Are you looking to do the Neural networks health industry?
Merehead is a leading software development company. Talk to our experts to get a turn-key solution!
Write to an Expert
The artificial intelligence market in healthcare has been estimated at $8.23 billion in 2020 and is expected to reach $194.4 billion by 2030, growing at an average of 38.1% from 2021 to 2030. The main driver of market growth will be artificial neural networks, because they are the reason how a digital "mind" can learn so quickly and efficiently. In this article, we will tell you exactly how artificial neural networks are used in medicine today.
Surgeon's procedures require a deep knowledge of medicine, a high level of accuracy and the ability to adapt quickly to changing circumstances along with constant attention over an extended period of time. Although well-trained surgeons typically have all of these qualities, but they are just ordinary people and therefore can make mistakes, especially if they have a busy schedule. According to research by Johns Hopkins University, surgical errors occur more than 4,000 times a year in the United States alone. For example, at least 39 times per week a surgeon leaves foreign objects inside patients, more than 20 times per week surgeons are performing the wrong surgery or operating on the wrong organ.
Smart Tissue Autonomous Robot or STAR is the first AI that has planned and performed soft-tissue surgery (pigs) without human intervention
Robot-assisted surgery can alleviate this problem. In particular, AI based on neural networks can be used to model and plan surgery, to assess surgeon skills and to simplify surgical tasks. Also, robots with mechanical arms can perform surgical procedures on their own, which will make these kinds of services faster and cheaper, as well as making highly professional surgery more common - a cool surgeon would hardly want to live in poverty-stricken areas of Africa or Asia, whereas a robotic surgeon can be placed in any room with electricity.
And these are not just words: the robotic surgery industry has already been valued at $40 billion and it's showing excellent results:
- Artificial neural networks analyze data from preoperative medical records to guide the surgeon's instrument during the operation, cutting the patient's hospital stay by 21% or more.
- A study including 379 orthopedic patients found that a neural network-based robotic procedure produced five times fewer complications compared to surgeons working alone.
- Cardiac surgeons are assisted by a miniature Heartlander robot. It enters a small incision on the chest to perform stable and localized probing, mapping and treatment over the entire surface of the heart. Its usage reduces the damage to the patient if it is necessary to access the heart.
- The Smart Tissue Autonomous Robotic Surgeon (STAR) independently performed a complex task on a pig's soft tissue (reconnecting two ends of the intestine) showing significantly better results than human surgeons.
Medical imaging refers to the process of creating visual representations of internal body structures for clinical analysis and medical intervention, as well as visual representations of certain organs or tissue functions. Visualization allows better diagnosis when using X-rays, CT scans, mammograms, MRIs, PET scans, ultrasound scans and other procedures.
The medical community recognized long ago that imaging can greatly increase the probability of detecting disease. The problem is that even the trained human eye of a doctor is often unable to recognize the fine details in images. Besides, also we shouldn't forget about the problem with "blurred eyes": the researchers from Harvard University added the image of a gorilla to x-rays and demonstrated them to doctors - 83% of radiologists didn't notice the image of a gorilla on the x-rays.
Only 2 doctors out of 10 noticed the gorilla in this picture.
Converging neural networks (CNNs) are great for medical visualization. According to scientists at Stanford University, CNNs are being developed with the expectation in mind to process images, which means they can be used in medicine, for example to analyze MRI or X-ray images. Moreover, CNNs are often superior to human accuracy. For example, professional dermatologists have a melanoma detection accuracy of 65% to 85%. Whereas solutions such as TensorFlow, scikit-learn or keras demonstrate accuracy levels of 87% to 95%.
Neural networks are 10% more accurate at detecting melanomas than human dermatologists.
At the same time, they work much faster than humans, without lunch breaks or days off: Researchers at Mount Sinai Icahn School of Medicine have developed a neural network
capable of diagnosing important neurological conditions, such as stroke and brain hemorrhage, 150 times faster than human radiologists.
Here are a few more examples of medical imaging with neural networks:
- In 2019, researchers used a convolutional neural network built on ResNet50 and InceptionV3 architectures to analyze medical image sets and dermatoscopy. This solution provided the same level of accuracy as professional human diagnosticians.
