suffer from data overload based on numerous connected medical devices and healthcare systems. More raw data given to them will overload them and prove useless for improving service.
Implementing clever algorithms is one technique for dealing with the data overload of digital health apps. These machine learning algorithms are a more effective way to collect actionable data without depending on human intervention since they process the data and detect patterns that are not immediately visible to the human eye.
Devices that monitor glucose levels are an excellent example of how smart algorithms work. These gadgets only alert the user when their action is necessary, not the raw data of glucose levels, and they do so by identifying trends. Modern gadgets like Medtronic's Continuous Glucose Monitoring device (CGM) can automatically modify insulin doses following crucial glucose data.
This is only one of several cases where businesses create new digital health solutions for monitoring illnesses using similar principles and AI.
4. Processing Large Data Sets for Diagnosis
Processing an enormous amount of data, another area where AI is transforming the world of medical equipment is a benefit of AI in the healthcare industry.
For example, a recent NIH study on detecting metastatic breast cancers using AI discovered that the system was up to 99% accurate. In this study, artificial intelligence (AI) was employed to find micrometastases, cancer cells that have spread but tend to be very difficult for pathologists to spot.
According to another study, pathologists may overlook up to 60% of small tumors when diagnosing without the help of artificial intelligence. This is another case in which the use of AI in medical devices can save lives. For improved health results, early detection and suitable treatment are always preferred.
5. Improving Healthcare in Under-Resourced Areas
Certain countries struggle more than others because they lack access to professional training or resources for qualified clinicians. Undiscovered or incorrect diagnoses of health problems are more common in these nations.
One study found that applying high-resolution microendoscope pictures for identifying esophageal tumors had significant potential for usage in countries with limited resources or human capital.
In contrast, the Human Research Programme at NASA is creating a platform that employs machine learning to recognize a wide range of concerns essential to space flight. In a hostile environment like space, this new technology enables continuous tracking of crucial parameters, including bone density, intracranial pressure, and cardiovascular diseases.
Additionally, NASA is working on smart guidance systems that will allow astronauts with little knowledge to operate ultrasound equipment properly. Astronauts can move the probe to take the best images using "GPS-like" navigation that shows them where to position it. Any community with limited resources could use the same technology that NASA has created. If the technology turns out to be accurate, this might accelerate the accurate diagnosis of these populations.
6. Speeding Up Drug Development With Artificial Intelligence
Before a drug is successfully developed, the typical development process takes an extremely long time, and it frequently involves multiple "misses" before companies find a formula that works. The lengthy drug development process contributes to the current high price of pharmaceuticals.
By incorporating AI into developing drugs, researchers hope to uncover the most promising discoveries early in the R&D process, enabling them to save time and resources during the initial discovery phase.
An example of how AI is used in drug development comes from a partnership between GNS Healthcare and REFS (Reverse Engineering and Forward Simulation) to parse and analyze complex medical data. Using patient data, researchers can create new models that identify hidden factors in cancer.
The collaboration between GSK and Exscientia to find creative and selective small compounds for up to ten disease-related targets is another example of AI in action. The objective is also to improve the accuracy of the early stages of drug discovery. In reality, every stage of the pharmaceutical development cycle can be impacted by AI.
A clinical trial could serve as a last hope for people who have tried all other options to address their medical condition. Finding and being accepted into a relevant clinical study is one of the main difficulties for patients.
Unless their doctor or another person they know already knows of a trial for them, patients have to look up the government's databases of clinical trials on their own. Then, as shown in the process map from CB Insights below, they go through a thorough review process for criteria for inclusion and exclusion.
Current Applications of Artificial Intelligence in Medicine
1. Cardiology
1.1. Atrial Fibrillation
The early detection of atrial fibrillation was one of the first medical applications of artificial intelligence. The mobile app Kardia from AliveCor was given FDA approval in 2014 and allowed smartphone-based ECG monitoring along with atrial fibrillation identification. According to the current REHEARSE-AF study, remote ECG monitoring with a Kardia on mobile devices makes individuals more likely to recognize atrial fibrillation than standard care. Furthermore, Apple's Apple Watch 4 was given FDA approval, enabling rapid ECG acquisition, atrial fibrillation detection, and sharing with the medical practitioner of your preference through a smartphone. Numerous critiques of wearable and portable ECG technologies were addressed, highlighting their limitations, like the false positive rate caused by movement artifacts and obstructions to their adoption in elderly patients more inclined to suffer from atrial fibrillation.
