- AI
- A
Applications of AI in Healthcare
Recently, there have been many changes in the field of AI, and the situation is changing every day. AI has begun to spread in almost all industries, including healthcare and life extension.
Healthcare is a special area because it is highly regulated and can have a direct impact on human lives. Therefore, the implementation of AI here is not as fast as in other areas. Even in the USA, which is the main center for the development of such technologies, regulators such as the FDA (Food and Drug Administration) often delay the approval of new approaches.
At the moment, AI in healthcare mainly helps people with routine tasks and is rarely used for diagnosis or decision-making.
Nevertheless, there are noticeable attempts to change this situation. For example, Hyppocartic.ai is trying to solve the problem of nurse shortages and get approval in the near future.
The FDA plays a key role in approving new treatments, including those that use artificial intelligence and machine learning, and is largely a role model for all other countries, as it faces new devices earlier than others. The role of the FDA is to ensure the safety and effectiveness of medical devices, treatments, and therapeutic methods for their intended use. The regulatory process for AI-based treatments generally follows the same principles as for traditional medical devices, but takes into account the unique features of AI technologies.
Here is what the regulator usually requires for approval
1. Device Classification
FDA classifies medical devices into three classes (I, II, and III) based on the level of risk to patients:
Class I: low-risk devices, often with minimal regulation (e.g., bandages).
Class II: moderate-risk devices, usually subject to more stringent regulations (e.g., diagnostic tools, some AI systems).
Class III: high-risk devices requiring the most stringent control, often including clinical trials (e.g., AI devices for intensive care or surgical interventions).
AI-based medical technologies typically fall into Class II or Class III depending on the level of risk. If an AI-based solution impacts clinical decision support or patient care and does not rely solely on human intervention, it may be classified as a medical device. This status affects compliance requirements, safety standards, and the speed of updates or improvements.
2. Pre-submission
510(k) submission (premarket notification): for most Class II devices, this is the most common regulatory pathway. The manufacturer must prove that the new device is substantially equivalent to a legally marketed device (predicate device). For AI, equivalence must be demonstrated in terms of safety and effectiveness.
De Novo classification: if a predicate device is not available for a moderate-risk product, a De Novo application can be submitted to classify the new device, including AI-based technologies. This is suitable for new but moderately risky devices.
Premarket Approval (PMA): for Class III (high-risk) devices, the PMA process requires significant clinical evidence of safety and effectiveness. This process is very stringent, and AI technologies used in critical applications (e.g., diagnostics or life-saving interventions) may fall into this category.
3. Clinical Studies and Trials
For AI-based treatments that pose higher risks, the FDA often requires clinical trials to confirm the safety and effectiveness of the technologies. Clinical studies should include:
- Robust data on how the AI/ML model performs in various conditions.
- Analysis of the model's accuracy, sensitivity, and specificity compared to existing methods or tools.
- Verification of the model's ability to generalize to different populations and conditions.
4. Good Machine Learning Practices (GMLP)
AI/ML-based technologies must follow GMLP principles, which include data quality, transparency, and validation. The FDA has begun developing specific guidelines to ensure that AI models are developed and tested on appropriate datasets, reduce bias, and ensure patient safety.
5. Post-market Surveillance
The FDA continues to monitor AI devices even after approval. AI systems that continue to learn or adapt post-deployment (adaptive AI) may require ongoing review to ensure that updates do not compromise safety or effectiveness. This is part of the FDA's evolving system for "Software as a Medical Device" (SaMD).
For most startups developing healthcare systems, this process is very expensive and usually takes a lot of time. And AI systems add their own problems, especially in three areas:
Explainability: AI systems must be transparent enough for doctors to understand how they make decisions.
Bias and fairness: The FDA carefully analyzes the data used to train AI models to ensure they are not biased towards certain population groups 🤔.
Adaptability: AI technologies that continue to learn (adaptive AI) are under special scrutiny, and developers must prove that learning processes do not compromise patient safety.
Compliance with data protection regulations: for example, HIPAA, GDPR, and so on. Most widely used tools, such as Open AI, however, are not inherently compliant with these regulations, and it is currently unclear how to create a fully compliant AI system at the model level, even if it is aware of compliance rules — they can be violated through skillful prompt engineering.
Therefore, as already mentioned, most startups focus on things that do not require approval and have minimal risk.
In which areas can AI have the greatest impact?
