For healthcare organizations, artificial intelligence (AI) has tremendous potential for increasing the quality of care and enhancing operational efficiency. As this technology continues to mature, healthcare organizations will be able to tap into an expanding array of use cases—from streamlining administration and boosting patient engagement to accelerating drug development and improving diagnostics.
Where should your healthcare organization begin its AI journey? Investigating potential applications for AI is an essential first step. But your organization should also be aware of key challenges that you will need to address before moving forward.
AI is technology that uses computers or other machines to simulate human intelligence. Though forms of AI have been in existence for decades, the increasing availability of large data volumes and enormous compute power is speeding the advancement of AI for healthcare and a range of other fields.
Several types of AI technologies can be used for healthcare—and a few have already been implemented.
Machine learning (ML) is a branch of AI in which machines (computers) emulate the way humans learn. ML systems analyze data, make predictions, evaluate the accuracy of those predictions, and then refine predictions as new data is ingested. These systems are used in a variety of healthcare applications. In public health, for example, ML can be used to predict the location of possible disease outbreaks by analyzing a variety of factors.
Deep learning is a form of ML in which computers simulate the layers of neural networks that the human brain uses to learn. As data moves through the layers, the output of one layer is used as input for the next. Deep learning could be used to help analyze medical images and assist physicians with diagnostics.
Natural language processing is a technology that enables humans and computers to communicate through human language. Natural language processing has given rise to generative AI (GenAI), in which humans can prompt machines with questions or commands and then receive responses in human language. Healthcare organizations are already employing GenAI in chatbots that provide the first level of customer and patient care.
AI promises to have a transformative effect on healthcare. Healthcare providers could use AI to accelerate diagnostics and deliver personalized treatments that produce better outcomes. Payers could use AI to reduce administrative costs and offer better subscriber experiences. Meanwhile, pharmaceutical companies could employ AI to deliver new, more targeted drugs, faster and more efficiently.
Despite the possible benefits of AI, all healthcare organizations will need to be careful with AI implementation. For example, they will need to ensure that patient data remains private even as it is analyzed by AI models. And they will need to minimize any potential biases in AI algorithms. Addressing these and other challenges will be critical for maximizing the benefits of AI.
AI has the potential for helping clinicians make better medical decisions, faster than before. In fact, some basic AI systems for clinical decision support are already in place.
A rule-based expert system represents one of the simpler forms of AI: It analyzes data by applying rules created by human experts. This type of system can assist with diagnostics and treatment, using a list of symptoms, observed signs, and test results to present potential diagnoses and treatment recommendations.
A rule-based system can improve the speed, consistency, and quality of decisions. However, it can have difficulty with more atypical or novel patient presentations, since it might lack the rules for interpreting particular collections of symptoms or measurements.
ML is a more sophisticated form of AI than a rule-based expert system. An ML system continuously refines its models and improves its ability to predict future results as it ingests more data.
For clinicians, ML-based systems can help enhance the accuracy of predicting disease outcomes—and those predictions can help clinicians develop better, more personalized treatment plans. So, for example, clinicians could use ML-based systems to predict the prognosis for patients with cancer, heart disease, or other illnesses. With new insights, clinicians can better determine what treatments should be applied and when.
Natural language processing can dramatically accelerate the acquisition of data from unstructured, text-based records. In healthcare, natural language processing enables providers to extract information from clinical notes and other data within electronic health records (EHRs). Clinicians could use that extracted information to support clinical decision making for individual patients. And they could input information into analytics or AI systems to generate new insights that would apply to larger populations.
AI is showing great potential for assisting with the review and assessment of medical images, including X-rays, CT scans, ultrasounds, mammograms, MRIs, and more.
Because deep learning systems excel at pattern recognition, they are a good fit for analyzing medical images. These systems can help clinicians better detect anomalies, focusing on subtleties that might otherwise be difficult to spot, especially early in the course of a disease. Importantly, deep learning systems can also automate image analysis, enabling providers to analyze a large number of images rapidly.
Computer-aided detection—or computer-aided diagnosis—(CAD) can help clinicians improve the accuracy and consistency of image analysis. Though CAD systems have existed for several decades, the rise of AI technology is helping augment the capabilities of CAD systems.
The integration of AI technology can help CAD systems speed image analysis, provide prognoses, and generate insights for treatment decisions.
