The transformative power of artificial intelligence(AI) is reshaping healthcare as we know it.
The industry's surge in data output has led to AI integration, turning the data into actionable insights. Already, AI is used in numerous aspects of patient care, healthcare systems, and pharmaceutical organizations.
The global AI market in healthcare is set to reach $20.9 billion in 2024. By 2029, it’s set to hit $148.4 billion, growing at a CAGR of 48.1% during the forecast period [1]. Factors like the increase in data volume, computing power, and the increasing adoption of digital technology in the healthcare industry will drive this exponential growth.
There are numerous potential use cases for AI in healthcare, ranging from early cancer detection to improved diagnosis accuracy. By leveraging AI and associated technologies, the healthcare industry is set to experience a leap in care delivery models. However, amidst the potential benefits, there are challenges and setbacks, including ethical concerns.
Our article explores healthcare applications benefiting from AI, the challenges, and the future of AI in this critical industry.
Early detection is key to improving disease management, helping to reduce mortality, costs and curb the spread of infectious diseases.
Misdiagnosis or failure to detect disease can be fatal. In the US alone, a study found that 400,000 patients suffer harm due to medical errors. Another study discovered that more than 200,000 patients die due to preventable medical errors [2].
AI has been a game-changer in diagnostics, especially in medical image analysis. Automating this process results in a significant improvement in the accuracy of disease detection. ML and DL algorithms like support vector machine(SVM) and convolutional neural network(CNN) are building models that identify and categorize anomalies in medical images like CT and MRI scan [3].
AI systems are capable of analyzing vast amounts of data quickly and accurately. Through pattern detection algorithms, an AI system can learn anomalies from data and images indicative of diseases. For instance, DL models based on CNN now detect complex details in medical images that may be undetectable to the human eye, helping identify diseases like rectal cancer [4].
AI is accelerating the analysis of medical images compared to traditional methods. These systems help reduce the impact of fatigue or oversight while still offering accuracy and efficiency. Using these tools, medics can spot minor anomalies, helping to quickly identify rare blood conditions and cancer [5].
Additionally, using multimodal analysis, an AI system can utilize multiple forms of data, including medical history, lab results, and imaging, to obtain detailed results. This process can be time-consuming and less comprehensive for human doctors.
AI-powered diagnostics in radiology: Such systems automatically analyze medical images, helping radiologists understand CT scans and X-ray images. The models are trained on vast amounts of data and use DL techniques to detect patterns indicative of diseases like cancer and organ abnormalities. For instance, an AI diagnostic tool showed an accuracy rate of 94% in detecting lung nodules [6].
AI-driven diagnostic tools: IBM Watson neural network software and Medtymatch are examples of AI engines powering diagnostic tools. For instance, the Medymatch engine utilizes deep vision and cognitive analytics to identify the rare anomalies invisible to the human eye [7].
AI-based skin condition diagnosis: Skin examination is the primary method of diagnosing skin conditions. AI assistive tools can help physicians diagnose these conditions, considering more than 2 billion people globally suffer from these issues [8].
For instance, using computer vision and image search capabilities, Google is building a tool to understand human skin, nail, and hair conditions. The tool can help detect 90% of the commonly searched conditions [9].
Personalized medicine continues to revolutionize healthcare by tailoring treatment plans to individuals. These plans leverage an individual’s genetic information, lifestyle data, clinical makers, and environment to develop targeted interventions likely to improve outcomes.
The integration of AI into personalized medicine has unlocked new possibilities. It can uncover hidden insights and identify correlations, helping medics accurately predict individual treatment responses.
Additionally, by analyzing the factors that predispose individuals to diseases, AI systems can predict people likely to get certain diseases, allowing a shift towards patient-centric approaches.
Precision medicines include promising targets and accurate therapies for patients. The impact is so effective, even prolonging and improving the lives of patients with fatal illnesses such as spinal muscular atrophy [10]. This strategy recognizes that a fit-all approach may not yield the best approach for everyone.
ML algorithms analyze large datasets of genetic sequences to uncover complex relationships in genomics. For instance, AI helps identify patterns, mutations, and genetic variations that may be linked to certain diseases and subsequent treatment responses.
Using multimodal data from previous patients treated for disease allows AI-based systems to identify future patients who can benefit from specific therapies.
Many models using algorithms like SVM have been developed and tested to check the effectiveness of treatment using genome data [11]. For instance, evaluations have been done for anticancer drug sensitivity, which could help avoid treating patients likely to be unresponsive to the treatment.