Accelerating drug discovery: AI can significantly reduce drug development timelines, saving resources and ensuring faster delivery of vital medications to patients. Good examples are AlphaFold and AlphaProteo.
Advancement in medical imaging: creating synthetic medical images for training AI models, improving the quality of real images, and increasing the accuracy of automatic anomaly detection.
Revolution in personalized medicine: AI can analyze extensive patient data to tailor treatments and predict individual responses to therapy, pushing the boundaries of personalized care. AI can analyze genomic data to identify genetic mutations and variations associated with diseases. Based on the patient's genetic profile, medical professionals can develop targeted therapies that are more effective and have fewer side effects. For example, AI algorithms are used to interpret genomic data in cancer patients, helping oncologists choose the most appropriate chemotherapy or immunotherapy based on the unique genetic characteristics of the patient's tumor.
Improvement in medical training: AI-based simulations of diseases and medical procedures provide medical professionals with a safer and more effective platform to hone their skills without the need for real clinical cases.
Optimization of telemedicine and virtual assistance: AI-based chatbots can answer basic medical questions and direct patients to the right specialist, reducing the burden on healthcare workers and expanding access to timely medical consultation.
Preventive medicine: AI-powered platforms analyze user data (e.g., data from wearable devices or mobile apps) and provide personalized health, nutrition, or physical activity tracking recommendations.
Mental health and well-being solutions: Tools that help users track mental health (e.g., mood monitoring, stress management) or offer personalized mental well-being programs.
AI for administrative tasks and operations: AI systems that optimize the work of medical institutions (e.g., reduce patient waiting times, improve appointment scheduling, increase supply chain logistics efficiency).
Support for doctors in routine tasks: AI allows healthcare workers to focus on patients by automating record-keeping, meeting preparation, and post-meeting research. This is especially important as healthcare is largely built on human interaction, and we are now losing part of it by focusing on data collection and protocol compliance. AI can help return attention to patients by freeing doctors from routine tasks.
It is also important to note that AI is multimodal (video, images, text, sound, including chemistry or, as in the case of AlphaFold, protein folding, etc.), which creates many synergies if disparate data held by medical institutions are combined. AI can connect the dots that a human cannot see due to the scale of the data. Thus, AI can work at the intersection of different fields.
From an integration perspective, in addition to regulatory hurdles, there are natural forces that slow progress. Existing workflows in healthcare systems are very different from technical systems (e.g., cloud computing systems). They are very difficult to change, and it is often easier to integrate a less advanced or even outdated solution if it fits well into the system than the most advanced solution.
This is an aspect that technical specialists often do not understand. Technologies change quickly, and people change very slowly. Often the speed of adoption is limited by the level of trust, so AI-based solutions are implemented at the speed at which trust is built.
Validation and measurement of success
Generative AI and predictive models require special approaches to testing compared to structured logical systems that predict exact results. In the case of generative AI, the same input data can lead to similar but still different results in different runs. Therefore, implementing a reliable testing plan is extremely important to ensure AI accuracy. Effective testing strategies should include setting confidence thresholds (asking AI to either provide contextual evidence or refrain from answering), incorporating human feedback into the process, using traditional models for cross-checking (e.g., NLP for verifying contextual data with source materials), and analyzing the error matrix to assess the proportion of correct answers relative to the number of attempts. These strategies help improve model accuracy and overall performance.
Examples of real-world implementations:
Moscow city project on AI diagnostics:
Moscow is actively developing a project to implement artificial intelligence for diagnosing medical images such as X-rays, CT scans, and MRIs. As part of this project, AI helps doctors detect diseases faster and more accurately, including COVID-19, pneumonia, lung cancer, and other pathologies. The AI system analyzes examination results and transmits them to doctors for final conclusions.
Sechenov Institute:
Actively applies AI technologies for the diagnosis and analysis of medical data. In particular, AI is used for the analysis of medical images in oncology, which allows for more accurate and early detection of cancerous tumors.
Mayo Clinic, USA:
Mayo Clinic is at the forefront of integrating AI into healthcare, especially in radiology and cardiology. AI algorithms have been developed to assist in the interpretation of medical images, such as detecting early signs of heart disease from echocardiograms or analyzing MRI and CT scans to detect tumors. For example, the clinic collaborates with companies like Google and Nference to create machine learning models that can predict outcomes and personalize treatment for patients with heart disease, cancer, and liver diseases.
