Rasmus R. Paulsen

Title: Neural implicit representations of complex 3D anatomies.

Abstract: It is not trivial to harness the power of modern deep learning methods on complex 3D data. In order to feed the data into a network a good representation of the data is essential. For certain types of 3D data, the interface between organs represented as surfaces is a good approximation of the physiology. An example is the cardiac chambers where the blood pools can be approximated using a bounding surface (The left ventricle shape can for example be seen as where there is contrast filled blood). The typical, triangulated surface, is however very difficult to use in deep learning frameworks. Methods exists but have hard limitations on the surface topology (for example constrained to be spherical). In this talk, I will focus on how surfaces can be represented using neural implicit descriptions, where the signed distance field is a commonly used representation. Implicit representations come with both benefits and pitfalls – not being rotationally invariant being one. Having effective and information preserving data representations opens up many advantages seen in other fields of deep learning for example generative modelling. I will demonstrate a few applications from the cardiovascular domain.

Related papers

• K. Sørensen et al. NUDF: Neural Unsigned Distance Fields for High Resolution 3D Medical Image Segmentation. 2023.
• K. Sørensen et al. Implicit Neural Distance Representation for Unsupervised and Supervised Classification of Complex Anatomies. 2023

Vedrana Anders Dahl

Title: Investigating the intersection between deep-learning-based segmentation and synchrotron 3D imaging

Abstract: Despite remarkable advances in deep-learning-based segmentation, scientists using 3D imaging at synchrotrons are struggling to analyze the precious data collected at these world-class facilities. There seems to be a large discrepancy between the perceived maturity of state-of-the-art methods and how those methods perform in practice. To investigate this, we conducted a comprehensive review of papers published in the field revolving around synchrotron imaging and deep learning. We determined the methods commonly utilized in practice, as well as the annotation strategies employed. We collected all available data and reproduced the reported results. We evaluated the performance of several commonly used architectures, alongside some emerging methods. Our results show a large variation in the benefits of using deep learning, but trends emerge.

Related papers

• https://doi.org/10.7554/eLife.71359
• https://doi.org/10.1073/pnas.2209938120
• https://doi.org/10.1063/5.0129324
• https://doi.org/10.1038/s41524-022-00734-6

Rikke Gade

Title: Acquisition and analysis of human motion data from thermal imaging

Abstract: Analysis of human motion is interesting for several domains, including surveillance, sports research and rehabilitation. This talk will draw mainly on examples from the sports domain, and cover different options for sensors and image modalities. I will go more into details with thermal imaging, specifically, how it works, advantages and limitations of this image modality, and the perspectives of privacy.

After covering the human detection task, I will show how this data can be analyzed to obtain information like occupancy, trajectories, sports type classification, and energy expenditure of individuals.

Related papers

• Multimodal and multiview distillation for real-time player detection on a football field
• Automatic Analysis of Activities in Sports Arenas Using Thermal Cameras
• Estimation of Energy Expenditure during Treadmill Exercise via Thermal Imaging
• Measuring Energy Expenditure in Sports by Thermal Video Analysis

Melanie Ganz-Benjaminsen

Title: Motion Analysis and Correction for 3D Medical Image Acquisition with a Focus on Brain MRI.

Abstract: Magnetic Resonance Imaging (MRI) is a critical tool for diagnosing and monitoring brain disorders, offering detailed anatomical and functional insights. However, patient movement during image acquisition poses a significant challenge, often resulting in artifacts that compromise image quality and diagnostic accuracy. This talk delves into advanced techniques for motion analysis and correction specifically tailored to brain MRI. We will explore state-of-the-art algorithms for detecting and quantifying patient motion, and discuss novel correction strategies that integrate real-time feedback mechanisms. The presentation will cover the application of machine learning and deep learning models in enhancing motion artifact reduction, as well as the implementation of hardware solutions such as motion-tracking devices. Case studies will illustrate the impact of these innovations on clinical outcomes based on a clinical study carried out at Rigshospitalet. Attendees will gain a comprehensive understanding of the current landscape and future directions in motion correction technologies for brain MRI, ultimately aiming to optimize the reliability and efficacy of brain MRI diagnostics.

Related papers and material

• Imaging Children Without Anesthesia
• Deep Learning for Retrospective Motion Correction in MRI: A Comprehensive Review: https://ieeexplore.ieee.org/abstract/document/10285512/
• Characterisation of Children’s Head Motion for Magnetic Resonance Imaging With and Without General Anaesthesia: https://www.frontiersin.org/articles/10.3389/fradi.2021.789632/full
• Evaluating the performance of markerless prospective motion correction and selective reacquisition in a general clinical protocol for brain MRI: https://osf.io/preprints/psyarxiv/vzh4g

Anders Bjorholm Dahl

Title: Medical CT at Extreme Scale using Synchrotrons.

Abstract: Imaging the microstructure of tissues has entered a new era thanks to the capabilities of the fourth-generation synchrotron facilities. Now, it is possible to first image a sample in an overview scan, and then zoom into much smaller regions than what has previously been possible. This has been the basis for the Human Organ Atlas project at ESRF, where full human organs are scanned at 25-micron resolution and subregions of the same organ are scanned down to 1-micron resolution. At MAX IV, we are also implementing hierarchical synchrotron CT scanning, but at resolutions in the range of 1 micron to 50 nanometers. The resulting images are TB in size, and extracting information from these images requires a new set of image analysis tools. In this presentation, I will give examples of the imaging method and our research in methods for segmenting and quantifying the imaged structures using biomedical examples

.

J. Andreas Bærentzen

Title: Parameterizing complex 3D shapes

Abstract: Often, the information that we seek in biomedical images pertains to the geometry of shapes. The goal of this talk is to provide an overview of how shapes can be represented digitally, methods for converting between different representations, and approaches to computing properties needed for e.g. segmentation, shape descriptors, or loss/energy functions. Initially, we cover three categories of shape representations: implicit representations such as 3D images, boundary representations, with an emphasis on triangle meshes, and, finally, skeletal representations where either the medial axis or a curve skeleton represents the shape. We will discuss the uses and advantages of these representations, pipelines for conversion between representations and geometry processing tools for simplification and noise reduction. Finally, we discuss how to compute curvature and topological properties. Many of the examples will be demonstrated using tools from the GEL library (https://github.com/janba/GEL).

Related papers

• T. V. Christiansen et al. Neural Representation of Open Surfaces. 2023.

Aasa Feragen

Title: Fairness of AI in medical imaging (FAIMI)

Abstract: During the last 10 years, the research community of fairness, equity and accountability in machine learning has highlighted the potential risks associated with biased systems in various application scenarios, ranging from face recognition to criminal justice and job hiring assistants. A large body of research has shown that such machine learning systems can be biased with respect to demographic attributes like gender, ethnicity, age or geographical distribution, presenting unequal behavior on disadvantaged or underrepresented subpopulations. This bias can have a number of sources, ranging from database construction, modeling choices, training strategies and even lack of diversity in team composition, but can also stem from differences in data quality, prevalence, or other hidden correlations. This tutorial will introduce the audience to the standard practices within algorithmic fairness through the lens of medical imaging, and provide case discussions, current research status, potential pitfalls, as well as data resources to enable medical imaging researchers to get started working on bias and fairness in medical imaging. The tutorial is rooted in the FAIMI community (https://faimi-workshop.github.io), an initiative dedicated to promoting knowledge and research about bias and fairness in the medical imaging community.

Summer school challenge