Artificial Intelligence and Medicine

• Read original article here

Artificial Intelligence (AI) is an emerging technology that almost everyone has heard of. This technology is already having a huge impact on the world from voice to speech bots all the way to the algorithm which detects faces in your phone. Now due to its significance in our everyday life, it is important to understand what this technology is and to not get confused by the huge amount of words involved in this area of technology like deep learning or machine learning which are mostly thrown around as synonyms of artificial intelligence. Thus in this article, I will teach you about AI and its subsets. On top of this, we will look at neural networks and the medical applications of artificial intelligence.

In 1955, John McCarthy coined the term “Artificial Intelligence” and is recognized throughout the world as the founder of this technology. Personally, he described this technology as “the science and engineering of making intelligent machines, especially intelligent computer programs”. AI is a computer program which can do things that normal programs cannot do. Generally, AI can be coded using a complex statistical model or a lot of if-then statements. The latter method of programming AI is normally referred to as Weak AI as it is not able to carry out very complex processes as it is constrained by an if-then architecture. The utilization of complex statistical models is referred to as Strong AI and it is mostly the way to go, as these systems can perform a wide variety of tasks in an intelligent way, depending on how they are trained. Now within this broad area of AI, there is a subset of AI called machine learning (ML).

Machine learning is a technology that has the ability to learn from data and can adapt its methods when exposed to higher volumes of data. This aspect of having a computer system “learn” without any human intervention is very efficient and can be used in a variety of ways. The huge category of machine learning can even be further divided into supervised learning, unsupervised learning and reinforcement learning.

Supervised learning is where a computer system learns from a dataset that has been accurately labeled. A labeled dataset is one which has a list of inputs and the correct outputs for each of those inputs. With these labeled datasets, machine learning algorithms identify the patterns in the dataset and use these patterns in order to predict outputs for new input data. This type of learning is very powerful and can make the computer system exceptionally accurate for predicting the correct output.

Unsupervised learning is a machine learning method that is programmed to identify patterns and structures in unlabeled data. In order to organize and analyze unlabeled data, many of the unsupervised learning algorithms cluster the data into many different categories making complex data very easy to analyze. In general, unsupervised learning is strong for organizing and identifying patterns in unlabeled data. The magnitude of data that can be provided to unsupervised learning models is infinite as almost all data is unlabeled.

Reinforcement learning is when a machine learns by reacting with its environment repeatedly until it is able to come up with the best possible algorithm for the task provided. Specifically, the computer system is encouraged to improve its strategy through a reward system where if it does good it is rewarded while if it does bad, it is penalized. In this way, the machine continually strives to maximise rewards by changing its strategy based on past experience.

Deep learning is a subset of machine learning and in a broader sense, it is a technique that is predominantly based on the human brain. Systems based on this method can scale up their performance with access to more volume of data making them highly useful in commercial applications. Explicitly, Deep learning uses a layered structure of algorithms called an Artificial Neural Network (ANN) which is based on the biological neural network in the human brain.

A neural network is the foundation of deep learning and it is based on the brain’s biological neuron arrangement. In a neural network we have different layers which include the input, hidden and output layer. Keep in mind that a neural network can only have 1 input/output layer, while it can have multiple hidden layers.

In each of the layers, there are nodes which act like neurons and are responsible for the computation we see in AI algorithms. When going from one layer to the next, each node in the previous layer is connected to every single node in the next layer. These connections have specific weights and the value of a node is the weights of it’s connections multiplied by the values of all their corresponding nodes, added together. For instance, if we consider a part of a neural network where one node is connected to 3 nodes of the previous layer, we compute value in the following way:

You might notice that in the above diagram, there is also a constant “b” being taken into account when calculating the value of each node. This “b” represents the bias for the node y0 and in general, biases help neural networks better fit the data they have been given. Now in this way, as we compute the values for nodes in each layer, they directly impact the values of the nodes for the next layer. At the end of the neural network, depending on the values of the second-last layer’s nodes, an output will be computed by the output layer.

With the neural network design, you might wonder how a neural network is trained by data to give accurate outputs. Well, a neural network’s accuracy is all dependent on the weights and biases in the network. Thus, the training or so called “learning” a neural network performs is simply the action of tweaking the weights and biases so that the network becomes accurate. This tweaking involves using a cost function and back-propagation which is a huge topic and I plan to discuss it in another article. For now here are some sources that you can use to learn about training a neural network:

There are a variety of applications of AI in many different industries. However, one industry which will be massively impacted by AI in the coming years is the field of medicine. Current applications of AI in the medical industry include the following:

Through the utilization of a deep learning method known as a convolutional neural network (CNN), doctors are able to input medical images into Deep Learning algorithms to get an accurate diagnosis. For example, Deep Learning algorithms which take MRI scans of the brain to diagnose Alzheimer’s have been developed and due to the continual advancement of technology, these algorithms are becoming more accurate every single day.

In 2020, the MIT laboratory was able to leverage AI in order to discover new viable antibiotic drugs. The usage of AI in drug discovery is very impactful as it makes the process of discovering drugs way more efficient and easy. In terms of how AI can discover drugs, it uses a dataset of effective drugs against a certain disease and then the algorithm is trained on this dataset. After training has taken place, thousands of compounds are tested to see whether they have drug-like properties against the disease. With this method of drug discovery, the process becomes much more efficient and allows researchers to discover new drugs at a much faster pace.

Overall, this is a brief overview about AI, it’s subsets, neural networks and finally the applications of this technology in the medical field.