It’s no secret that artificial intelligence is playing a bigger and bigger role in the automation of routine business tasks. And deep learning is a big part of this.
In 2014, Google spent over $500 million to acquire DeepMind Technologies, an AI startup working on deep learning software. Many businesses around the world need to decrease operational costs and because of this, the demand for deep learning technology is increasing.
Even the medical industry is investing in deep learning technology. One of the areas where deep learning can make a huge impact is in the field of whole slide image analysis.
Pathologists spend a lot of their time making a diagnosis by carefully analyzing digital images. But what if computers could do it for them? Keep reading to find out more about the role of deep learning in digital pathology.
What Is Deep Learning?
Deep learning is part of machine learning – the branch of artificial intelligence focused on performing routine tasks and learning through data.
Machine learning involves the use of computer algorithms that learn through data and increase their accuracy over time. Machine learning software underpins many of the apps and devices we use on a day to day basis.
Typical machine learning algorithms are excellent at solving many problems. However, they are poor at dealing with certain kinds of data – especially images. That’s where deep learning comes in.
Deep learning is a subset of machine learning that uses artificial neural networks to analyze complicated data and solve problems. an artificial neural network is a type of software architecture modeled on the human brain.
An artificial neural network (ANN) consists of a number of interconnected nodes (‘neurons’). These nodes are arranged in layers. The first layer is the input layer and it receives the data (such as an image).
This information is then passed through a secondary layer, called the ‘hidden’ layer before it reaches the final layer, called the ‘output layer’. The output layer then tells the computer what to do in response to the data.
The subset of deep learning focused on image analysis is called ‘computer vision’.
Deep Learning and Computer Vision
Computer vision is one of the hottest areas of AI research. Teaching computers to make sense of photos and digital images is a problem scientists have been grappling with for a while. But recently, they have begun to make real progress.
There is a specific kind of ANN used for image analysis. It’s called a convolutional neural network (CNN). A CNN is also composed of layers of neurons but each layer detects different things within the image.
The first layer might detect basic things like edges. Deeper layers in the network detect things like shapes and corners. The output layer of the CNN then estimates the probability that a certain thing was found in the image.
Deep learning and computer vision have the potential to revolutionize the medical field of pathology. How exactly? By automating whole slide image analysis.
You see, pathologists diagnose cancer by analyzing digital images of a patient’s tissue. This type of imaging is called whole slide imaging and it replicates what a pathologist would see under a microscope.
What Is Whole Slide Image Analysis?
First, it’s important to understand what a whole slide image is. Whole slide images are high-quality, digital images, created by scanning a glass microscope slide. A whole slide image scanner is used for this.
The digital image depicts a patient’s tissue sample. This is then sent to a pathologist for review and diagnosis. This type of tissue analysis is the only definitive method for cancer diagnosis and prognosis.
Pathology labs all across the world have adopted digital whole slide imaging.
Deep Learning and Digital Pathology
It takes considerable time and energy for pathologists to analyze these images. Due to the large size of these images and the increase in cancer rates, it’s become important to create an automated solution.
An early cancer diagnosis significantly increases the probability of survival. It’s for this reason that scientists are working to create a deep learning framework for whole slide image analysis.
An effective AI model that can accurately analyze and interpret histological images, will save lives.
Moreover, because of the difficulty of the task, pathologists often misdiagnose. One study found that pathologists disagree with each other on diagnosis almost 25% of the time.
This further stresses the need to create accurate AI models that can assist pathologists.
In recent times, there has been huge progress made in building deep learning systems capable of whole slide image analysis. However, there are still a number of challenges that researchers must overcome.
Whole slide images are typically very large and require a lot of preprocessing. This can slow down the process. Training the ANN to analyze whole slide images can also take a long time. ‘Training’ is when large amounts of data are fed into the ANN – this is how it learns.
Lastly, Deep learning models must be able to complete a number of different pathological tasks. This makes it challenging to build an AI framework for histopathological analysis.
There are two main kinds of analysis. The first is classification and involves classifying whole slide images as benign or malignant. The second kind is segmentation and involves detecting and separating normal cells from tumor cells.
A successful deep learning model will need to perform both of these tasks with reasonable accuracy. This would aid pathologists greatly and increase the chance of an accurate diagnosis.
Revolutionizing Pathological Analysis
The use of deep learning models for conducting whole slide image analysis is still a relatively new field. It’s extremely exciting to ponder how this new technology is going to transform the healthcare industry.
There’s no doubt that digital pathology is the future. If you’re interested in learning more about whole slide imaging systems, you can find more information here on our website, or contact us today.
