By almost 18 times – that’s how much the size of the artificial intelligence (AI) healthcare market is expected to increase by 2028 according to a report by Grand View Research. Additionally, Intel reports that the adoption of AI technologies nearly doubled [PDF] after the COVID-19 outbreak.
Though the healthcare industry is actively adopting innovative AI technologies, developing AI-powered healthcare solutions requires specific skills, resources, and knowledge of the peculiarities of such software and systems. In this article, we talk about the future of AI in healthcare, discuss how AI can improve the healthcare industry, what risks and challenges come along with AI adoption, and what steps to take to build a successful medical AI solution.
NASA’s Human Research Program is working on a machine learning-powered platform that’s supposed to identify a wide variety of issues critical during space flight, including the health of the crew. Yet in the future, the results of this project might also be applied to people on Earth, who similarly to astronauts have limited access to doctors.
This is how the healthcare industry improves itself — by embracing innovative technologies. Artificial intelligence takes a special place among innovations that can revolutionize healthcare, as it can bring significant improvements in various areas:
Most importantly, implementing AI technologies can be of great help to medical professionals, and, therefore, to patients. Successful AI applications are currently helping medical professionals in different fields and with different tasks. Let’s take a look at some examples of artificial intelligence applications in healthcare.
Technologies powered by artificial intelligence and machine learning (ML) solutions transform the way healthcare is delivered, enabling beneficial technology–human collaboration. But what are the main AI benefits in healthcare? Let’s take a look at some of the key fields where AI can shine alongside doctors and researchers.
Provide actionable insights from unstructured data
Healthcare organizations accumulate tons of data every day: health records, medical claims, images, data from medical wearables, clinical trial data, and so on. It’s just too much data to be processed and analyzed by a human.
AI and machine learning algorithms can lift the burden of making sense of all this data off of clinicians’ shoulders. AI can process huge amounts of unstructured data at a high speed, identifying unobvious patterns and providing valuable insights.
Assist in making medical decisions
Insights generated by AI systems enable medical professionals to make quality evidence-based decisions in a particular case: choose a more fitting procedure or examination, prescribe a better drug, etc.
For example, AI can be used to calculate target zones for head and neck radiotherapy, reducing the risk of unnecessary patient exposure to radiation. AI systems can also assist physicians by providing a second opinion on a patient’s diagnosis or possible treatment.
At Apriorit, we’ve recently developed a complex AI-powered system that can detect and measure follicles in ultrasound videos with as high as 90% precision and a recall rate of 97%.
Personalize patient care and assistance
Solutions like AI-powered medical wearables, medical chatbots, and telemedicine applications bring healthcare experiences to a new level, increasing the personalization of patient care.
The AI-powered Babylon chatbot can dispense medical advice similarly to helpline professionals, enabling patients to quickly check their symptoms and learn the best course of action to cure common medical conditions.
Improve the accuracy of diagnosis
While AI systems aren’t going to replace medical professionals completely, they can outperform clinicians at specific tasks. Common examples of such tasks include using artificial intelligence in medical diagnosis to detect skin and breast cancer, brain tumors, and tuberculosis.
A recent trial has shown that AI solutions can identify eye diseases just as well as human physicians. Meanwhile, an algorithm developed by a group of scientists from the University of Central Florida shows promising results in detecting COVID-19.
Apriorit’s AI engineers have created and successfully trained deep learning networks capable of performing complex tasks like detecting brain abnormalities or classifying specific types of skin cancer lesions. You can read more about this in the article below.
Enable predictive analytics for early interventions
Insights generated by AI systems enable medical professionals to not only make more accurate diagnoses but to predict potential health issues a particular patient may encounter in the future.
For example, an AI model developed by IBM and Pfizer can predict Alzheimer’s disease in healthy individuals with 70% accuracy. Verily, an application created by Google, can forecast eye disease, diabetes, and some genetic diseases.
Assist in surgery and patient care
AI-powered robotic systems have great potential in assisting and educating medical professionals.
In patient care, AI-powered exoskeleton robots can assist caretakers working with people who are paralyzed or have limited mobility. Much smaller nanorobots may assist medical professionals in combating infection in pre-surgery patients and enabling more efficient drug delivery. And in complex surgeries, dedicated AI systems can help doctors ensure absolute accuracy of the tiniest movements, reducing the risk of extensive blood loss and complications.
Speed up clinical research and drug development
Developing new drugs and treatments is a long, multi-stage process. In particular, this is due to the inability of medical professionals to ensure timely matching of patients with clinical trials. AI, however, has the potential of significantly speeding up this process.
For example, GNS Healthcare designed an AI-powered technology that builds models of particular diseases based on patient data and modulates possible responses to the drug in development. Deep 6 AI extracts and analyzes patient EHR records to choose the most fitting participants for a particular trial.
Prevent human error
Every year, the US Food and Drug Administration (FDA) receives over 100,000 reports on potential medication errors. Making the wrong diagnosis, choosing the wrong drug, or miscalculating the dosage may all have devastating effects on people’s lives. AI technologies show great potential for reducing human error and preventing fraud in the healthcare industry.
