A5056603838- Medical Image Segmentation Techniques
- Radiomics and Machine Learning in Medical Imaging
- Advanced Neural Network Applications
- AI in cancer detection
- Explainable Artificial Intelligence (XAI)
- Fetal and Pediatric Neurological Disorders
- Retinal Imaging and Analysis
- Machine Learning in Healthcare
- COVID-19 diagnosis using AI
- Domain Adaptation and Few-Shot Learning
- Acute Ischemic Stroke Management
- Artificial Intelligence in Healthcare
- 3D Shape Modeling and Analysis
- Advanced X-ray and CT Imaging
- Cell Image Analysis Techniques
- Artificial Intelligence in Healthcare and Education
- Image Retrieval and Classification Techniques
- Magnetic properties of thin films
- Industrial Vision Systems and Defect Detection
- Physics of Superconductivity and Magnetism
- Topic Modeling
- Cardiac Valve Diseases and Treatments
- Radiology practices and education
- Cleft Lip and Palate Research
- Digital Imaging for Blood Diseases
University of Tübingen
2019-2024
Hertie Institute for Clinical Brain Research
2023-2024
ETH Zurich
2017-2018
Board of the Swiss Federal Institutes of Technology
2018
Imperial College London
2014-2017
Institute of Group Analysis
2015-2017
Johannes Gutenberg University Mainz
2003
Delineation of the left ventricular cavity, myocardium, and right ventricle from cardiac magnetic resonance images (multi-slice 2-D cine MRI) is a common clinical task to establish diagnosis. The automation corresponding tasks has thus been subject intense research over past decades. In this paper, we introduce "Automatic Cardiac Diagnosis Challenge" dataset (ACDC), largest publicly available fully annotated for purpose MRI (CMR) assessment. contains data 150 multi-equipments CMRI recordings...
Identifying and interpreting fetal standard scan planes during 2-D ultrasound mid-pregnancy examinations are highly complex tasks, which require years of training. Apart from guiding the probe to correct location, it can be equally difficult for a non-expert identify relevant structures within image. Automatic image processing provide tools help experienced as well inexperienced operators with these tasks. In this paper, we propose novel method based on convolutional neural networks,...
Attributing the pixels of an input image to a certain category is important and well-studied problem in computer vision, with applications ranging from weakly supervised localisation understanding hidden effects data. In recent years, approaches based on interpreting previously trained neural network classifier have become de facto state-of-the-art are commonly used medical as well natural datasets. this paper, we discuss limitation these which may lead only subset specific features being...
In this study, we construct a spatio-temporal surface atlas of the developing cerebral cortex, which is an important tool for analysing and understanding normal abnormal cortical development. utero Magnetic Resonance Imaging (MRI) 80 healthy fetuses was performed, with gestational age range 21.7 to 38.9 weeks. Topologically correct models were extracted from reconstructed 3D MRI volumes. Accurate correspondences obtained by applying joint spectral analysis cortices sets subjects close...
Distribution shifts remain a problem for the safe application of regulated medical AI systems, and may impact their real-world performance if undetected. Postmarket can occur example algorithms developed on data from various acquisition settings heterogeneous population are predominantly applied in hospitals with lower quality or other centre-specific factors, where some ethnicities over-represented. Therefore, distribution shift detection could be important monitoring AI-based products...
Interpretability is essential for machine learning algorithms in high-stakes application fields such as medical image analysis. However, high-performing black-box neural networks do not provide explanations their predictions, which can lead to mistrust and suboptimal human-ML collaboration. Post-hoc explanation techniques, are widely used practice, have been shown suffer from severe conceptual problems. Furthermore, we show this paper, current techniques perform adequately the multi-label...
Multi-atlas segmentation is a widely used tool in medical image analysis, providing robust and accurate results by learning from annotated atlas datasets. However, the availability of fully images for training limited due to time required labelling task. Segmentation methods requiring only proportion each be labelled could therefore reduce workload on expert raters tasked with annotating images. To address this issue, we first re-examine problem common many existing approaches formulate its...
To efficiently establish training databases for machine learning methods, collaborative and crowdsourcing platforms have been investigated to collectively tackle the annotation effort. However, when this concept is ported medical imaging domain, reading expertise will a direct impact on accuracy. In study, we examine of amount available annotations accuracy outcome liver segmentation problem in an abdominal computed tomography (CT) image database. controlled experiments, study different...
Abstract In medical image classification tasks like the detection of diabetic retinopathy from retinal fundus images, it is highly desirable to get visual explanations for decisions black-box deep neural networks (DNNs). However, gradient-based saliency methods often fail highlight diseased regions reliably. On other hand, adversarially robust models have more interpretable gradients than plain but suffer typically a significant drop in accuracy, which unacceptable clinical practice. Here,...
Retinal blood vessel segmentation can extract clinically relevant information from fundus images. As manual tracing is cumbersome, algorithms based on Convolution Neural Networks have been developed. Such studies used small publicly available datasets for training and measuring performance, running the risk of overfitting. Here, we provide a rigorous benchmark various architectural choices commonly in literature largest dataset published to date. We train evaluate five models FIVES image...