Get In Touch
Xeraya Capital Sdn Bhd
26.03–26.08, G Tower
#199 Jalan Tun Razak, KL
+603 2381 8700
Submit a Proposal

Connecting Humans to Machines – What You Should Know About Deep Learning Models.

We often take for granted advancements in technology that make it easier for us to use the tool and devices present in our everyday lives. One such innovation is the development of Deep Learning Models. In this article, Xeraya explores the definition of a Deep Learning Model, the top five models most in use today, and takes a brief look at how these models are helping to shape the work of healthcare and life sciences players today.

Deep Learning Models – What Are They?

In the realm of scientific computing, deep learning has been growing in popularity, and its algorithms are now used by a wide variety of industries to solve complex problems. Each deep learning algorithm uses different types of neural networks to perform specific tasks.

In essence, deep learning models use artificial networks to perform complex computations on large quantities of data. As a subset of machine learning, these models work based on the structure and function of the human brain.

As mentioned above, they are utilised to perform specific tasks – repetitive and routine jobs – that they can do more efficiently than the average human being. Deep learning models also take this one step further by guaranteeing the quality of the work executed for these tasks and provide key insights for future learning. Companies that have embraced deep learning have been able to cut both costs and labour time, which frees up their employees to perform creative tasks that require a human touch.

Undoubtedly, deep learning has become a disruptive piece of technology; with the increasing adoption of automation tools – such as robots, IoT, cybersecurity, industrial automation, and machine vision – by organisations, the number of data being generated has grown exponentially and the best way to manage all this data is to use deep learning models. They use this information to serve as training modules and to create algorithms, which in turn are used to diagnose issues and test purposes.

The most accurate deep learning algorithms can learn from past experiences and create a consolidated data environment, spurring better data management. These models rely on data, so the more data there is, the more accurate the results will be. While we are still figuring out the ways deep learning can be used in different situations, they are currently being utilised for machine translation, chatbots, and service bots.

No alt text provided for this image
The various types of machine learning technology available, categorised by unsupervised, supervised, and reinforced learning.

The Top 5 Deep Learning Models

1. Convolutional Neural Networks (CNNs)

Developed by Yann LeCun back in 1988 (when it was named LeNet), CNNs or ConvNets were initially used to recognise characters such as ZIP codes and digits. Consisting of multiple layers, CNNs today are mainly used for image processing and object detection.

2. Long Short-Term Memory Networks (LSTMs)

LSTMs (a type of Recurrent Neural Network or RNN) possess a chain-like structure, with four interacting layers communicating in a unique way. They retain information over time and are useful for time-series prediction scenarios because they recall previous inputs, thus meaning that they can learn and memorise long-term dependencies. Because the default behaviour of LSTMs is to remember past information for long periods of time, they’re used for speech recognition, music composition, and pharmaceutical development.

3. Recurrent Neural Networks (RNNs)

RNNs have directed cycles that are formed with connections, allowing outputs from the LSTMs to be used as inputs for the current phase of the process. RNNs are mostly used for image captioning, time-series analysis, natural-language processing (NLP), handwriting recognition, and machine translation.

4. Generative Adversarial Networks (GANs)

A neural network that has becoming increasingly utilised in several systems over time, GANs are deep learning algorithms that create new data instances to resemble training data. There are two components in GANs: a generator which generates fake data, and a discriminator, which learns from this fake information.

At present, GANs are used in several sectors; in astronomy, they’ve been used to improve images of space and the simulate gravitational lensing for dark-matter research. In entertainment, game developers have been using GANs to improve low-resolution graphics, and to convert 2D textures from older video games into 4K or higher resolutions via image training. Animation players have also been using GANs to generate realistic images, cartoon characters, photographs of human faces, and to render 3D objects.

5. Radial Basis Function Networks (RBFNs)

RBFNs are special types of feedforward neural networks that use radial basis functions as activation functions. They have an input layer, a hidden layer, and an output layer and are mostly used for classification, regression, and time-series prediction.

RBFNs consist of an input layer, a hidden layer, and an output layer and are a special type of feedforward neural network. They’re most used for classification, regression, and time-series prediction.

