A5057748958- AI in cancer detection
- Advanced Image Processing Techniques
- Generative Adversarial Networks and Image Synthesis
- Renal Diseases and Glomerulopathies
- Natural Language Processing Techniques
- Radiomics and Machine Learning in Medical Imaging
- Geophysical Methods and Applications
- Digital Radiography and Breast Imaging
- Adversarial Robustness in Machine Learning
- Systemic Sclerosis and Related Diseases
- Speech and dialogue systems
- Privacy-Preserving Technologies in Data
- Topic Modeling
- Image and Signal Denoising Methods
University Hospital Cologne
2023-2024
University of Cologne
2023-2024
Institut für Medizinische Informatik, Biometrie und Epidemiologie
2023
RWTH Aachen University
2023
Gradient inversion attacks can reconstruct the victim's private data once they have access to model and gradient. However, existing research is still immature, many are conducted in ideal conditions. It unclear how damaging such really be effectively defended. In this paper, we first summarize current relevant researches their limitations. Then design a general gradient attack framework, which both FedSGD FedAVG. We propose approaches enhance label inference image restoration, respectively....
The Oxford Classification for IgA nephropathy is the most successful example of an evidence-based nephropathology classification system. aim our study was to replicate glomerular components scoring with end-to-end deep learning pipeline that involves automatic segmentation followed by mesangial hypercellularity (M), endocapillary (E), segmental sclerosis (S) and active crescents (C).A total number 1056 periodic acid-Schiff (PAS) whole slide images (WSIs), coming from 386 kidney biopsies,...
Currently, the medical field is witnessing an increase in use of machine learning techniques. Supervised methods adopted classification, prediction, and segmentation tasks for images always experience decreased performance when training testing datasets do not follow independent identically distributed assumption. These distribution shift situations seriously influence applications’ robustness, fairness, trustworthiness domain. Hence, this article, we adopt CycleGAN (generative adversarial...
Supervised methods, such as those utilized in classification, prediction, and segmentation tasks for medical images, experience a decline performance when the training testing datasets violate i.i.d (independent identically distributed) assumption. Hence we adopted CycleGAN(Generative Adversarial Networks) method to cycle CT(Computer Tomography) data from different terminals/manufacturers, which aims eliminate distribution shift diverse terminals. But due model collapse problem of GAN-based...
The paper discusses biases in medical imaging analysis, particularly focusing on the challenges posed by development of machine learning algorithms and generative models. It introduces a taxonomy bias problems addresses them through data infrastructure initiative: PADME (Platform for Analytics Distributed Machine-Learning Enterprises), which is part National Research Data Infrastructure Personal Health (NFDI4Health) project. facilitates structuring sharing health while ensuring privacy...
Black-box context-free grammar inference presents a significant challenge in many practical settings due to limited access example programs. The state-of-the-art methods, Arvada and Treevada, employ heuristic approaches generalize rules, initiating from flat parse trees exploring diverse generalization sequences. We have observed that these suffer low quality readability, primarily because they process entire strings, adding the complexity substantially slowing down computations. To overcome...