Objectives
The overall objective of MAIBAI is to develop a metrological framework necessary to support standardisation in image-based AI systems for disease detection. Using breast screening as an exemplar, the performance of explainable and traceable AI tools will be analysed on a large real-world database of mammographic images, informing the design of the standardised assessment framework.
To develop a technical infrastructure to be able to query and extract the relevant data from medical imaging databases, such as the OPTIMAM Mammographic Image Database (OMI-DB). To establish a methodology for centralised and common metadata indexes, creating an open-source middleware for translation of metadata from different imaging databases.
To identify image acquisition key factors and population subgroups, and use these to categorise the clinical data into subsets, determining where there is sufficient data for training and validation of AI tools for disease screening. To develop a methodology to generate synthetic data derived from physics-informed models and data augmentation techniques based on measurement knowledge.
To use explainable and traceable AI tools for disease screening, providing the capability to train and retrain the tools as necessary. To benchmark the AI tools in terms of prediction performance, robustness, fairness and uncertainty quantification, under at least three scenarios, including low versus high image quality data, validation for specific patient demographics, presence of machine-based artefacts and noise sources. To develop and validate methods for the explainability and interpretability of the trained AI tools.
To provide an AI validation toolbox for diagnostic imaging, to summarise the performance testing evaluations and to give recommendations for the assessment of explainable and traceable AI tools for disease screening, with a focus on understanding their generalisability and sensitivity to varying populations, manufacturers, image processing, and acquisition techniques. To use the recommendations to design a global, standardised, and impartial AI assessment framework.
To facilitate the take up of the technology and measurement infrastructure developed in the project by the measurement supply chain, standards developing organisations (British Standards Institution, ISO/IEC JTC 1/SC 42 – Artificial intelligence), and end users (e.g., clinical stakeholders, manufacturers of medical and healthcare products, regulators).