For example, an Israeli AI-powered patient safety platform can detect if a prescribed drug doesn’t match typical treatment patterns for similar cases and alert the physician of possible errors. Another AI-powered medication safety system, MedEye, assists nurses by verifying the accuracy of both a medication itself and its dosage using machine learning and visual recognition.
While having promising potential in the healthcare industry, AI technology is, however, challenging to implement. In the next section, we go over some of the biggest pitfalls you may face when working on a medical AI project.
There are two groups of common pitfalls you may face when trying to solve medical tasks with the help of AI technologies: technical and non-technical.
Let’s start with exploring key technical issues of implementing AI in healthcare.
Lack of technical skills
Many industries crave AI talent. Artificial intelligence and machine learning developers, engineers, researchers, and data scientists are in high demand.
According to the 2020 RELX Emerging Tech Executive Report, 39% of organizations can’t leverage AI capabilities due to a lack of technical expertise. It’s the second leading reason for companies to not use AI in their projects, right after budget limitations.
How to solve it: Organizations working on long-term projects may consider growing their own talents. But for companies that need to deliver quality results fast or don’t have resources for educating internal experts, it’s best to outsource AI tasks to qualified technical experts from outside their company.
Not enough structured data
Quality data is the heart and soul of any AI system. Data fed to an AI system must be relevant, well-structured, accurately labeled, and unbiased. Finding or generating a dataset that meets these requirements is a challenge on its own.
Healthcare organizations and institutions have accumulated massive volumes of data such as health records and images, insurance claims, population information, and clinical trial data. The problem is that all this data is often poorly organized and spread across multiple organizations and siloed systems. In countries that don’t use an electronic healthcare system, getting access to relevant first-hand data becomes even more difficult.
How to solve it: Look for open access datasets relevant to your specific task. Start with datasets provided by Medicare, Open Access Series of Imaging Studies (OASIS), or dataset aggregators like Kaggle and Healthcare.ai.
If a dataset doesn’t contain enough data, you may supplement it using augmentation techniques.
10 Nuances of Testing AI-Based Systems
Data security concerns
Patient data is considered one of the most sensitive types of information. Furthermore, fines for disclosing personal data are some of the highest:
- up to €20 million (around $23.5 million) for violating the requirements of the General Data Protection Regulation (GDPR)
- up to $1.5 million for violating the requirements of Health Insurance Portability and Accountability Act (HIPAA)
Therefore, you need to think about ensuring proper data security and preventing any data leaks at the early stages of your AI project.
How to solve it: All sensitive data used by an AI system should be encrypted, with access strictly limited and monitored.
Errors and biases in AI
AI systems are prone to errors and biases. Furthermore, who builds an AI system is just as important as what data and algorithms are used to build it.
Developers may accidentally implement their cultural prejudices and misconceptions into the algorithms they write. AI systems trained on unbalanced datasets with overrepresented gender, age, or racial groups can provide biased outcomes when faced with data on patients from groups who are underrepresented in the training dataset.
For instance, a recent study shows that widely used datasets with chest X-ray data are biased in terms of patients’ gender, race, and income levels. And in the healthcare sector, an error or bias in an AI system may lead to the wrong diagnosis or assigning the wrong treatment plan.
How to solve it: Make sure to use well-balanced datasets filled with diverse data and cross-check the algorithms your AI system relies on. Also, run regular audits and tests of your AI system to ensure timely detection and elimination of possible biases.
Now, let’s talk about some of the key non-technical issues of solving healthcare tasks with the help of AI.
High implementation costs
AI solutions are cost-intensive to build. A study by Everest Group shows that for one in three organizations, a lack of funds is the key struggle in AI implementation.
Among the factors that influence the cost of AI system implementation are:
- Complex, time-consuming development
- Need for rare talents
- High demand for computing resources
- Acquisition and pre-processing of quality data
How to solve it: You may try looking for ready-to-go AI models, pretrained algorithms, and open access datasets that can be applied in your project. However, be prepared to spend some time on improving and adjusting these solutions to your needs.
In addition to technical challenges, you need to be ready to face several non-technical complications.
Lack of public trust
AI-based medical solutions still need to earn the trust of both patients and doctors.
Patients often doubt the accuracy of decisions made by an AI system. They tend to see their cases as unique and believe medical AI solutions are unable to take into account all of their specific circumstances. Furthemore, a study by scientists from the NYU Stern and Boston University found that when randomly assigned to either an AI or a human healthcare provider, more people were willing to pay to switch from an AI to a human doctor than vice versa.
For medical professionals, on the other hand, it’s crucial to get a better understanding of the way medical AI works and what tasks it can and can’t solve. Yet the healthcare industry currently lacks faculty with expertise in teaching AI to medical students and practitioners.
How to solve it: Ensuring and demonstrating the fairness and transparency of an AI system’s decisions is one way for AI technology to gain public trust. Through continuous education and training, both patients and doctors can better understand how AI works and grow to trust it.
Patient privacy concerns
Training an AI system requires a lot of data. In the healthcare industry, this is usually patient data and medical records, including MRIs, X-ray scans, and blood test results.