The Market Size of Deep Learning Models

Thanks to the advancements in data centre capabilities and high computing power as well as their ability to perform tasks without relying on human input, the global deep learning market hit a valuation of US$49.6 billion in 2022 and has a CAGR of over 33.5% for the period 2023-2030. Fuelling its growth is also the rapid adoption of cloud-based technology across different fields, including healthcare.

No alt text provided for this image
The US/Global market size of deep learning models.

Applications in Healthcare and the Life Sciences

Over the past five years or so, the application of machine learning technology in healthcare has grown; deep learning models have gained a lot of attention, mainly due to the growing complexities of healthcare data. Machine learning has given healthcare professionals a way to create efficient and effective data analysis models which can help uncover hidden patterns and other meaningful information in a relatively faster amount of time in comparison to conventional models. Most importantly, they help foster pattern recognition in healthcare systems.

With the rise of personalised medicine, analysing healthcare data has become crucial. A good example is when an oncologist wants to form a personalised treatment plan for their patient. Data such as genomics variants, environment, imaging genomics, current drugs, and lifestyle must all be considered to create the right plan, and this is where technology, such as imaging and life monitoring has become very useful. However, despite the huge amount of data these methods have been able to generate, healthcare professionals still need to glean a better understanding of diseases and how to treat patients.

This is where deep learning models come in, specifically in data analysis. When implemented in diagnostic tools, these models help identify the associations between all the different types of patient data and draw patterns and insights physicians can use to decide the course of treatment.

No alt text provided for this image
Some of the use cases in healthcare that researchers have found for deep learning models.

Now, healthcare players are using the computing capability of deep learning models such as computer vision, natural language processing, and reinforcement learning in a few key areas:

Patient Care

1.     Medical Imaging

Image recognition and object detection are both used in processes for Magnetic Resonance (MR) and Computed Tomography (CT), specifically for image segmentation, disease detection, and prediction. Thus far, deep learning models have helped with diabetic retinopathy detection, early detection of Alzheimer’s, and ultrasound detection of breast nodules.

2.     Healthcare data analytics

As mentioned above, personalised healthcare is becoming a key catalyst for the future of patient care. Deep learning models are thus being used to analyse electronic health records, including clinical notes, laboratory test results, diagnosis, and medications at superhuman speeds and the best possible accuracy.

Smartphones and wearable devices with health monitoring capabilities also fall into this category. At present, there are devices available to patients and their physicians in the market that measure a person’s vitals such as their pulse, respiration, oxygen saturation, temperature, and mobility.

3.     Mental health chatbots

COVID-19 brought mental health into the spotlight and started a movement towards interactive and online mental health consultation services that is showing no signs of slowing down two years later. Some of these services are now exploring the use of deep learning models to create more realistic conversations with patients.

Research and Development

1.     Drug Discovery

In this space, deep learning algorithms have been helping researchers to identify viable drug combinations by processing genomic, clinical, and population data at a rapid rate. The researchers are then able to focus on patterns in these large data sets and discover new drugs as well as predict interactions.

2.     Genomics Analysis

The use of deep learning models can increase the interpretability of biological data; these models essentially help scientists in their study of genetic variation and genome-based therapeutic development. CNNs are widely used because they can obtain attributes from fixed-size DNA sequence windows.

3.     Mental Health Research

To help improve clinical practice in the realm of mental health, academicians and researchers alike are using deep neural networks (a.k.a. deep learning models) to understand the impact of mental illness and other disorders on the brain. They are also using deep learning algorithms in their work to determine meaningful brain biomarkers.

Deep Learning Models: An Inevitable and Essential Component of the Future

There continues to be a plethora of use cases and evidence that speaks to the importance of using deep learning models in healthcare and other industries moving forward. Challenging issues, such as disease detection, feature extraction, and personalised courses of treatment are just some of the problems deep learning models can help with. In the future, it is likely that researchers will also try their hand at developing and integrating different types of efficient technologies to hardware requirements of decision-making systems, including those with a deep neural network structure. All in all, machine learning and deep learning models have tremendous potential in the healthcare sector, especially given how data in this industry continues to grow in both size and complexity.