- One breast cancer early detection tool developed by the Houston Medical Research Institute interprets mammograms with 99% accuracy and provides diagnostic information 30 times faster than humans.
- A team of Chinese researchers has developed a neural network system to analyze radiographs for early signs of COVID-19-induced pneumonia. This tool saves physicians up to 40% of their diagnostic time, allowing them to identify, isolate and treat infectious patients more quickly.
- A research team from Spain has developed a neural network-based deep learning algorithm to improve MRI resolution. It helps to identify complex brain-related pathologies, including cancer, speech disorders and physical trauma.
- Facebook AI (now Meta) and NYU Langone Health have developed AI called fastMRI. It offers a new approach to MRI scanning by speeding up the imaging process 4 times faster. When such images were provided to radiologists, they couldn't tell the difference between a traditional scan and one created with fastMRI.
Virtual Care Assistants
Another thing that neural networks can help you with is interacting with patients in things like patient requests, managing confidential patient medical information, scheduling appointments with doctors, sending test reports, reminding them of appointments to medical facilities and so on. If these tasks were handed over to AI, it would save $20 billion annually in the U.S. healthcare market alone.
Virtual nurse assistants such as Care Angel are capable of conducting first appointments, arranging hospital visits, partially maintaining medical records, reporting test results and even conducting health checks using voice. Another example of such a solution is Sensely. This virtual assistant is an interactive application designed to streamline the medical workflow, automate routine processes and reduce the cost for monitoring patient care. Studies show that the use of Sensely can reduce labor costs for patient monitoring by 66% compared to the usual procedure.
Workflow and administrative tasks
The New England Journal of Medicine (NEJM) survey among British physicians found that more than 80% of respondents considered physician burnout to be a problem in their organization. At the same time, half of these respondents reported that "relieving administrative tasks" could fully or partially solve this problem by allowing physicians to share more time with patients as well as spending additional time on professional development. Moreover, the biggest complaint about the necessity of maintaining Electronic Health Records (EHRs) for patients, which are being used by many hospitals around the world.
The prevalence of burnout among medical personnel in the United States.
Artificial intelligence based on neural networks could take over much of the routine clerical work of filling out documents and reports, saving up to $18 billion in healthcare. For example, an algorithm for transcribing voice to text could help to fill out EHRs, assign tests, prescribe medications, take notes and other things using the voice during a patients' check-up.
One example of such solutions is Olive, an AI-powered platform that automates several administrative processes, such as verifying the eligibility of unwarranted medical claims, transmitting necessary medical data to appropriate health care providers, etc. Olive integrates easily with existing medical software tools.
K2 Process Automation provides similar services. This platform uses artificial intelligence to optimize the management
of medical facilities and their personnel. Specifically, K2 AI can automatically update data in patient records, handle patient requests by sending them notifications when they receive test results, log calls and save information appropriately.
Predicting infectious disease outbreaks
At the end of December 2021, the BlueDot platform predicted a coronavirus outbreak in China. BlueDot's artificial intelligence, which studied news in 65 languages, people's social media conversations and disease reports, reported the outbreak of a new flu-like infection a week earlier than the World Health Organization (WHO) had done. And this is not the first such case. BlueDot had previously predicted a Zika virus outbreak in Florida in 2016, six months before it happened. BlueDot had previously predicted the Ebola outbreak in 2014 and its spread outside of Africa.
Another startup that has shown its ability to predict infectious disease outbreaks is Metabiota. According to TechCrunch, the platform has helped to stop the spread of Ebola in Sierra Leone.
The HealthMap neural network managed to catch CoVID-19 when Wuhan doctors first started discussing the strange disease on the ProMED-mail site. Any Internet user can use that site's data.
Creating drugs and new therapies
Atomwise, an American pharmaceutical company using artificial intelligence capabilities, in partnership with IBM, screened compounds capable of binding to a glycoprotein that prevents Ebola from entering cells inside the body during the 2015 Ebola virus outbreak in West Africa. The Atomwise neural network performed this analysis in less than a day, although this process usually takes months or even years.
This was the first time that AI had accelerated the drug development process so much. In another five years, the United States will approve the first drug almost entirely created by AI - the drug DSP-1181 for the treatment of obsessive-compulsive disorder. This drug was created at Exscientia and according to its management, it took only 12 months.
