AI unbounded

Convolutional neural networks exceed human capability in image analysis and can be trained for any medical or pathology task with the Aiforia AI software.
Written by Aiforia

What type of artificial intelligence do we offer at Aiforia and how does it perform image analysis?

Many companies and organizations are claiming they use or offer solutions based on artificial intelligence (AI), machine learning (ML), or deep learning (DL). Note, the latter two are subtypes of AI. This hype has reached a point that many of you might be desensitized to these terms, even skeptical of what AI really is and who is actually offering “it”. 

I would like to be transparent and describe to you what the Aiforia AI offering is and how it performs image analysis. If you want to delve into the definitions of AI. For now, I will just focus on the Aiforia AI: deep learning. 

The layers run deep

Deep learning is a subset of machine learning and often referred to as the next generation of ML because it involves, very simply put, more math, accuracy, and computing power. As you may have come to realize the world of computer intelligence involves increasingly deeper layers of different subtypes and architectures. The taxonomy runs deep. So let’s dive into the Aiforia deep learning AI. 

Our AI is built as a convolutional neural network (CNN) — more abbreviations for you to remember. CNNs are formed of multiple layers of mathematical functions. These functions process and calculate inputs and outputs, such as pixels in an image. 

Pixel based analysis 

In image analysis a CNN processes pixels differently at each layer, starting with more basic features at the first layer and working on increasingly complex features the deeper the layers are situated. CNNs are incredibly powerful at analyzing images and are able to perform tasks such as image recognition, object detection, and segmentation. 

These neural networks interpret images through different processes — I told you there are a lot of taxonomic layers in the world of AI. The Aiforia CNN mainly uses semantic segmentation. This is a process of linking each individual pixel in an image to a class label. This is incredibly technically complex but also the most fruitful form of image analysis. 

Quality over quantity is the mantra of this type of image analysis as each individual pixel of an image is considered and evaluated to produce a result. Therefore eliminating the need for many more images to train deep learning AI models, as each image used is reviewed thoroughly. Aiforia also offers an active learning tool, Annotation Assistant, which reduces the amount of training images even further. 

More networks, more knowledge

Then the explanation can go deeper, there are different structures to how semantic segmentation is formed or works. There are also different ways in which these networks can learn. However for sake of sanity I will not delve into that here. We will open up these networks in another article.  

Hopefully you now have an understanding of the Aiforia AI, how it is formed, and how you can utilize it for image analysis. If you are looking for more AI knowledge, visit our resources here.