SANTA CRUZ DE TENERIFE, Dec. 19 (EUROPA PRESS) –
The Pathological Anatomy Service of the Complejo Hospitalario Universitario de Canarias (HUC) and the Instituto de Astrofísica de Canarias (IAC) collaborate in the development of a computer tool that adapts artificial intelligence methods applied in astronomy to digitized images of human samples.
The project, called ‘Patolog-IA’, aims to speed up the interpretation of results and the diagnosis of colorectal cancer. It is anticipated that it could also be used in precision personalized medicine targeting other types of cancer.
Colorectal cancer is one of the most aggressive and the most frequently diagnosed in Spain in 2021. Its early detection is critical and for this reason samples are taken systematically, but the interpretation of the results is slow and complex. Analyzes are performed manually by expert pathologists with more than ten years of training, which takes a significant amount of time and limits the pace of diagnoses.
The problem is especially pressing in outermost regions, where the technical means for taking samples exist, but there is a lack of human means for their interpretation. The shortage of pathologists to study the large number of samples generated and the need to prioritize the most important prompted the HUC Pathology service to contact the IAC, where there is a group specialized in the application of artificial intelligence methods to astronomical images. .
An initial problem is the enormous size of the digitized images of the human samples, with hundreds of millions of pixels, equivalent to more than a dozen high-definition televisions. The typical images from astronomical surveys are usually smaller, although very numerous. A common problem in astrophysics and the diagnosis of pathologies is the lack of already classified images, since it entails enormous manual work by experts. Based on previous research results, the team identified a technique that solves both problems.
AUTOMATIC LEARNING
Thanks to the availability of previously published methods, codes and even training data from jobs, it has been possible to develop an artificial intelligence model based on the ‘Middle Master’ technique. In this technique two identical and very deep neural networks are combined. A network acts as a ‘student’, and with a few labeled images it learns the characteristics to detect tumors. The other network acts as a ‘teacher’, receiving some of the information from the student and trying to classify the unlabeled images.
In the first attempt the result is random, but iteratively, cycle after cycle, it improves its predictions. As the original images are too large, they are divided into thousands of more manageable segments, suitable for the memory of the graphics cards used in the analysis. “In the end, a network is obtained that is capable of identifying whether an image is a tumor with high reliability and with very few false positives,” said Carlos Westendorp, an astrophysicist at the IAC.
According to market.us, The Artificial Intelligence Market is projected to grow exponentially over the next decade, and it is expected to grow around USD 2,967.42 billion by 2032 from USD 94.5 billion in 2021 at a 36.8% CAGR.
A computer server has already been installed, which incorporates a graphics card donated by NVIDIA to the IAC, for the exclusive use of the HUC. The service is still in the testing phase, but soon the images obtained daily at the hospital will be able to be uploaded to the server and it will return a prioritized list to be reviewed by doctors. “The final diagnosis will be made by a specialized pathologist, but the most urgent cases will be identified more quickly thanks to the help of artificial intelligence,” stressed Carlos Luque, a researcher at the EuroCC project in which the development of this tool is framed.
The current design has been carried out in a period of just a few months thanks to the fact that the tissues have common characteristics for the entire human species. “In the future we want the network to pay more attention to specific details identified by the HUC,” said Andrés Asensio, a researcher at the IAC. “As this technique can also be applied to other data, such as astronomical data, we believe it is a great example of the benefits of knowledge transfer between different fields linked by artificial intelligence techniques,” he added.
Carlos Allende, the coordinator of the project at the IAC, advanced that this program “will continue with funding from the Carlos III Health Institute for personalized precision medicine aimed at breast cancer, and also within IACTEC, the IAC space dedicated specifically to technology transfer, under the name of Patolog-IA”.
