Digital pathology has become widespread across the world in more than just academic labs. Contract research organizations (CROs) and pharmaceutical companies reap the benefits of digitization due to the nature of their work which is managing and analyzing millions of samples while seeking to reduce cost and time to develop viable drug candidates. Clinical labs on the other hand find it particularly helpful for producing quantitative results and taking part in remote diagnostic consultation.
The benefits of digital pathology
The benefits of digital pathology are manifold, the following graph summarizes these for the general pathology lab. These benefits are applicable to all areas and industries including academic research, preclinical labs of pharmaceutical companies and CROs, as well as the clinical pathology laboratory.
How to use AI for image analysis in pathology
Deep learning (DL) is a subset of machine learning (ML) which falls under the category of artificial intelligence (AI). The formation of layers is not only pervasive in the nomenclature of these different types of computer intelligence but also in their architecture. DL is based on artificial neural networks, a layered and connected system of algorithms receiving and processing information.
Artificial neural networks (ANNs) are what form and drive deep learning. They are computing systems designed to find patterns that are too complex to be manually taught to machines to recognize. Hence why deep learning is so adept at image analysis and in some regards more powerful than machine learning.
What are neural networks and how do they work?
ANNs are made up of a collection of connected units of mathematical functions, referred to as artificial neurons. In deep learning models the neurons can range in amount from dozens to millions of units always arranged in a series of layers. The neurons are connected to each other between the different layers with a series of connectors, called weighted connections.
A very simplified explanation of how the layers work together when processing information through deep learning is broken down here:
The input layer is fed information or data to process and learn from
It transmits its output to the subsequent layers called the hidden layers
The output layer is ultimately responsible for producing the final result
The information being processed in neural networks can travel through the layers, neurons, and connections in different ways: only in one direction from input to output, or in multiple directions, and more permutations of these. These ways differ between different types of ANNs. We will now dive into the type of neural network deployed by Aiforia: convolutional neural networks.
How to train AI for image analysis
Pantanowitz L. Digital images and the future of digital pathology. J Pathol Inform 2010.
Pantanowitz L. Digital Pathology. In: Pantanowitz L, Parwani AV, editors. Chicago, USA: ASCP Press;