With COVID-19 pandemic hitting hard, the healthcare industry is at the forefront of our defense. It needs all help it can muster, both in terms of physical aid and applying the latest technology to find the solution. People across the globe ask what can Machine Learning & AI do for healthcare? AlphaGo from Google Deep Mind has defeated the world Go champion back in 2017 — can all that investment and effort now be put to good use and help stop the coronavirus?
Actually, yes. There are multiple ways to use Machine Learning (ML) and Artificial Intelligence (AI) in healthcare. Many companies develop successful products on this market and today we briefly cover 10 examples of Machine Learning in healthcare. Read on to learn how these products work — and maybe find a way to augment your product or service with AI and ML!
Machine Learning and Artificial Intelligence: these are not the droids you are looking for
You think you know what AI is, right? There are tons of movies about smart machines capable of independent thinking and decision-making. However, the way Artificial Intelligence and Machine Learning are currently developed and applied differs a lot from the idea described in sci-fi books and movies.
Artificial Intelligence is a broad concept aimed at developing computer systems capable of mimicking human thought processes and behavior.
Machine Learning is a subset of AI aimed at providing algorithms with access to data sources and letting them analyze them according to some models. This allows data scientists to discover patterns and establish connections between previously disparate facts.
For starters, you’ve definitely heard of Google Echo, Apple Siri, Alexa and other smart assistants. These use voice recognition to transform your spoken words into commands for your smart device ecosystem.
You’ve probably also heard about how Google uses optical character recognition (OCR) to make all the road signs and other textual elements on the Street View of their Google Maps easily readable. Developments in text recognition and natural language processing (NLP) allowed Google Translator to become the great tool we use to translate big chunks of text, instead of a word salad producer it started from.
Nevertheless, really useful cases with a commercial value of applying Machine Learning and Artificial Intelligence in healthcare involve Deep Learning Networks or DNN and Reinforced Learning or RL. These types of algorithms are used to analyze huge volumes of data, find similarities and visualize patterns that were very hard to uncover without these ML models. We explore more on these below.
However, while it sounds great, AI algorithms lack the ability to make moral choices. They learn based on the data they are given, so the choice of the training data made by the data scientist greatly affects the effectiveness of the resulting ML algorithm. In addition, while there are multiple ML models, each of them is best suited for solving particular tasks (mostly analyzing huge volumes of data), so we are very far from Jarvis (or Ultron-like system) yet.
Artificial Intelligence in Healthcare: transforming the clinician’s role
Medical research and development of new treatments have long been hindered by the speed a human can absorb and process information. With the introduction of ML & AI tools in healthcare along with the infrastructure to support them, the speed of innovation can grow greatly, along with the efficiency of data analysis, drug development, Electronic Health Records (EHR) management, and other tasks. This can lead to saving more than $150 billion by 2025! Let’s take a look at examples of machine learning in healthcare that have proven their effectiveness already.
1. NLP for EHR
Keeping the patient’s records up to date is a mundane but vital task, which can lead to burnout in 83% of physicians. This routine can be successfully performed by AI algorithms that can perform administrative tasks: classify, extract, transform, and load the textual data from a variety of sources to store it in EHR systems.
2. Brain tumor or breast cancer detection using DNN
Radiology is another healthcare domain where ML can be of huge use. The official report on modern radiology workloads published in the Academic Radiology journal states that an average radiologist has to analyze a new tumor image once every 3–4 seconds to meet the demand. This is where deep learning networks come to aid, like InnerEye computer vision platform from Microsoft. When these Artificial Intelligence algorithms are trained by analyzing millions of X-ray images of brain tumors, they can diagnose the brain tumor, breast cancer, or any similar case with 99% accuracy in seconds.
3. Convolutional Neural Networks (CNNs) diagnose skin cancer
A subset of DNNs, CNNs are centered on the classification of images using OCR methodologies. This helps in making a data-driven diagnosis in complex cases like skin cancer. While experienced dermatologists can guarantee only 65–85% accuracy of their predictions, CNNs can achieve 95% accuracy in diagnosing skin cancer.
4. Drug development and trials
Releasing a new medication to the market takes around 12 years of development, pre-clinical, and clinical trials compliant with multiple regulatory protocols. Machine Learning tools like Microsoft’s Project Hanover help speed this process up immensely by ensuring automated real-time collection of study results, as well as an in-depth analysis of the existing data.