AI-based genome analysis has also been used for disease prognostication, identifying genetic markers associated with specific conditions or treatments. Another area is pharmacogenomics, where AI models can predict the customization of medication dosages [12].
The World Health Organization(WHO) estimates a global shortage of 4.3 million medical practitioners [13]. Using AI, healthcare organizations can improve service provision by augmenting processes like customer services and administrative duties.
Medics can concentrate on less repetitive tasks by automating processes like patient scheduling, issuing bills, and filling out medical records. AI virtual assistants and chatbots can increase patient engagement to increase customer satisfaction.
According to a study by Syneos Health Communications, 64% of patients surveyed said they would be comfortable using virtual assistants [14]. Chatbots offer instant responses and round-the-clock availability. This can be crucial during emergencies or when patients need quick responses about their condition.
Moreover, virtual assistants can reduce the administrative work burden in healthcare facilities. For instance, the assistants can answer questions about medication or forward reports to doctors and surgeons.
Advancements in digital medical devices and electronic health records are helping us overcome traditional challenges in providing care, such as geographical barriers. The data can help in continuous monitoring, helping reduce the frequency of hospital visits.
Integrated wearable and smart gadgets track patients' vitals daily and store data in cloud servers. AI can analyze this data to identify abnormalities or critical health trends. With such tools, caregivers can intervene quickly, helping reduce emergencies and hospital readmissions.
Additionally, AI-driven cognitive behavioral therapy(CBT) platforms make mental health services more personalized and accessible. They are designed to help users tackle challenges such as panic and anxiety. The AI creates a safe ecosystem for capturing user issues safely, without prejudice, assisting the patients in working through their problems using CBR strategies [15].
Other examples include platforms like IBM Watson, which collects patient data, analyzes it, and provides personalized care plans.
Poor medication adherence contributes to 125,000 deaths annually in the US alone, costing the healthcare system about $300 billion [16].
IoT gadgets and AI can help combat this significant challenge. Integrating AI into remote patient monitoring tackles these challenges by providing smart medication reminders, personalized education, and gamification to motivate patients to stick to treatment plans.
Some of the elements of an AI-driven patient adherence system include:
Chatbots: Using natural language processing(NLP), chatbots can engage with patients in conversations. It can answer questions about medications and respond to any concerns.
Behavioral analysis: ML algorithms can analyze patient behavioral patterns to determine factors contributing to non-adherence. These tools can offer healthcare workers insight into when patients will likely skip their medications and personalize strategies to overcome this.
Personalized reminders: AI-powered apps can provide context-aware and personalized reminders to prompt patients to take their medicine, tailored to their preferences.
Dosage error reduction: Errors in self-administering medicine using devices like insulin pens and inhalers have led to non-adherence and even hospitalization. AI-based unobtrusive frameworks can help identify cases of improper drug administration at home [17].
Current drug discovery methods are expensive and slow, and they’re generally a hit-and-miss approach. Drug development can take up to 14 years, and costs can eclipse $1 billion for successful drug candidates [18]. AI can help accelerate different aspects of this process to make it cheaper and faster.
The explosion of data from digitizing medical records, drug research, and precision medicine has made data available for study by AI tools. Here are some of the processes AI is reshaping in the drug discovery pipeline:
Target-based discovery: involves identifying novel protein molecules that could alter the disease state if modulated. Graph neural networks (GNN) and tree-based ML algorithms like decision trees can analyze proteomic and genomic data to identify potential targets [19].
Molecular simulations: Traditional molecular simulation methods are expensive and time-consuming. The use of AI accelerates the testing of candidate drug compounds [20].
Drug properties prediction: AI can forecast key physiochemical of potential drug compounds, including their solubility, toxicity, and bioavailability. Researchers can focus on the compounds with a high probability of success, reducing the time needed to identify promising compounds early. ML and DL neural networks, like generative adversarial networks(GANs), can be trained on databases of known compounds to detect patterns and correlations between the chemical structure of compounds and their physiochemical properties [21].
De novo drug design: DL methodologies are replacing the traditional method of de novo drug design, which involves screening large libraries of candidate modules. AI offers various ways to design new compounds using algorithms like CNN, GANs, and recurrent neural networks(RNN) [22].
Drug efficacy is crucial because negative drug responses result in drug failure, which can lead to drug withdrawal and side effects.
