Algorithms, datasets, machine learning, deep learning, cognitive computing, big data, and artificial intelligence: IT expressions that took over the language of 21st-century healthcare with surprising force. If medical professionals want to get ahead of the curve, they should get familiarized with the basics of A.I. and have an idea of what medical problems they aim to solve. So, let’s take a closer look at machine learning and deep learning in medicine.
The ante-room of artificial intelligence
The term “artificial intelligence” might be misleading as due to the overuse of the expression, its meaning started to get inflated. It implies software with cognition and sentience, a far far more developed technology than it stands at the moment. For example, Facebook announced an A.I. to detect suicidal thoughts posted to its platform, but closer inspection revealed that the “A.I. detection” in question is little more than a pattern-matching filter that flags posts for human community managers.
At best, current technology with various algorithmic methods is able to reach artificial narrow intelligence (ANI) in some fields, the most advanced areas being computer vision and natural language processing. In very simple terms, ANI has incredible pattern recognizing abilities in huge data sets, which makes it perfect for solving text, voice, or image-based classification and clustering problems. But while more complex data analysing methods sound exciting and appealing, sometimes you can arrive at great results by using less advanced techniques, too.
For example, in a small Hungarian hospital, the pre-treatment waiting time for oncology patients dropped drastically from 54 to 21 days only by optimizing patient management processes with the help of simple mechanisms such as recording and following-up cases closely. The first step for data analytics is recording data appropriately – and then carrying out the necessary follow-up actions. The second step is using various statistical methods, such as data mining for collecting, analyzing, describing, visualizing and drawing inferences from data, for example from electronic health records or the ‘OMICS’ universe. The focus is on discovering mathematical relationships and properties within big data sets and quantifying uncertainty. Data mining usually means when insights and patterns are extracted from large-scale databases.
However, this is only the ante-room of artificial intelligence: machine learning and deep learning go far beyond that.
Pattern recognizers rule the world: supervised machine learning and measles
Machine learning is the field of computer science that enables computers to learn without being explicitly programmed building on top of computational statistics and data mining. As with traditional statistics, machine learning requires sufficient training datasets (also known as sample size in traditional statistics) and the right algorithms to optimize its performance on the training dataset before testing. However, in contrast to the tradition methods, machine learning is focused on building automated decision systems.
The field has different types: it could be supervised, unsupervised, semi-supervised or reinforcement learning, among others. The first is typically used for classification problems, e.g. pairing pictures with labels. You have an input and output data – the image as well as the label -; and the algorithm learns general rules how to categorize. It is the most popular type of machine learning in medicine, and in a few years, it will be widely used in medical imaging in radiology, pathology, and other image-intensive fields. Although it certainly has its limitations: it requires large data sets to become accurate enough, and the data has to be appropriately labelled. That’s why the work of data annotators is so crucial.
Nevertheless, supervised machine learning can also be effectively deployed to predict health events based on various input data, which can be classified in a linear way. For example, from statistics on measles vaccination rates and disease outbreaks from the Centers for Disease Control and Prevention, as well as non-traditional health data, including social media and syndromic surveillance data generated by software that mines a huge range of medical records sources, an algorithm can provide a reliable map of future measles outbreak hotspots.
Unsupervised machine learning and drug interactions
In the case of unsupervised machine learning, the computer program is asked to discover inherent structure and patterns that lie within the data. Unlike in the case of supervised machine learning, the data sets are unlabeled and unstructured. Thus, the algorithm has to make up its own groups, clusters, and categories based on “similarities” in huge data sets. It is usually used to predict unknown results and to determine how to discover hidden patterns. Unsupervised machine learning has subtypes: clustering algorithms and association rule-learning algorithms.
Unsupervised learning is often used in deep learning, and has been implemented in self-driving vehicles and robots as well as being used in speech- and pattern-recognition applications. In medicine, for example, tissues samples can be clustered based on similar gene expression values using unsupervised learning techniques. As an example of association rule-learning algorithms, the testing of novel drug-drug interactions can be mentioned.
Reinforcement learning and the magic of AlphaGo
The last category, reinforcement learning constitutes probably the most known type of machine learning: when the computer program learns from its mistakes and successes; and builds its experiences into the algorithm. The most famous example for it is AlphaGo, the machine developed by Google that decisively beat the World Go Champion Lee Sedol in March 2016. Using a reward and penalty scheme, the model first trained on millions of board positions in the supervised learning stage, then played itself in the reinforcement learning stage to ultimately become good enough to triumph over the best human player.
However, the problem with applying reinforcement learning to healthcare, especially for optimizing treatment, is that unlike with AlphaGo, we cannot play out a large number of scenarios where the agent makes interventions to learn the optimal policy – as the lives of patients are at stake. Luckily, we already have examples where this issue can be mitigated. In a study published by MIT researchers, the authors reported a successful formulation of clinical trial dosing as a reinforcement learning problem, where the algorithm taught the appropriate dosing regiments to reduce mean tumor diameters in patients undergoing chemo- and radiation therapy clinical trials.
Deep learning in medicine used for very complex issues
Deep learning is the subfield of machine learning where computers learn with the help of layered neural networks. However, there’s no strict line between machine and deep learning, usually, the cleanliness of data and the complexity of the problem determine which one is more applicable. Deep learning algorithms usually deal with messy data sets, unstructured piles of information to try to give answers to difficult questions.
What are neural networks? Their operation basically imitates the neurons in the brain. Quartz formulated the explanation as the followings: algorithms which are roughly built to model the way the brain processes information, through webs of connected mathematic equations. Data given to a neural network is broken into smaller pieces and analyzed for underlying patterns thousands to millions of times depending on the complexity of the network. A deep neural network is when the output of one neural network is fed into the input of another, chaining them together as layers. Typically, the layers of a deep neural network would analyze data on higher and higher levels of abstraction, meaning they each throw out data learned to be unnecessary until the simplest and most accurate representation of the data is left.
Deep learning has different types based on the ways of connecting layers and the ways ‘neurons’ act. There’s also unsupervised, supervised and reinforcement learning in deep learning algorithms, as these signify the way the algorithm is fed with data by researchers. Beyond those, convolutional neural networks (CNN) are typical for recognizing images, video, and audio data, due to their abilities to work with dense data. Recurrent neural networks (RNN) are used for natural language processing, while long short-term memory networks (LSTM) are variations of RNNs meant to retain structured information based on data. For instance, an RNN could recognize all the nouns and adjectives in a sentence and determine if they’re used correctly, and an LSTM could remember the plot of a book.
As an example of deep learning in medicine, researchers proposed an approach to deduce treatment policies for septic patients by using continuous state-space models and deep reinforcement learning. In another study, experts attempt to solve the difficult problem of estimating polyp size using colonoscopy images or videos, which is crucial for making a diagnosis in colon cancer screening. Moreover, unsupervised deep learning may facilitate the exploration of novel factors in score systems or add hidden risk factors to existing models. It can also be used to classify novel genotypes and phenotypes from pulmonary hypertension, cardiomyopathy, and many other factors.
Navigating in the sea of information about artificial intelligence is tough. As everyone realized that the technologies leading to A.I. could revolutionize healthcare, there’s a lot of experiments, research, but also bogus information and overhype out there. We, at The Medical Futurist, aim to provide you with context, to interpret study results and attempt to make sense of the digital health revolution. Feel free to reach out to us for questions, comments or just a conversation. We’d love to hear from you!