Various Laboratory Information Management Systems (LIMS) are successfully used to collect and manage laboratory data, helping pharmaceutical companies shorten drug development time. On the other hand, platforms like Pfizer and AtomNet leverage the power of IBM’s Watson AI to predict the effects of new drugs through the recombination of their molecules. This helps in immuno-oncology research aimed at fighting cancer.
5. Robotic surgery
Robots will not be able to perform neurosurgery any time soon, but they are excellent when we talk about low-invasion surgical procedures like laparoscopy. Robotic instruments ensure the precision and speed of such procedures while retaining flexibility and control hard to achieve for human surgeons. “da Vinci” systems are among the latest additions to the market that work under the surgeon’s supervision to perform high-speed and high-precision surgical operations.
6. AI-powered patient care
Patient care begins long before and ends long after the surgery ward. It involves regular healthcare procedures, therapeutic treatments, leveraging various benefits the patient is entitled to, etc. AI can be of help here too. Orderly Health helps patients and their insurers locate the least expensive and most reliable healthcare providers for required procedures based on the EHR data. This helps to ultimately save time and money on healthcare.
7. Chronic diseases analysis
Another important aspect of EHR is that it is a colossal array of data, which grows bigger — and less structured — with every passing day. Applying DNNs and other Machine Learning tools helps track patterns that were hard to define earlier — like the ways the chronic diseases respond to treatments in huge groups of patients. This way, physicians can leverage the power of Machine Learning in healthcare to automate EHR analysis and identify patterns in a population, not in single patients.
8. Early disease detection
Treating a disease early is much easier than dealing with the consequences. Thus, it is very important to detect the earliest symptoms, interpret them correctly, and assign the fitting treatment in time. Prognos AI platform claims to leverage more than 19 billion EHR cases collected and updated from more than 190 million patients. This allows Prognos to pinpoint therapy requirements and highlight gaps in trials and treatments to facilitate early disease detection for a number of conditions.
9. Precision medicine
The doctors have to base their diagnoses on the available data — symptomatic and genetic history. Thus, they have a limited set of options they can recommend for every case. However, IBM Watson Oncology is a great example of an Artificial Intelligence healthcare tool, which uses the patient’s medical history records along with vitals from various sensors to measure the response to treatments in real-time and recommend precision medicine for every patient. In the coming years, it will definitely support more and more sensors and wearable devices to increase the range of data available to physicians.
10. Epidemic outbreak predictions
2020 has reminded us of the importance of timely predicting and localizing the epidemic outbreaks. Governments and healthcare providers should leverage any means at their disposal to determine and quarantine the potential epicenters of epidemics. While not directly based on healthcare data, such solutions can use satellite data, live social media updates, records on various websites, etc. to create a real-time picture of disease outbreaks. This is especially useful in third-world countries that lack a developed medical infrastructure.
Conclusions: smart EHR management paves the way for better application of Machine Learning in healthcare
As you can see, there are various scenarios when utilizing Machine Learning & AI in healthcare can provide multiple benefits for all parties involved. The hardship here, however, is the sheer volume of the data involved in the training and daily operations of such systems. Octabyte-sized data arrays are being digitized and transferred to the cloud by leading governmental and healthcare organizations worldwide — but they are constrained with the existing EHR infrastructure and architecture.
One of the reasons why EHR management is so complex is because the current system was designed and implemented when “cloud” meant nothing more than water vapor in the skies. Mainframe systems built in Cobol and Fortran simply can’t cope with the amount of data they need to process in real-time, not to mention the need to type in hand-written documents.
As EHR management is a crucial component of providing effective healthcare services, Google, MIT, and MATLAB combined their efforts to create the next generation of EHR systems. It will use OCR technology and Google DeepMind AI to translate handwritten records into a machine-readable format in real-time. Deploying such systems to the cloud will greatly reduce the amount of effort needed to operate and maintain EHR in healthcare.
This will pave the way for the ever-growing implementation of ML & AI-based applications in all areas of healthcare — from customer experience and personalized medicine to robotic surgery and cancer treatments. Maybe, your healthcare product or service uses AI to some extent already? Or you plan to implement it in the near future? Let us know!