Is it not magical how Siri on your phone listens to your voice, talks to you, and helps you out? Or have you ever wondered how your phone recognises people in pictures? Or maybe how Netflix gives you spot on recommendations for the next binge-watch-worthy tv show? The answer is – unsurprisingly – artificial intelligence (AI). Most do not realise how many times a day AI makes our lives easier. Imagine having all those emails announcing you as the next lottery winner in your inbox instead of your spam folder.
Mistakenly, artificial intelligence (AI) is often thought to be a way of making machines human, but this is not quite true. Intelligence is defined as the ability to learn, understand, and act rationally. Hence, AI enables computer systems to learn from experience, recognize patterns, and act accordingly. More specifically, through AI technology computers can learn from robust, labelled datasets to solve problems and make predictions. But we can all agree that these abilities alone do not make a machine human. Basically, in AI you provide the system with an algorithm and then give the system well prepared data sets to learn from.
Algorithms are the base of AI. It is like giving the computer system a cooking recipe which it follows. You will have to define the ingredients and quantities (data input), the recipe / methods (process), and the desired final product (output). Only that in AI the recipe is written in programming language (e.g., Java, Python, etc.) and the final product is a prediction or a decision. But not all algorithms are AI algorithms!
Maybe you can already see the great advantage and benefits of AI. Computer systems are enabled to perform tasks that normally require human intelligence. However, through AI they can analyse and learn from thousands of lines of data input in milliseconds, which makes AI so much more efficient than a human.
Deep Learning and Machine Learning
As you can see from the graphic below, machine learning (ML) is a sub-field of AI and deep learning is a sub-field of machine learning.
Machine learning is made up of a series of algorithms. These algorithms will adapt and enhance as they are fed with more datasets. This evolution and self-improvement is called model-based learning. The AI algorithm will find patterns, and create assumptions and predictions based on those patterns. The algorithm will get better and better at making those predictions and decisions, the more data it analyses. It is like a human – the more it learns and practices, the better it gets.
While machine learning algorithms are rather “simple” (e.g., a decision tree or linear regression) and need human intervention when programming how to extract and classify data, deep learning is based upon multi-layered artificial neural networks resembling the human brain.
Let us try to understand the difference by looking at an example. Imagine we want to build an algorithm that identifies dogs in pictures. With machine learning, first we would have to extract the attributes and characteristics from a picture with a dog. What makes a dog a dog and which attributes should be analysed? After this we will be able to write an algorithm that incorporates those attributes and then the algorithm can analyse images, classify the data, and detect dogs in a new picture. If the outcome is erroneous, we can adjust the algorithm and correct mistakes. In the beginning a cat might be classified as a dog but over time the algorithm will learn and evolve, and it will become increasingly precise.
Now, with deep learning / deep neural networks the algorithm will be able to extract and characterise the attributes from the picture by itself. The algorithm will learn from its own errors. It will go back and forth – basically trial and error – and will over time become better and better. Look at the graphic below to understand this difference. Neural networks and deep learning are still quite new, but already immensely powerful.
Applications in Healthcare
Why is artificial intelligence important and interesting for medicine and MedTech?
Especially deep neural networks might be able to solve long standing problems and issues within medicine and healthcare. These include severe diagnostic errors followed by incorrect treatment, and massive waste of resources. Further, a lot of workforce is lost due to time consuming administrative tasks as well repeatedly performed standard diagnostics that have to be assessed manually. Time between patients and their clinicians remains inadequate . Deep learning could automate a lot of these processes and make diagnostics and hence clinical decision making more precise. During a time where smartwatches and fitness trackers have become quite popular, artificial intelligence could provide patients with even greater autonomy and could bring us closer to a more patient centred healthcare with highly personalised treatment plans .
AI is being widely investigated in all areas of healthcare and medicine. Radiology is probably the field where the most potential for deep learning and automation exists. For example, AI is applied to analyse chest X-rays achieving expert level performance in a task which is time consuming, requires expert knowledge, and is prone to fatigue based diagnostic error . In ophthalmology AI algorithms predict loss of sight and visual acuity outcomes in patients with neovascular age-related macular degeneration – the major global cause of blindness . In dermatology a deep learning tool was built that can distinguish between and diagnose 26 of the most common skin conditions seen in primary care . In cardiology AI augments analysis of electrocardiograms (ECG) enabling early detection of atrial fibrillation and heart weakness and detecting patterns occult even to experienced cardiologists . In Psychology AI enhances the prediction, monitoring, and treatment of post traumatic stress disorder (PTSD) [7,8].
But concerns also exist. It will be important to preserve the human element of medicine when implementing these tools – empathy, teamwork, communication, and emotional factors. Further, as AI and especially deep learning algorithms become increasingly complex and self evolving, it will become truly important to find ways to control the actions of the system and to be able to retrace how the algorithm reached a decision. This is often described as the “Black Box” of AI [9,10].
These are just a few of the possible applications of artificial intelligence in medicine. The field is rapidly evolving and in the long-run medicine will benefit immensely from integrating deep learning tools into clinical practice. Ultimately, it will not just save practitioners time but will also lead to more accurate and earlier diagnosis of disease and hence better treatment. This results in better conditions and outcomes for patients – which is the ultimate goal of healthcare and medicine. A healthy and happy patient.
That is why we at MedTech UCL have taken a huge interest in artificial intelligence and have been hosting the AI in Medicine conference for the past years. If you want to hear about the latest advancements of AI in Medicine from leading researchers and professionals in the field, the conference will not disappoint you. You will be able to ask questions, network, and meet fellow students that are like-minded.
3. Rajpurkar P, Irvin J, Ball RL, Zhu K, Yang B, Mehta H, et al. Deep learning for chest radiograph diagnosis: A retrospective comparison of the CheXNeXt algorithm to practicing radiologists. PLoS Med. 2018 Nov 20;15(11):e1002686.
4. Abbas A, O’Byrne C, Fu DJ, Moraes G, Balaskas K, Struyven R, et al. Evaluating an automated machine learning model that predicts visual acuity outcomes in patients with neovascular age-related macular degeneration. Graefes Arch Clin Exp Ophthalmol. 2022 Feb 5;
8. Schultebraucks K, Shalev AY, Michopoulos V, Grudzen CR, Shin SM, Stevens JS, et al. A validated predictive algorithm of post-traumatic stress course following emergency department admission after a traumatic stressor. Nat Med. 2020 Jul;26(7):1084–8.