1.2. Cardiovascular Risk
When applied to electronic medical data, AI is more accurate than conventional scales at estimating the risk of cardiovascular disorders like acute coronary syndrome and heart failure. However, recent in-depth evaluations have shown how the sample size utilized in the research report could influence the outcomes.
2. Pulmonary Medicine
According to reports, the analysis of pulmonary function tests is an appealing field for creating artificial intelligence applications in pulmonary medicine. According to the latest research, AI-based software improves the accuracy of interpretation and functions as a decision-support tool when evaluating the results of pulmonary function tests. The study was subjected to several critiques, one of which noted that the study's pulmonologists had substantially lower rates of accurate diagnosis than the country's average.
3. Endocrinology
Diabetes patients can use continuous glucose monitoring to view interstitial glucose measurements in real-time and get information on the rate and direction of change of their glucose levels in the blood. FDA approval has been granted for Medtronic's Guardian smartphone-connected glucose monitoring system. Through repeated measurement, the company collaborated with Watson (an IBM AI) for their Sugar. IQ solution in 2018 to help their customers better avoid hypoglycemic episodes. Continuous blood glucose monitoring could help patients improve their blood glucose management and lessen the stigma associated with hypoglycemic episodes. Despite this, a study conducted on patients' experiences using glucose monitoring found that, despite participants' statements to the contrary, they felt personally unsuccessful in maintaining blood sugar control.
4. Nephrology
Various clinical nephrology settings have used artificial intelligence. For example, in polycystic kidney disease, the patient's glomerular filtration rate drops, and there is a possibility of developing IgA nephropathy. It has been demonstrated that its use in these situations benefits them. However, the sample size needed for inference has recently been shown to be a research limitation.
5. Neurology
5.1. Epilepsy
Intelligent seizure detection devices can enhance seizure management through continuous ambulatory monitoring. In 2018, the FDA gave Empatica's wearable Embrace its approval. When used with electrodermal captors, it can detect generalized epilepsy seizures and indicate them to a mobile application, alerting close relatives and a trusted doctor and providing more information about the patient's location. Following a patient experience report, epileptic patients did not experience any adoption barriers for seizure detection wearables, unlike those for heart monitoring. They have shown a lot of interest in using wearable technology.
5.2. Gait, Posture, and Tremor Assessment
Patients with multiple sclerosis, Huntington's disease, and Parkinson's disease may be able to quantify their tremors, posture, and gait using wearable sensors.
6. Computational Diagnosis of Cancer in Histopathology
The FDA has given Paige.ai's AI-based algorithm breakthrough status for its capacity to precisely diagnose cancer through computational histopathology, freeing up pathologists' time to focus on critical slides.
7. Medical Imaging and Validation of AI-Based Technologies
Radiologists' and deep learning systems' abilities in imaging-based diagnosis were compared in a long-awaited meta-analysis. Although deep learning appears to be as effective as radiologists for diagnosis, the authors noted that 99% of studies did not have an accurate design. In addition, only 1% of the papers validated the results using strategies to diagnose medical imaging in different populations. These findings suggest that extensive testing is necessary to verify AI-based technology.
Future Directions and Challenges of AI in Medicine
1. Validation of AI-Based Technologies: Toward a Replication Crisis?
One of the biggest difficulties in applying AI to medicine in the coming years will be the clinical validation of recently created fundamental principles and technologies. Even though many investigations have already proven the value of AI with apparent potential based on promising outcomes, several well-known and often published restrictions on AI studies tend to make such validation more difficult. The three restrictions will be covered in the following, along with possible solutions.
Most studies comparing AI and doctors' efficacy have been found to have unreliable designs and require primary replication, or validation, of the algorithm produced in samples from sources other than those employed to train the algorithms. The open science era offers an approach to this problem because open data and open methodologies will become more popular as best practices in research. But for medical AI businesses that depend significantly on software development, transitioning to open research might be challenging.
Second, studies describing the application of AI in clinical practice are recognized to have constraints due to retrospective methods and sample sizes; such designs can involve selection and spectrum bias, which occurs when models are created to have the optimal fit for a specific data set (a phenomenon referred to as overfitting) but fail to duplicate the same results in other datasets. It is essential to continually reevaluate and calibrate the software to adjust for changes in patient demographics after adopting algorithms that are overfitting. There is also an increasing recognition that algorithms must be created and customized to handle larger communities while considering subgroups.
Third, only a few studies compare AI and clinicians utilizing similar data sets; in these situations, criticisms have highlighted a lower diagnostic rate of accuracy than anticipated in specialized doctors. Though strongly supported in the scientific literature, there are better ways to address the performance problem in medical expertise than opposing AI and doctors. Several studies are investigating the interaction between algorithms and clinicians, as humans and artificial intelligence work more effectively together than separately.
