Healthcare, Insights, ML and AI

Using ML and AI for Disease Predictions

ML and AI for disease predictions: visualization of outbreak clusters

Machine Learning (ML) and Artificial Intelligence (AI) are finding their ways in numerous industries. They have already won a share of attention form eSports, Fintech and security, and eLearning. However, the industry that was the first to dive into the innovative application of this technology is healthcare that is using ML and AI for disease predictions among many other applications powered by data science.

In this article, SOLVVE will take a look at how exactly ML and AI are helping medical professionals to estimate, predict and prevent diseases.

What is the current role of ML and AI in healthcare?

While there are many theoretical ways in which AI and ML could help health providers, the key aspects of ML and AI application at the moment are predictions aiming to make better diagnoses faster and cheaper as well as to shift from acute treatment to prevention. This would require a very accurate mechanism of forecasting disease probability not only in a certain person but also in a region or a country. 

Ideally, we are striving here for the ability to draw predictions for a variety of diagnoses that are influenced by peculiarities of the area and its people, from climate to nutrition habits, to lifestyle, to personal history of illnesses or diseases that run in one’s family.

However, according to a paper by Fei Jiang, Yong Jiang et al. in the National Center for Biotechnology Information (NCBI) of the United States, the major areas where AI tools can help with predictions are predominantly cancer, neurology, and cardiology. Nevertheless, scientists are working on extrapolating predictive capabilities to other types of diseases as well. For example, exploring the forecasts for Alzheimer’s disease progression.

How can we achieve such capabilities?

In a nutshell, creating ML and AI for disease predictions is relatively easy as a concept. AI must be trained on the sets of data to find necessary markers and learn to distinguish them from irrelevant information, then learn to interpret the finding. To do it, scientists need to feed volumes of data from the area where AI will operate.

However, in practice, it is not that easy to achieve. Firstly, it is difficult to keep supplying new data sets. AI needs a lot of information to be trained on and to adjust its accuracy later. However, medical institutions and research centers are not able to constantly supply the necessary amount of information. Not to mention that there are difficulties in data sharing between institutions in the first place, up to the point where there are no electronic healthcare records systems to harvest data. Thus, before we even move into collecting data, it needs to be digitized and properly handled first.

Secondly, if we talk about making assumptions about the future and not just providing positive/negative diagnostics, we need to also feed such data as patients’ age, gender, habits, medical test results, family history, current symptoms, and myriads of other information. Then we also need to train AI to understand correlations and co-dependencies of all these variables. At this point, the initially simple concept becomes rather difficult to implement. 

Here is how NCBI Fei Jiang, Yong Jiang, et al. put it:

“A successful AI system must possess the ML component for handling structured data (images, EP data, genetic data) and the NLP component for mining unstructured texts. The sophisticated algorithms then need to be trained through healthcare data before the system can assist physicians with disease diagnosis and treatment suggestions.”

From “Artificial intelligence in healthcare: past, present and future”

Nevertheless, we can already see the first attempt at a complex analysis of disease spread. Reality has presented scientists with gloomy, yet potentially fruitful situations as COVID-19 pandemic provides an excellent training and testing environment for ML and AI that hopefully will help us understand one day why and when diseases occur.

Predicting the spread of COVID-19 with AI and ML

BlueDot has been in the news, including the Wall Street Journal and Wired, a while ago with their solution for AI and ML to predict and track infectious diseases, including COVID-19. The company claims they were one of the first to identify the hazard of its spread and were the first to publish a scientific report about it. Currently, it serves private companies along with governments with information and alerts about a variety of epidemic and pandemic threats.

How were they able to deliver such results? Mainly by partially solving the above-mentioned issues of data collecting and processing. BlueDot harvests its information from animal farms, hospitals, flight tickets sales, etc., and then pulls needed information and models the spread and outbreaks of a particular disease.

Unfortunately, while BlueDot’s efforts are necessary and its current capabilities are stunning, the world needs more of such solutions to deal with all sorts of diseases and health conditions, both physical and mental. There is also a need to deliver these solutions to as many domains as possible at an affordable price. If you have any questions or ideas related to ML and AI for disease predictions, do not hesitate to contact us. Let us make this happen!