Artificial Neural Networks: Solving Challenges in Health Sciences - Perficient Blogs
Blog
  • Topics
  • Industries
  • Partners

Explore

Topics

Industries

Partners

Artificial Neural Networks: Solving Challenges in Health Sciences

There is a lot of buzz in healthcare and life sciences right now around Artificial Intelligence, and the potential uses for Artificial Neural Networks (ANN) and Deep Learning to solve for all manner of messy and complex problems. Deep-Learning software attempts to mimic the activity in layers of neurons in the neocortex[1], this includes cognitive processes such as pattern recognition and classification, concept association, learning, sensorial perception, and optimization.

ANN’s are being implemented today to address a myriad of applications. For example in healthcare:

  • Informing clinical diagnosis
  • Predicting future disease
  • Analyzing images

And in life sciences:

  • Signal interpretation
  • Real World Evidence/Drug Development
  • Market Research and Customer Service

What is an Artificial Neural Network and how does it work?

An Artificial Neural Network is a computational approach also referred to as a Connectionist System used in machine learning. ANN’s are loosely modeled after the biological neural network in an attempt to replicate the way in which we learn as humans. Think of it as a computing system, structured as a series of layers, with each layer composed of one or more neurons. The layers comprise input, output and hidden layers as follows:

  • Input neurons receive various forms of external information that the network will learn about, recognize, and process
  • Output units sit on the opposite side of the network and signal how it responds to the information it’s learned
  • Hidden units sit in between input and output units serving as the mechanism to transform signals for interpretation between input and output

Figure 1: Depiction of a Neural Network, where each circle is a neuron and the arrows indicate the connections between neurons in consecutive layers.[1]

One technique that led to the broader development of ANN’s is the Backpropogation algorithm, which allows the neural network to be trained, and provides detailed insights into how changing the weights and biases across all the connections, changes the overall behavior of the network to allow the ANN to arrive at the right answer. Backpropogation uses the comparison between the outputs a network actually produces with the output it was supposed to produce, applying the difference between the two to modify the weights of the connections between the units, so the network can get it right.
Figure 1: Depiction of a Neural Network, where each circle is a neuron and the arrows indicate the connections between neurons in consecutive layers.[2]

Deep learning neural networks, use different layers within a multilayer network to extract different features until it can recognize what it is looking for. A number or weight represents the connection between one unit and another, which can be positive or negative. The higher or lower the weight, the more or less influence one unit has on another. In an ANN the data is fed forward through the network layer-by-layer, until it reaches the final layer, and it is only when the data reaches the final layer’s activations that the network’s predictions are made.

How is an Artificial Neural Network or Deep Learning System Trained?

ANN’s or Deep Learning Systems can be trained in one of three ways[3]:

  1. From Scratch

This is by far the most time consuming method and involves gathering a very large data set containing metadata and the design of a network architecture that will learn the features and model. This approach is used for new applications, or applications with a large number of output categories. Because of the amount of data and learning involved, this method of training the network can take days or weeks to fully train the model.

  1. Transfer Learning

This involves tuning a pre-trained model and is the more common approach and much faster than #1. Most deep learning applications use the transfer learning approach, a process that involves fine-tuning a pre-trained model. You start with an existing network, and feed in new data containing previously unknown classes. After making some tweaks to the network, you can now perform a new task.

  1. Feature extraction

While less common than the other two more specialized approach to deep learning is to use the network as a feature extractor. Since all the layers are tasked with learning certain features from the inputs, we can pull these features out of the network at any time during the training process and these features can then be used as input to a machine learning model.

There are two modes of learning in an ANN: 1) Supervised learning where the network is trained using a set of input-output pairs. The goal is to’ teach’ the network to identify the given input with the desired output; and 2) Unsupervised learning whereby the network is trained using input signals only and the network organizes internally to produce outputs that are consistent with a particular stimulus or group of stimuli.

Where are Artificial Neural Networks and Deep Learning Systems Being Used Today?

There is a host of applications in production, as well as in active research as proof of concepts, or being contemplated as future possibilities across healthcare and life sciences, and we have listed several below to provide perspective on the opportunity for ANN’s in this space, however the list is by no means exhaustive:

Disease Identification and Diagnosis[4]

  • In radiology for disease identification and diagnosis, deep learning systems are trained to detect the presence or absence of disease in medical images and from unstructured text in radiology reports, helping doctors come up with better interpretations

Personalized Medicine

  • In personalized medicine to treat cancer patients, establishing standards of care and cancer treatment recommendations based upon the latest medical research literature, evidence based medicine, in combination with the patient diagnosis and medical history
  • Matching patients based upon their diagnosis, medical history and other factors to the optimal clinical trials available locally and nationwide

Drug Discovery and Manufacturing

  • Remote monitoring and real-time data access for increased safety; such as monitoring biological and other signals for any sign of harm or death to participants
  • Early stage drug discovery e.g. from initial screening of drug compounds to predicted success rate based on biological factors

Predicting and Managing Epidemic Outbreaks

  • Monitoring and predicting epidemic outbreaks based on data collected from satellites, historical information on the web, real-time social media updates, and other sources

 

In summary, Artificial Neural Networks and Deep learning are key technologies to achieving results never before seen, allowing us to realize a host of use cases that span research, drug development, diagnosis, treatment, prevention, patient safety and beyond. It is the key to voice control in consumer devices like phones, tablets, TVs, and hands-free speakers enabling speech based and other natural language inputs and structure data interactions.

Training neural networks requires a great deal of data, and that is just to allow them to recognize known features. Beware that not all applications are well-suited to ANN’s or deep learning, so it is important to understand when to leverage ANN’s and Deep Learning, verses other type of Machine Learning techniques to achieve the desired results. Also don’t underestimate the effort, particularly when training an Artificial Neural Network from scratch.

The list of opportunities is for ANN’s and Deep Learning is endless, and innovators across healthcare (payer and provider) and life sciences (pharma, biotech and medical device manufacturing) are beginning to invest in this area to achieve state-of-the-art accuracy, with many hoping to exceed human-level performance.

Sources:

  1. Robert D. Hof, MIT Technology Review 2013. Deep Learning: https://www.technologyreview.com/s/513696/deep-learning/
  2. Rohan & Lenny #1: Neural Networks & The Backpropagation Algorithm, Explained, March 3rd, 2016: https://ayearofai.com/rohan-lenny-1-neural-networks-the-backpropagation-algorithm-explained-abf4609d4f9d
  3. https://www.mathworks.com/discovery/deep-learning.html
  4. Deep Learning Applications in Medical Imaging, September 14, 2017 by Abder-Rahman Ali: https://www.techemergence.com/deep-learning-applications-in-medical-imaging/

 

 

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.

Subscribe to the Weekly Blog Digest:

Sign Up