AI can predict drug efficacy by analyzing large datasets of molecular structures and biological targets. Algorithms like generative autoencoders can leverage patterns from existing drugs to explore new chemical spaces, helping to accelerate the discovery of pharmaceuticals with improved efficacy profiles [23].
Identifying side effects of new drugs through solid clinical trials is often expensive, slow, and unfavorable for large-scale tests. Predicting drug-related side effects early enough leads to efficient time and resource use.
Leveraging outcomes from previous clinical trials and drug compounds, predictive models can be used to infer side effects.
Drug formulation is an intensive process that can leverage AI’s ability to analyze vast datasets and identify patterns and correlations. Techniques like artificial neural networks(ANN), fuzzy logic, and genetic algorithms(GA) can help forecast the attributes of a drug candidate. By predicting the solubility, bioavailability, and stability, attributes critical in determining the potency of a drug formulation, researchers can find candidates likely to succeed during clinical trials [24].
AI-powered quality control systems, such as meta classifiers and tablet classifiers, are crucial in ensuring the final product in drug manufacturing is acceptable [25]. ML algorithms can analyze various attributes of the tablets during manufacturing to identify potential defects and inconsistencies.
Clinical trials are critical in ensuring the drugs developed are safe and effective. The main challenges associated with clinical trials include data-related limitations, manual efforts, and patient monitoring. Additionally, the data generated requires extensive manual manipulation. Other clinical trials require extensive rework, repetition, and data transcription.
In an interview, Dr. Khair ElZarrad, the director of the FDA's Office of Medical Policy, admits that AI has the potential to influence clinical trial design. Dr. ElZarrad, in charge of drug evaluation and research, explains that AI can leverage multiple forms of clinical trials, including decentralized strategies, using data collected from digital health technology(DHT) devices [26].
The DHT devices can gather data remotely and help monitor patients’ vitals in real-time.
Some of the other ways AI can impact clinical trials include [27]:
Since developing new drugs is expensive and time-consuming, repurposing drugs that can target multiple sites is easier. AI can reduce the efforts required to identify new use cases for existing drugs. By analyzing datasets of diseases and their targets, researchers can uncover correlations between existing drugs, potential treatments, and disease targets.
This is especially useful for diseases that don’t receive adequate funding for drug development. AI can help accelerate drug repurposing in light of the growing threats posed by drug-resistant infectious diseases. For instance, researchers were able to repurpose Baricitinib, a drug developed for Rheumatoid Arthritis, to fight COVID-19. AI was able to predict the drug's interaction and potential side effects [28].
The global market for robot-assisted surgical devices is growing at a CAGR of 17.1% and is expected to reach $34.1 billion by 2031 [29]. Robot-assisted surgical devices allow surgeons to use technology that uses surgical instruments during minimally invasive procedures. Emergent robotic solutions using AI, cloud connectivity, and augmented reality have led to these devices handling more complex procedures with better patient outcomes.
Such robots are designed to help surgeons with the manipulation and positioning of surgical instruments during operations.
AI has proved effective in guided laparoscopic operations, where the AI robot provides video feedback to surgeons highlighting things like abnormal tissue size.
Moreover, ML techniques analyze vast datasets of previous surgical operations to uncover insights and practices critical for operations. The robots learn by demonstration, accumulating enough information to conduct specific tasks. A task can be divided into smaller tasks, allowing the AI robots to identify, model, and conduct subtasks sequentially.
In an interview, Dr. Christopher Tignanelli, a surgeon and the director of the Program for Clinical AI at the University of Minnesota, explained that AI will provide decision support by analyzing surgeries as they are being performed [30].
Reinforcement learning(RL) is employed in surgical robots to improve them further, enabling them to tackle other complex subtasks like tube insertion and soft tissue manipulation. Compared to other analytic models, RL algorithms are effective in teaching effectiveness and accuracy [31].
These intelligent robots have helped surgeons perform telerobotic operations remotely. These systems use high-resolution cameras to provide a detailed close-up view of the surgical site, allowing surgeons to observe every detail.
The AI-driven system helps translate a surgeon's exact movements into high-precision actions by the robotic arm on the patient [32].
AI surgical assistants are also being used to help surgeons avoid mistakes by absorbing shakings during the operation. The system decreases the possibility of injuries through correct positioning, which helps eliminate the chances of wrong cuts [33].
