Developing a metrological framework for assessment of image-based Artificial Intelligence systems for disease detection
Short Name: MAIBAI, Project Number: 22HLT05
Using imaged-based AI software for improved cancer diagnostics
It is estimated that cancer incidence will increase by 21 % by 2040. Preventative screening is integral to the early detection and treatment of many forms of this disease. However, this process is labour-intensive, costly, and time-consuming.
Machines use Artificial Intelligence (AI) systems to make decisions that usually require human intelligence. In a healthcare setting, AI systems possess the potential to automate processes by making diagnostic or risk assessment decisions, greatly decreasing the time required to identify cancerous tissues and help to handle future increased demand on the healthcare system. The survival rate after a cancer diagnosis varies significantly between European countries, and by using AI systems for cancer diagnosis human subjectivity could be reduced.
Currently, however, there are no standardised guidelines for evaluating AI systems and most suffer from significant methodological weaknesses such as limited data sets and lack validation procedures, hindering their use in a diagnostic setting.
This project will design an imaged-based AI programme that will be used for diagnostic breast cancer screening.
As part of this goal, the needed technological infrastructure to access information from a range of databases will be developed and the AI software will be programmed to be able to differentiate between clinical subgroups. Ideally, this AI software will be used in Europe as a standardised diagnostic method and remove some human subjectivity that currently affects cancer diagnosis.
The development of thorough and robust validation protocols for AI systems is essential to ensure that high quality healthcare is accessible across Europe.