Zebra Medical Vision, Israel:
Zebra Medical Vision is a company specializing in AI-based medical imaging analytics, providing AI tools for radiological image analysis to medical institutions worldwide. Their AI models are used to detect a wide range of diseases, including cardiovascular diseases, lung diseases, liver diseases, and bone fractures. The system is implemented in several hospitals, helping radiologists quickly and accurately diagnose diseases by analyzing a large volume of images in significantly less time than manual analysis would require.
John Radcliffe Hospital, UK:
John Radcliffe Hospital in Oxford collaborates with Oxford University and Google DeepMind to use AI in detecting acute kidney injury (AKI). The system analyzes blood test results and other patient data to predict the onset of AKI 48 hours before it manifests. Early detection of AKI can significantly reduce patient mortality rates as it allows doctors to intervene more quickly and prevent further kidney damage.
Chang Gung Memorial Hospital, Taiwan:
Chang Gung Memorial Hospital has implemented AI to improve personalized medicine in the treatment of liver diseases, which are particularly prevalent in Taiwan. The hospital's AI system analyzes genetic information and medical history to predict which patients are most likely to benefit from certain treatments. This AI-based approach helps doctors make more accurate treatment decisions, leading to improved outcomes for patients with liver cancer and chronic liver diseases.
Apollo Hospitals, India:
Apollo Hospitals collaborated with Microsoft to develop an AI-based system designed to predict cardiovascular diseases. The system analyzes patient data, including medical history and lifestyle, to predict the risk of heart disease. This AI-based tool is especially important in India, where heart disease is one of the leading causes of death. The system helps healthcare workers identify high-risk individuals at early stages, allowing for preventive measures to be taken and reducing the likelihood of cardiovascular incidents.
Karolinska University Hospital, Sweden:
Karolinska University Hospital has implemented AI to optimize the process of diagnosing and treating brain tumors. Using AI-based imaging tools, the hospital's medical teams can detect brain tumors faster and plan treatment strategies. This has led to more accurate and timely interventions, improving patient outcomes. Additionally, Karolinska Hospital uses AI in drug discovery research, accelerating the development of new therapies for a wide range of diseases.
Mount Sinai Health System, USA:
Mount Sinai has integrated AI into its clinical workflows, especially in the areas of imaging and disease prediction. For example, Mount Sinai uses deep learning algorithms to analyze chest X-rays and predict the likelihood of various diseases, including pneumonia and COVID-19. The system provides fast and accurate diagnoses, helping doctors make more informed decisions. Additionally, Mount Sinai has used AI to predict outcomes in patients with conditions such as sepsis and heart failure, allowing clinicians to intervene in a timely manner and improve patient care.
Beth Israel Deaconess Medical Center (BIDMC), USA:
BIDMC has integrated AI algorithms into its clinical workflows, especially in intensive care settings, to predict patient deterioration. These predictive models use data from various monitoring devices and electronic medical records to identify early signs of potential complications, allowing medical teams to intervene in a timely manner. The use of machine learning and predictive analytics at BIDMC has contributed to improved patient outcomes by reducing mortality rates and shortening hospital stays.
Moorfields Eye Hospital, UK:
In collaboration with Google DeepMind, Moorfields Eye Hospital has implemented AI to analyze retinal scans. This AI system, trained on thousands of retinal images, is capable of detecting more than 50 different eye diseases with accuracy comparable to leading ophthalmologists. By significantly speeding up the diagnostic process, this technology allows for the early detection of eye diseases and the initiation of treatment sooner, helping to prevent vision loss in many patients.
Memorial Sloan Kettering Cancer Center (MSKCC), USA:
Memorial Sloan Kettering Cancer Center (MSKCC) collaborates with IBM Watson for Oncology to support personalized cancer treatment decisions. Watson for Oncology analyzes patient data and matches it with the latest medical research to provide evidence-based treatment recommendations. However, despite initial promise, the system faced challenges in clinical practice, and its effectiveness was questioned. Although Watson for Oncology advanced the use of AI in oncology, MSKCC subsequently reduced its reliance on the system due to concerns about the accuracy and practical usefulness of its recommendations.
Conclusion
AI is a very promising tool in healthcare that will significantly change the industry in the future. Currently, AI applications are limited and still in the research stage, so we have a long way to go.
Wishing everyone well and a positive mood!
Write comment