But that's not the limit, either. Insilico Medicin claims that their GENTRL AI is capable of creating new drugs against certain pathologies in just three weeks. It will take about 25 more days to select the best option and test it. Thus, it takes only 46 days to create drugs with GENTRL.
However, the most ambitious project in this direction is Deep Genomics. The company is developing an AI
called Project Saturn to discover new methods for repairing the effects of genetic mutations. Deep Genomics has already estimated 69 billion oligonucleotide molecules compared to 1 million targets in silico to create a library
of 1000 compounds that are experimentally validated to manipulate the cell biology on purpose.
Personalized medicine is another resource-intensive area in healthcare where artificial neural networks can be useful. This is because each treatment case is unique, as each person has a unique immune system, medical history and current state of health. And all of this has to be taken into account to maximize the effectiveness of the treatment. The only thing is that a doctor, firstly, cannot always see all the nuances in the patient's condition and secondly, the doctor does not have the necessary competence to make a comprehensive decision to build the most effective treatment plan taking into account all these nuances - this requires the deepest knowledge in all fields of medicine, which is unrealistic.
On the other hand, artificial intelligence can do it. It is able to study and analyze huge amounts of data and make decisions that consider all the available data on both the patient's condition and treatment options. Currently, there are not a lot of such projects (they are too complex and require huge amounts of high-quality medical data). But there are already the first successes in this direction: IBM's AI Merative (formerly called Watson Oncology) is designed precisely to solve such problems.
Fraud is another important problem in the healthcare industry. Counterfeit drugs, insurance fraud as well as substandard, illegal and deliberately ineffective medical services cause as much harm to the industry as fake and substandard clinical trials. In the United States alone, it is conservatively estimated that healthcare fraud costs the country about $68 billion a year or 3% of the nation's total healthcare spending.
Artificial intelligence based on neural networks can also combat this problem. For example, the medical company Aetna has about 350 neural network models
for security and fraud protection. Highmark Inc also has its own NN to combat fraud and according to the company, their neural network-based product has already helped the company to save $245 million.
How often have you missed medication appointments, forgotten or ignored preventive exams with your doctor? Surveys show that this is a very common problem and doctors have almost no control over it. Moreover, admonitions and criticism of patients' behavior by doctors only exacerbate the problem because patients become ashamed of their illnesses and simply stop going to the hospital.
Different gadgets and apps that correct people's behavior can help with this. The most obvious example is fitness bracelets and smart watches, which help set goals (for example, to walk 4 kilometers a day) and track their achievement, as well as reminding about preventive check-ups and the importance of physical activity. More sophisticated apps, such as Somatix, track patients' daily activities and point out their habits and routines so they can focus on eliminating them and getting better.
Somatix integrates easily with smart gadgets that track patient activity to adjust the patient's behavior.
Also, don't forget the various applications for eliminating< bad habits and creating good ones. These create the "external control" needed by the patient, which is frequently more effective as a motivator than doctor's advice, requests or even threats from relatives. At the same time, such apps are especially effective if they use elements of gamification - using game practices (rewards, achievements, encouragements, levels and account growth) in a non-game context.
Data collection and analysis
Finally on our list, but not the least important problem is collecting and analyzing qualitative medical data. And it is not only collecting the patient's medical history, but also various related data from where the patient works and how often he goes to the gym until the movies he watches and the food he eats, which ideally should also be taken into account when constructing a treatment plan.
The problem is that it is very difficult and costly to collect this kind of data. People won't be willing to report on every purchase, run or horror movie watching. However, artificial intelligence-based platforms can do that. If such a solution is integrated with a smartphone
, a smartwatch, a refrigerator, online banking and other sources, the artificial intelligence will collect such statistics for each patient - automatically and with privacy.
There are many examples of such solutions. Current Health, for example, has a gadget with artificial intelligence for medical monitoring of the human condition. This wearable device was one of the first to be approved by the U.S. Food and Drug Administration (FDA) for home usage. The gadget can measure a patient's pulse, breathing, oxygen saturation, temperature and mobility to provide this data to the doctor, who will make recommendations for the patient.
Current Health allows healthcare organizations to personalize and scale the way they provide home health care with a single, flexible solution.