Artificial intelligence, or AI, is an umbrella term for machine learning and deep learning. It is where a machine takes in information from its surroundings and, from that, makes the most optimal decision appropriate to the situation.
Types of AI
In machine learning, a machine can take a dataset, analyze it, and make a decision or prediction based on what it has learned. Deep learning is a more complex version of this, where there are several layers of process features and each layer takes some information.
These are both based on neural networks, which are algorithms acting similarly to the human brain in that they take an input and provide an output based on what they have learned.
However, the algorithm does not need to have the more cognitive problem-solving abilities of machine learning and deep learning to be considered an AI, it just needs to lead to the most ideal solution.
Those which are considered AI but not machine learning can be machines which just use decision trees for logic or ones that have been given rules and instructions.
Rise of artificial intelligence in medicine
AI has started making its way into life sciences due to the digitized and often massive datasets that need analysis. It allows for easier programming of machines because it removes a big portion of describing cell features. The AI only needs to be shown a set of two cell types, for example, cancerous and noncancerous, without in-depth descriptions of every feature.
For one, AI can be used during screening processes. Analysis of images, lifestyle and other health data can help in the diagnosis or prediction of onset of diseases at an early stage.
This opens up the increased possibility of using preventative measures before disease manifestation. Additionally, it may be possible to reduce false positive diagnoses.
Perhaps most importantly, AI has the possibility to help in areas with less hands-on healthcare. It is believed that geographically isolated areas can benefit from AIs which could replace physicians.
Alternatively, in areas lacking specialized staff, AI could replace the roles of for example radiologists to carry out screening routines. Such AI already exists in Korea, where tuberculosis can be identified from chest x-rays.
Applications of artificial intelligence
There are a myriad of examples of how AI can potentially be used, and the technology is currently experiencing a renaissance with several large companies focusing their efforts on AI applications in life sciences. However, some methods have been implemented already and deserve mention.
Deep Patient is a machine learning method modeled from the electronic health records of 700,000 patients from the Mount Sinai data warehouse. The algorithm was modeled using unsupervised machine learning, wherein the machine can learn to make decisions based on how similar the data is, even if no output is provided.
Deep Patient was successful in predicting diseases such as severe diabetes, schizophrenia, and certain cancers. Schizophrenia is notoriously difficult to predict for physicians, making this a very impressive feat. Deep Patient screened for around 80 diseases and managed to predict onset with 80-90% accuracy.
Further, unlike previous approaches, Deep Patient is not highly specific to a certain disease but covers a broad range of diseases while still retaining accuracy. It also shows that Deep Patient can learn descriptions and features that are not specific to a certain domain.
AI has been applied relatively extensively to cancer research. The Mayo Clinic have trained one neural network to recognize genetic mutations by analyzing the MRI image in brain cancer.
This removes the need for biopsies, as the system can identify with 95% accuracy. However, because the system does not give a reason for its identification, researchers are still not sure what exactly the algorithm is picking up on in those MRI images.
Good data and potential pitfalls of artificial intelligence
As mentioned, AI is capable of providing outputs with high accuracy. However, at its core AI is not fully understood. The math behind how an AI learns is not fully understood, and, as in the Mayo Clinic example, researchers sometimes do not know why the algorithm gives certain outputs, even if they are correct.
Machine learning and deep learning rely to a large extent on the dataset it learns from. They are susceptible to the so-called “garbage in, garbage out” principle, meaning the outputs are only as accurate as the learning sets were. Therefore, identifying the correct set of features to look at and having certainty that the data is reliable is critically important.