2. Ethical Implications of Ongoing Monitoring
Medical technology, expected to reach a market value of nearly $1 billion in 2019, is one of the century's most promising industries. Despite not being the primary goal of the customer profile (because health problems like atrial fibrillation are less likely to emerge), an increasing percentage of the revenue comes from the sale of medical supplies to younger people, such as heart monitoring devices. As a result of this phenomenon, the Internet of Things (IoT) has redefined what constitutes a healthy person as a combination of the quantified self and a variety of lifestyle parameters supplied by wearables (weight control, activity monitoring, etc.).
Additionally, many wearable companies have signed significant agreements with insurance firms or governments over the past several years to organize a wide-scale distribution of their products. These kinds of initiatives primarily seek to change people's lifestyles. The moral consequences of constant medical monitoring using medical devices over the Internet of Things are often discussed, even as Western countries continue moving towards health systems emphasizing the patient's self-sufficient responsibility for their health and well-being. For instance, continuous monitoring and confidentiality violations may increase stigma towards chronically ill or more impoverished people and penalize those who cannot accept new healthy lifestyle norms by limiting access to healthcare. However, little to no debate has been given to these significant and potential pitfalls in establishing health policy.
Although it has been around for more than 20 years, the problem of data ownership and protection is becoming increasingly important in the modern techno-political context. The consensus is moving towards patient ownership as it positively impacts patient engagement and could enhance information sharing if a data use agreement is developed between the patient and healthcare professionals. However, some works argue for sharing patient data ownership to improve personalized medicine approaches.
3. The Need to Educate Augmented Doctors
Many universities are developing new medical curricula, such as a doctoral engineering program, due to the requirement to educate upcoming medical leaders about the issues brought about by artificial intelligence (AI) in medicine. Such curricula emphasize physics and mathematics more than other subjects, including computational sciences, code, algorithmic thinking, and mechatronic engineering. These "augmented doctors" would depend on their clinical competence and digital knowledge to address contemporary health issues, develop digital strategies for healthcare organizations, manage the digital transition, and educate patients and peers.
These experts may act as a layer of protection for every procedure, including the application of artificial intelligence in healthcare, in addition to being a source of creativity and research that would benefit society and healthcare organizations. Besides basic medical education, continuous educational programs about digital medicine focused on graduated doctors are also required to make way for retraining in this growing profession. In many modern hospitals, these experts hold the Chief Medical Information Officer role.
4. The Promise of Ambient Clinical Intelligence: Avoiding Dehumanization by Technology
According to several studies, electronic health records are a huge administrative burden and cause burnout. Both recently graduated medical students and experienced doctors are increasingly experiencing this problem. Although artificial intelligence tools like natural language processing can already help clinicians offer complete medical records, new tools are still needed to solve the issue of the increased amount of time used for indirect patient care.
A sensitive, adaptable, and responsive digital environment that surrounds the doctor and the patient is known as ambient clinical intelligence (ACI). It can automatically fill out a patient's electronic health records and analyze the interview. The development of an ACI, which would be an essential application of AI in medicine and be desperately needed to address current problems with the medical workforce, is now the focus of several studies.
Fear of dehumanization of medicine is one of the main challenges facing doctors using intelligent medical technologies. The increasing administrative burden placed on doctors is the primary cause of this. The administrative burden issue will eventually be resolved, and improvements in technology like ACI and natural language processing will allow clinicians to give greater time to the patient.
5. Will Artificial Intelligence Replace Doctors
According to a recent literature discussion, Doctors won't likely be replaced by artificial intelligence; instead, smart medical technology exists to support doctors and help them better manage their patients' care. However, recent studies have shown that it often compares medical professionals and artificial intelligence solutions as if they competed. Translational clinical trials should be included in future research comparisons between physicians who use AI applications and those who do not. Artificial intelligence will be considered a medical complement after then.
Although a major rethink of medical education is required to equip future leaders with the skills needed, healthcare professionals are privileged to endorse digital transformation and be the primary agents of change.
Conclusion
Artificial intelligence in medical practice is a potential area of study and development that is expanding as quickly as other contemporary subjects like medical precision, genomics, and teleconsultation. The moral and financial difficulties posed by this aspect of medical progress must be tackled in health policies. Simultaneously, scientific studies should remain robust and transparent to develop fresh strategies for improving contemporary healthcare.