An artificial intelligence system capable of detecting possible tumors by imitating the human gaze.

The premise from the beginning of the research is the human ability to distinguish different parts of an image, so a group of researchers from Cardiff University in Wales used, in this research, a deep learning computer algorithm, better known as a convolutional neural network, designed to simulate the interconnected network of neurons in the human brain and is modeled specifically on the visual cortex.

The algorithm is designed to take images as the basis of an analysis and assign hierarchy of importance to specific aspects or points within the same image.

Among the various scenarios where this technology could be used, is the detection of possible tumors in medical images, which is why the Multimedia Computing Research Group, belonging to the same university, is working on proving that the system could help radiologists find lesions in medical images, improving the speed and accuracy of diagnoses.

The study worked with a huge image bank, where each image had already been previously analyzed or simply viewed by humans, where the various areas of interest were identified using eye-tracking software.

The images were then integrated into the algorithm, and using the deep learning AI system, the system began slowly learning details in the images, until it acquired the ability to accurately predict the highlights of the image.

To prove its effectiveness, this new system was tested against seven other similar state-of-the-art systems in use. The new system demonstrated its superiority in the metrics obtained.

Dr. Hantao Liu, from Cardiff University's School of Computer Science and Informatics, commented that this new AI is "able to successfully predict where people are looking in natural images could unlock a wide range of applications, from automatic target detection to robotics, image processing and medical diagnostics."

The team continues to develop and improve the system to materialize the medical injury aid, moreover, the code used is freely available. In Dr. Liu's words "so that everyone can benefit from the research and find new ways to apply this technology to real-world problems and applications".

You can access the source code used for free at https://github.com/LJOVO/TranSalNet 

And read more about the study at https://www.sciencedirect.com/science/article/pii/S0925231222004714?via%3Dihub