Project short cut: IPA-211–AI-ICOERD
Viability study for the development of deep learning-based artificial intelligence (AI) to support standardized pneumoconiosis classification in X-ray and CT images of the lungs. The aim to support the complex diagnosis and classification of dust-related lung diseases such as asbestosis and silicosis and to faciliate established procedures.
The number of pneumoconiosis cases in follow-up care remains relevant. At the same time, the number of radiologists with the relevant expertise and willingness to participate in imaging is declining.
In collaboration with the Bioinformatics Department at Ruhr University Bochum and Gesundheitsvorsorge (GVS), research is being conducted to determine whether AI can be developed and trained to automatically perform standardized pneumoconiosis classifications for X-ray and CT images of the lungs.
The basis for this is approximately 60,000 X-ray and 15,000 CT images from the GVS database with associated ILO or ICOERD classifications, in particular high-quality image-findings pairs from a group of experienced second evaluator. The completely anonymised data is used in a protected, locally restricted area for AI development. The results are then checked against a validation data set with findings from experts.