Radiologists often need to segment medical images into distinct regions and extract information about the features of objects in those regions. AI can help enhance the efficiency and precision of those processes. “Radiomics” is the approach of using analytics and algorithms to segment captured images into areas of interest and then extract quantifiable features, such as size, shape, texture, and more. With the incorporation of AI, radiomics systems can handle a much greater volume of images and find more subtle anomalies.
Pharmaceutical companies spend an enormous amount of time and money on drug discovery and development. AI can help those companies find promising targets faster, explore new compounds, and improve testing.
Pharma companies need to find molecules that will bind to a particular drug target. AI can help increase the speed and accuracy of virtual screening, which is the process of using computers to analyze modeled compounds instead of running costly lab experiments. With AI, companies could examine millions of molecules rapidly, greatly accelerating the virtual screening process while reducing costs.
As companies work to design and optimize drugs, they could use AI to explore an incredibly wide array of molecules—including molecules that do not exist in the natural world. Using AI in this way can dramatically accelerate drug discovery and ultimately produce better drugs.
AI can help pharmaceutical companies predict how newly discovered or designed drugs will affect humans. Companies can explore the efficacy of drugs in addressing particular diseases or conditions, and they can evaluate the safety of those drugs—all before conducting expensive, time-consuming clinical tests. They can eliminate some drugs early and focus their time and resources on the most promising candidates.
In healthcare and other fields, AI is already demonstrating its potential for providing responsive digital experiences for individual users. By implementing GenAI technology into chatbots and using it to send personalized messages, for example, healthcare organizations can improve patient engagement and adherence to treatments.
GenAI-based chatbots and virtual assistants can be used in multiple ways to support patients. For example, healthcare providers or payers can employ chatbots as a first line of responding to patient or subscriber questions, before engaging human customer service agents.
In addition, GenAI-based tools can enhance patient education. Organizations can provide ways for individuals to learn more about their diseases, review treatment options, understand medication side effects, and more. These tools deliver fast responses, conveniently, while minimizing costs.
AI tools can also be used to monitor patient health remotely and help track adherence to treatment regimens. When patients use wearable devices, providers can continuously collect and analyze vital data, looking for small changes that could signal emerging health issues or missed doses of medication. With AI-driven insights, providers can intervene quickly and early, before patients face emergency situations.
The same AI tools that provide remote patient monitoring can also generate reminders for patients to take their medicines or alert patients when they should contact physicians. Alerts can also be sent to providers, spurring them to reach out to patients or helping them further tailor treatment recommendations for specific individuals.
While the use of AI for diagnostics and drug discovery might garner more headlines, AI can also have a substantial, positive impact on healthcare administration and operations. By enhancing efficiency, AI can help providers and payers deliver better service while driving down costs.
AI can enhance robotic process automation (RPA)—technology used to automate manual and repetitive tasks. AI-driven RPA can help streamline billing and claim processing, enabling providers and payers to automate and accelerate complicated workflows. In addition, RPA can use AI to improve appointment scheduling, matching patients with the right providers, at the right locations, while avoiding scheduling conflicts. By using AI with RPA, organizations can provide timely, responsive service, reduce manual errors, and cut administrative costs.
Healthcare organizations can use AI to predict resource needs and optimize resource allocation. Resources can include staff, facilities, equipment, and supplies. For example, a hospital could use AI-driven insights about patient trends to optimize staff scheduling over the course of each day and across a typical year. Similarly, the hospital could use insights to adjust operating room schedules and ensure that on-premises pharmacies have the right amount of key medications stocked. Optimizing resources helps ensure that patients receive the care they need, when they need it, while allowing organizations to eliminate wasteful spending.
For decades, healthcare providers have used robots to assist with surgery, rehabilitation, and lab work. Using AI in conjunction with robotics can enhance the benefits that robotic systems deliver.
AI could augment the use of robotics for surgical procedures in several ways. For example, surgeons could use AI-assisted image enhancement to better visualize anatomical structures. AI models could help guide surgeons in real time to improve the precision of cancer-related surgeries. And AI could automate potentially simple tasks, such as tying a suture. By supporting decision making during surgeries and enabling surgeons to focus on the most critical tasks, AI-infused robotics could help produce better surgical results.
Robots are often used in rehabilitation to help patients improve movement in limbs that have impaired functioning. Incorporating AI into those robotic systems can help better tailor movements to each patient. Using AI with lower-limb exoskeletons designed to assist with walking, for example, can refine movement coordination for each patient, enabling them to more quickly improve their mobility.