AI in training healthcare practitioners offers numerous possibilities, including simulating complex scenarios and adapting the training to individuals.
AI systems can analyze students' learning patterns to tailor an appropriate learning path based on their attributes. Using recommender models, the system can actively suggest to the learners the areas they need to focus on. Custom tailoring education can increase learning effectiveness, helping students become more engaged.
Additionally, predictive models can forecast future learning outcomes using previous performance records and current learning progress. This can help educators provide appropriate support and intervene early for students who may be struggling
Learners can enhance their clinical skills through AI-driven simulations. By combining AI with virtual and augmented reality platforms, students can have immersive experiences that mimic the complexities of dealing with actual patients. Learners can reinforce their theoretical knowledge by practicing real-life scenarios like emergencies and surgical techniques.
AI simulation can create more complex and rare scenarios, helping learners broaden their experience.
Here are some medical domains in which AI is already in use:
AI in cancer treatment: AI holds immense potential in cancer research, diagnosis, treatment, and patient care. AI excels in analysis and segmentation tasks, helping reveal hidden insights in medical images and blood profiles, which form the basis of screening and evaluations.
AI in mental health treatment: Mental health illness is quickly becoming a global crisis. AI’s natural language capabilities are excelling in this space, with virtual therapists and chatbots helping many rapidly access help.
AI in infectious disease control and treatment: Disease outbreaks are challenging to control because of their uncertainty and complexity. However, AI has found multiple applications in infectious disease prediction and control, helping with decision support, drug discovery, and research.
The integration of AI in healthcare continues to enhance multiple processes in the industry. However, adopting AI in the industry has resulted in challenges that must be tackled.
Data privacy and security concerns: As AI increasingly integrates into healthcare processes and protocols, the volume and sensitivity of data being captured and analyzed raise security concerns.
For instance, in 2019, Google partnered with the hospital chain Ascension, which gave the tech company access to millions of patient records to train AI models. This was done without the patient's consent, which raised privacy concerns [34].
Hospitals and medical research institutes using AI systems must institute robust cyber security to prevent breaches. Other potential solutions include federated learning, where decentralized devices hold data locally and implementing techniques that make it hard to extract individual data.
Ethical concerns: AI models trained on data with biases like race, gender and socioeconomic status can perpetuate bias. For instance, in 2019, a study revealed bias in a widely used AI algorithm. It significantly underestimated the medical needs of black patients by over 50% compared to white patients with similar health conditions [35].
AI researchers must use diverse and representative data to alleviate bias and ensure it does not discriminate against specific demographics. Other possible solutions include using bias detection tools and ensuring AI decisions are explainable.
Regulatory compliance and trust in AI-driven healthcare: The US and the EU already have different regulations to govern AI and AI-based medical equipment use. However, the rapid advancements in AI are outpacing the laws. Issues of contention include the lack of clarity in determining liability between entities collaborating to develop, implement, and use AI systems.
Another critical issue is building trust in the use of AI in healthcare. To spur adoption, researchers must develop robust systems with failsafe mechanisms. Other considerations include ensuring explainability, keeping humans in the loop, and monitoring continuously.
It’s undisputed that AI holds great potential in the healthcare sector. However, there is uncertainty and excitement about its potential.
AI is expected to be even more integrated into administrative workflows as models and developers push boundaries. For instance, Mt Sinai uses an AI tool to code pathology cases autonomously [36].
Other possibilities include using deep generative foundation models to design novel molecules for drug discovery.
Virtual screening of billions of compounds for potential therapies will speed up drug discovery [37]. For instance, Atomwise’s AtomNet AI platform has shown the ability to undertake comprehensive high-throughput screening consistently, helping to identify novel chemical structures for drug development even faster.
The impact of AI in healthcare is undeniable. The technology improves patient outcomes, pushes medical boundaries, and streamlines medical processes. Leading use cases include drug discovery, AI-assisted surgeries, and disease diagnosis.
The use of AI is helping researchers accelerate the screening of compounds for drug discovery, analyze patient data, and uncover insights from medical images. Medical practitioners like radiologists can leverage AI for decision support and help diagnose diseases.
Despite all these use cases and benefits, ethical and regulatory concerns persist, such as algorithmic bias, which discriminates against certain demographics. Going forward, there’s a need for robust standards, security, and regulations governing the use of AI in healthcare.
Written By Baptiste Aelbrecht & Victoria Sihabouth