AI system that mimics human gaze could be used to detect cancer

Cardiff University scientists have developed an advanced artificial intelligence (AI) system that can accurately predict which parts of an image a person is most likely to look at.

Based on the mechanics of the human brain and its ability to distinguish between different parts of an image, the researchers say the new system displays more accurately human vision than all the foregoing.

Applications of the new system range from robotics, multimedia communications and Camera surveillance to automated imaging and finding tumors in medical images†

Cardiff University’s Multimedia Computing Research Group now plans to test the system by helping radiologists find lesions in medical images, with the overall aim of improving the speed, accuracy and sensitivity of medical diagnostics.

The system has been presented in the magazine neurocomputing†

Being able to focus our attention is an important part of the human visual system that allows us to select and interpret the most relevant information in a given scene.

Scientists around the world have used computer software to try to recreate this ability to distinguish the most eye-catching parts of an image, but so far with mixed success.

In their study, the team used a deep learning computer algorithm known as a convolutive neural network designed to mimic the interconnected web of neurons in the human brain and modeled specifically on the visual cortex†

This type of algorithm is ideal for taking images as input and for assigning importance to different objects or aspects within the image itself.

In their study, the team used a huge database of images in which each image had already been rated or viewed by humans and assigned to so-called “areas of interest” using eye-tracking software.

These images were then fed into the algorithm and using a type of AI known as deep learningthe system slowly began to learn from the images to a point where it could then accurately predict which parts of the image stood out the most.

The new system has been tested with seven state-of-the-art visual saliency systems already in use and shown to be superior on all metrics.

Study co-author Dr. Hantao Liu, of Cardiff University’s School of Computer Science and Informatics, says that “this study has shown that our advanced system, which uses the latest advances in machine learningis superior to the existing state-of-the-art visual saliency models that currently exist.”

“Being able to successfully predict where people will look in natural images could unlock a wide variety of applications, from automatic target detection to robotics, image processing and medical diagnostics.”

“Our code has been made freely available so that anyone can take advantage of the research and find new ways to apply this technology to real-world problems and applications.”

“The next step for us is to work with radiologists to determine how these models can help them detect lesions in medical images.”

The code underlying the new system has been made freely available to the public and can be downloaded from GitHub.