Artificial Intelligence, Deep Learning and Medical Treatment: Exciting Potential for Ground-Breaking Developments Ahead
Have you ever wondered about the potential efficiencies and applications of Artificial Intelligence (AI), or how AI has evolved so quickly in recent years? One explanation is the notion of Machine Learning (ML, a subset of AI), which enables computers to learn from data without being explicitly instructed or programmed to do so.
Machine Learning is viewed as one of the most pivotal sectors of AI, and experts in the field are constantly searching for new ways to uncover ground-breaking insights through its use. At the most fundamental level, ML consists of training computers to learn from data so that they can recognize patterns or make decisions.
A popular analogy used to explain this concept is how children are shown various shapes or colours, and over time, learn how to identify the differences amongst them. Similar to how the goal for children in this respect would be to assert, for example, “that is a red triangle”, the goal for machine learning engineers and data scientists is to build an algorithmic model that asserts an insight of importance, such as “the most profitable product is product B, and customers belonging to segment D are buying the most of it”.
One significant result of these efforts has been the rapid development of Deep Learning — a subset of ML which incorporates human-like brain function into computer-programming. This integration enables the processing of complex data, such as audio input and visual imagery.
You may ask: how effective is Deep Learning (DL) and what are some of its potential applications in tackling complex problems?
Imagine a world where the only barriers to solving such problems are a lack of creativity and critical thinking. Deep Learning allows computers to build upon and refine these skills, so that the future solution of such problems may become a reality.
One recent example of this reality was realised by Google DeepMind, which built a Deep Learning model that found 2.2 million undiscovered crystals, 380,000 of which have the potential to power the development of new technology. Experts estimate that this knowledge is roughly equivalent to 800 years worth of human knowledge and exploration!
First developed in the 1960s, DL was applied to forecast weather and recognize speech. We have since come a long way in the realm of AI and DL, recognizing that certain types of Deep Learning models best interact with certain data types (like text, audio or images). Today, some of the most impactful and widely used models are:
- Convolutional Neural Networks (CNNs): inspired by the human visual cortex, recognizes patterns and features in images and videos.
- Recurring Neural Networks (RNNs): excel in recalling previous input data to analyze and understand sequential patterns.
- Generative Adversarial Networks (GANs): generates artificial data from training input to augment understanding and improve overall data quality.
- Deep Reinforcement Learning: inspired by how humans learn, allows machines to make optimal decisions through trial and error and positive reinforcement.
Deep Learning: A Promising Future For Healthcare
So, why should you care about DL and what are some promising discoveries that have been made in recent years?
One industry that has been revolutionised by this technology is healthcare. As someone whose life has been touched by the impact of cancer, I wanted to understand how these developments could pave the way for a promising future of disease recognition, patient treatment and drug discovery. The findings are fascinating!
One of the most exciting applications of DL is the use of CNNs to analyze patient-scan abnormalities that are indicative of malignant tumours.
One CNN model, developed by Google Health and DeepMind, has outperformed radiologists in detecting breast cancer with 11.5% greater accuracy.
Another promising discovery was made by the Hindawi, an Egyptian-based healthcare research organization. Hindawi found that CNNs could identify colon cancer at a level of 99.8% accuracy (surpassing the average radiologist’s accuracy, equipped with CT and MRI equipment, of 85%).
According to statistics collected by the Canada Cancer Society, this means that radiologists (without the use of CNNs) currently misdiagnose 3381 breast cancer cases per year (10 cases per day) and 2338 colon cancer cases per year (6 cases per day) in Canada, respectively.
Another groundbreaking development in the realm of cancer treatment involves the use of Deep Reinforcement Learning (DRLs). Researchers at Stanford University have developed DRL algorithms that analyze MRI scans and biological patient responses to predict individual tumour growth and mitigate the negative impacts of chemotherapy treatment. Although exploration of this type of DL is in its early stages, it paves the way for enhanced tumour control, personalized treatment plans and the acceleration of new drug discovery.
It is challenging to know how widely this new technology will be used for medical diagnoses and treatment in the near future. However, it is safe to say that the potential applications of AI and DL in the healthcare field are extremely promising. Who knows, with accelerated advancements, Artificial Intelligence and Deep Learning may help us discover cures for diseases that we, as a society, had previously thought were not possible!
