The AI in Medical Image Analysis (AIM) group, led by Professor Marleen de Bruijne, develops novel techniques for quantitative analysis of medical images, with a focus on machine learning – and especially deep learning – techniques and on large-scale image-based studies. An important theme is the development of machine learning techniques to predict disease directly based on imaging data. Using prediction models derived from a database of images for which the diagnosis has already been established or for which the future course of the disease is known from clinical follow-up, such techniques are more widely applicable and often more sensitive and robust than conventional image analysis methods. Another important theme is the development of robust and fair image analysis models based on clinically realistic situations. Machine learning techniques often work well on large, well-curated, fully annotated datasets, but how do we learn reliable models if datasets are small or heterogenous and have few, weak, or noisy annotations?
Currently our main application areas are in neuro-, vascular-, and pulmonary image analysis.