Service robots can handle many routine tasks in healthcare settings, such as preparing patient rooms, disinfecting areas, or restocking supply cabinets. Integrating AI capabilities could enable robots to start work or customize processes without human intervention. Consequently, staff can focus more on directly providing care to patients.
There’s no doubt that AI has immense promise for improving decision making, increasing the efficiency of tasks, and ultimately helping to deliver better health outcomes. Still, healthcare organizations must be aware of the important legal and ethical risks of AI adoption as they move forward.
AI tools generate insights by processing large amounts of data. In the case of healthcare, that data can include protected health information (PHI). To maintain compliance with HIPAA and other regulations, healthcare organizations must ensure that AI in no way jeopardizes data privacy and security. When using AI for providing diagnostic support, analyzing images, or answering patient questions through a chatbot, for example, organizations will need to make sure they are de-identifying data before uploading it to an AI model.
Organizations that plan to adopt AI for healthcare use cases should also be aware of the possibility that bias has been inadvertently incorporated into models through the training process. The models used for diagnostic support, for example, might not include sufficient data from certain underrepresented communities. Consequently, the results that AI tools generate might not have the same accuracy or utility for diagnosing or treating people in those communities. Healthcare organizations must prioritize eliminating these biases as they select or train AI models.
In healthcare, patients have the right to know and understand their diagnoses, treatment plan, costs, and a variety of other information pertaining to their care. When AI is used for any aspect of patient care, patients should also have the right to understand how AI is being employed. Moreover, they should know whether and how their data will be used to train models. At a high level, to meet ethical standards, AI should not diminish an individual’s dignity, autonomy, or privacy.
What if something goes wrong when using an AI-based diagnostic tool or AI-based imaging analysis? A diagnostic tool could recommend a particular diagnosis or suggest a course of treatment that fails to work or harms the patient. Who is liable? The physician using the AI tool? The company that trained the model?
These questions are not yet resolved. Until they are, healthcare organizations should be cautious in their use of AI, especially for clinical diagnostics and decision making. Providers should remember that, more than likely, they will be held responsible for medical decision making whether or not they use AI tools.
As your healthcare organization begins its AI journey, you should first identify potential use cases and contemplate the legal and ethical implications of AI adoption. When it’s time to evaluate different AI models and tools, you will have a range of factors to consider—including how they will comply with regulatory frameworks, whether they have been subjected to real-world testing, and whether they provide sufficient visibility into how they work.
As interest in AI has grown among healthcare organizations, government agencies and industry organizations have begun to develop frameworks and guidelines for how to use AI. To some degree, AI solutions have been covered under rules for software-based medical devices. But some government and industry entities are also advocating for new rules that are more specific to AI.
For example, a regulatory council in the UK published suggestions for regulating AI-based medical devices to help ensure the effectiveness of AI, eliminate bias, and detect potential harms that could arise from AI. The EU, meanwhile, proposed the AI Act, which defines a legal framework for AI products and services from their development to their application. In the United States, the FDA implemented a proposed regulatory framework for AI-based medical devices that makes developers accountable for the performance of their AI systems.
AI can help healthcare organizations accelerate clinical trials and analyze real-world evidence (RWE)—data acquired through everyday clinical practice. But as organizations evaluate AI models, they should make sure these models themselves have been subjected to controlled clinical trials and real-world testing. Many AI models have been evaluated only using retrospective analysis. Until the models are rigorously evaluated, there will continue to be doubts about the accuracy, reliability, and effectiveness of AI-generated results.
AI models should offer the right balance between interpretability and explainability. Interpretability is the degree to which a model allows you to observe how it works and how decisions are made. Explainability is a measure of how easily a model allows you to generally explain how the model produces results, even without knowing the details of how the model works.
Given the concerns about the accuracy of AI-generated insights and the need to ensure regulatory compliance, interpretability is often critical for healthcare organizations. They want to clearly observe, understand, and demonstrate how a model is arriving at its conclusions. Still, the price for that transparency can be performance: Organizations might be able to deliver results faster if they have less visibility into how those results are produced.
As AI continues to mature, many healthcare organizations will be increasingly eager to implement AI to enhance operational efficiency and improve healthcare outcomes.
Want to learn more? Read the free Guide.