When it comes to patient data privacy, there are two common problems:
- Nonconsensual use of patient data
- Use of personally identifiable information
How to solve it: It’s important that patients know who uses their medical records and for what purposes. Data anonymization techniques can help you ensure there’s no way to link sensitive information used to train an AI system to a particular person based on identifying data such as names, dates of birth, or timestamps on medical records.
No clear rules
AI technologies are progressing faster than relevant laws and regulations. Even though healthcare is a strictly regulated industry, it lacks clear unified standards for building an AI- or ML-based medical system.
In the US, the FDA sees AI and ML systems as software as a medical device (SaMD). The FDA offers a number of guidelines regarding risk categorization and clinical evaluation of such systems. And in the UK, the government is still working on regulations relevant to the use of artificial intelligence in healthcare.
How to solve it: Start with following the requirements of relevant industry- and region-specific laws, regulations, and data security standards, including HIPAA and the GDPR. And watch for changes in local requirements regarding the use of artificial intelligence in the medical field.
Tips on Writing GDPR Compliant Applications
Now you know exactly what pros and cons to expect when integrating AI into a healthcare solution. But what should be your first step on the long journey towards creating a successful and efficient AI-powered healthcare product?
We can outline three major points that can help you make the right decision and ensure the success of your AI-based medical project:
1. Does your project really need AI?
While many healthcare projects may benefit from implementing AI, not all of them have to do it to achieve their goals. Sometimes, it’s best to save both time and money by choosing a solution that doesn’t leverage costly AI technologies.
Before making the final decision on whether your project needs AI technologies, think about these aspects:
- Average data processing volumes. The more data your solution will have to process on a regular basis, the more benefits you’ll get from enhancing it with AI capabilities, and vice versa.
If your project doesn’t involve daily processing of thousands of files, you’d better look for a non-AI alternative: data analytics tools, cloud-based services, and so on.
- Task complexity. AI solutions are excellent at automating repetitive tasks of any kind, from filtering patient data to extracting specific patterns in MRI scans. But as with data processing, not all of these tasks should be solved with AI.
AI- and ML-based solutions work better for cases that can’t be described with simple rules and require making intelligent decisions. In other cases, you might want to consider simpler, rule-based automation solutions. Sometimes, you can even combine them: delegate a more complex task to AI while solving smaller subtasks with non-AI algorithms and technologies.
- Resource sufficiency. Due to the complexity of their development, training, and testing, AI solutions are time- and money-consuming to build.
You must have a clear view of the entire project and the scope of work to be done in order to make accurate time and budget estimates. And if your project’s resources are strictly limited, consider looking for non-AI-based alternatives or ready-to-go AI solutions, such as pre-trained models and ready datasets, that can be used in your product.
2. What data, technologies, and technical skills will you need?
Depending on the task at hand, your AI-based solution can rely on different technologies, technical skills, and data. Your goal is to determine these core sets of data, skills, and technologies long before the actual development will begin.
Here’s what you need to pay special attention to:
Quality and relevance of data. The garbage in, garbage out principle works for any AI system, especially those to be deployed in the healthcare industry. You need to determine the type and volume of data needed to build, train, and test your AI-based healthcare solution.
A quality dataset must meet several requirements:
- Include only data gathered from reliable sources
- Be properly labeled
- Contain zero damaged, poor quality, or duplicated data
- Maintain patient privacy
- Be stored securely
If you have enough resources, you can try composing such a dataset from scratch. In other cases, you may use a ready dataset from a reliable source as a basis and improve it. In both cases, you can turn to expert data scientists and AI engineers to help you prepare a quality dataset for your project.
Skilled AI development team. It’s best to form your team out of experts with hands-on experience in building the type of AI solution you need. Look for professionals familiar with the technologies that will be at the core of your solution.
Aside from true AI experts, you might need to engage specialists from other fields: data engineering, cybersecurity, DevOps, etc.
Collaboration with field experts. Form your team out of both AI and healthcare professionals, or at least make sure to consult industry experts while working on your medical AI solution. Experienced clinicians can help your development team get a better understanding of industry needs, specifics, and relevant requirements, as well as the particular problem your solution is supposed to solve. In some cases, medical professionals might even assist your team with technical tasks like labeling data.
Building an AI-based Healthcare Solution
3. What industry standards and requirements should you comply with?
Healthcare is a strictly regulated industry. Depending on the geographical reach of your solution, you might need to meet the requirements of both local and international laws, standards, and regulations.
In the US, make sure to check if your solution is subject to the FDA’s SaMD requirements. Your solution would also have to meet HIPAA requirements, as they are obligatory for all healthcare providers and companies working with protected health information in the US. And if your AI system is going to process patient data of EU residents, it will also have to meet GDPR requirements.
AI technologies have promising potential to improve the quality of healthcare services and automate daily processes for physicians. They also can empower medical professionals to make more accurate diagnoses and save more lives.
If you are determined to make AI a part of your medical solution, start with making a clear plan, gathering enough data, taking proper security and privacy measures, and consulting relevant medical professionals along the way.
At Apriorit, we have a team of passionate AI engineers who will gladly turn your ambitious ideas into real life-saving solutions. Get in touch with us to start discussing your next project!