Medical Image Analysis

Medical image analysis takes place at the intersection of image acquisition, processing, analysis and interpretation. While our main focus is on the modelling and analysis of data, this can only be done through a detailed understanding of the image data and the underlying anatomical context. Our vision is to push the frontiers of integrated image acquisition, modelling and its interpretation. Our research is grounded in industrial, clinical and societal needs.

 

Interpretability, transparency and safe AI

Tractography based on diffusion MRI is a non-invasive way of visualisation the brain network in the human brain and can be used to predict the impact of disease. However, tractography is a model-based approach meaning that connections are not guaranteed to be correct. Together with an international team we created a physical brain phantom of the human brain and made a tractography challenge. In total 96 distinct submissions applied their tractography pipeline to the synthetic diffusion MRI data set of the human brain. The tractography results provided unique insights into how the various steps in the data processing can influence reliability of the tractography results. Essentially, we obtained clearer insights into the challenges of today tractography methods when tracking brain connections through complex fibre regions. Published in Maier-Hein et al. 2017, Nature Communication

Predictive models are an integral part of medical image analysis, where they currently form the backbone not just for image guided diagnosis and prognosis, but also for image processing tasks such as denoising, segmentation and registration. Given that the output returned from these models are used to improve diagnosis and treatment, as well as to enhance our anatomical knowledge, it becomes crucial for these models to also come with information on their own limitations. We develop methods for quantification of uncertainty, interpretability, and for detecting bias and enhancing fairness in predictive models for medical imaging.

Shape and appearance modelling

The statistical modelling of shape and appearance was pioneered by the active shape and appearance models. This has been and is still a very active research topic, where recent developments within deep learning has further increased the potential in these methods. We are using these methods in different fields including modelling of the human face, ear and skull for hearing aid design. Another example is shape modelling of the human cochlea and the surrounding structures for surgical planning. A major research theme is also the modelling of the human heart for risk assesment and intervention planning.

Geometric machine learning

Recently, deep learning has been applied to non-structured 3D geometries as for example surface meshes and a new area of research named geometric deep learning has emerged. We have been using methods from geometric deep learning to analyse facial geometries, heart anatomies and are also developing methods for the analysis of the human inner ear system.

 

We develop methods for analyzing data that respects known invariances or constraints, for example encoded in a Riemannian manifold structure (Mallasto et al, CVPR'18 and AISTATS'19).

Cross-disciplinary applications

A highly cross-disciplinary teamwork resulted in the development of the Carbo-gel: A multimodal soft tissue marker that bridges the gap between high-resolution diagnostic imaging and therapeutic intervention. During the project, medical image analysis was used to quantify and validate the behavior, e.g. efflux of solvent, surface volume, gel degradation and image contrast.

Paper in Science Advances, 2020

Microstructure imaging - Biophysical modelling and validation

 

Diffusion MRI combined with biophysical modelling enable indirect insights into tissue microstructural features along brain connections such as axon diameters, cells sizes, shapes and density - although the low image resolution of MRI. To improve and validate our underlying biophysical model-based assumptions to estimate axon diameters we need direct 3D measurements of tissue microstructure in ultra-high image resolutions. In recent work, we took a tissue sample from a monkey brain being whole-brain MRI scanned and mapped the estimates of axon diameters across the Corpus callosum (A). We also used tractography to map the interhemispheric V1/V2 brain connection projecting through the genue of corpus callosum (green streamlines). A tissue sample from the monkey brain (B) was 3D scanned with phase-contrast synchrotron imaging in ultra-high image resolution at the ID16A beamline at the large-scale ERSRF synchrotron facility, Grenoble, France (C). 3D analysis of axon segmentations from synchrotron data shows that axons have a complex morphology and are not like cylinders as thought today (C). Surprisingly, we find that axon diameter variation is introduced by neighbouring structures physically pushing on the axons such cells clusters (blue), vacuoles (green), blood vessels and other axons (D). Finally, we speculated how our findings can impact the signal conduction properties along brain connections known as the structure-function relationship. Published in Andersson et al., 2020 PNAS

Predictive models are an integral part of medical image analysis, where they currently form the backbone not just for image guided diagnosis and prognosis, but also for image processing tasks such as denoising, segmentation and registration. Given that the output returned from these models are used to improve diagnosis and treatment, as well as to enhance our anatomical knowledge, it becomes crucial for these models to also come with information on their own limitations. We develop methods for quantification of uncertainty, interpretability, and for detecting bias and enhancing fairness in predictive models for medical imaging.

Contact

Aasa Feragen
Professor
DTU Compute
+45 26 22 04 98
Anders Nymark Christensen
Associate Professor
DTU Compute
+45 45 25 52 58
Rasmus Reinhold Paulsen
Associate Professor
DTU Compute
+45 45 25 34 23
Tim Bjørn Dyrby
Associate Professor
DTU Compute
+45 45 25 34 24