Deep learning enables a novel magnetic resonance imaging contrast that unveils chemical and microstructural brain tissue changes through nondipolar larmor frequency shifts. - In: Biomedical engineering, ISSN 1862-278X, Bd. 68 (2023), S. 160
https://doi.org/10.1515/bmte-2023-2001
Mapping the anisotropy of tissue magnetic susceptibility from single-orientation magnetic resonance imaging. - In: Biomedical engineering, ISSN 1862-278X, Bd. 68 (2023), S. 156
https://doi.org/10.1515/bmte-2023-2001
Real-time smartphone-assisted EEG electrode localization and augmented reality application. - In: Biomedical engineering, ISSN 1862-278X, Bd. 68 (2023), S. 153
https://doi.org/10.1515/bmte-2023-2001
Cobalt ferrite nanoparticles as thermal markers on lateral flow assays. - In: Biomedical engineering, ISSN 1862-278X, Bd. 68 (2023), S. 126
https://doi.org/10.1515/bmte-2023-2001
Novel MRI technique reveals subtypes of paramagnetic rim lesions and predicts 5-year rim disappearance. - In: Multiple sclerosis journal, ISSN 1477-0970, Bd. 29 (2023), 2, P031
https://doi.org/10.1177/13524585231169437
Fast fictitious surface charge method for calculation of torso surface potentials. - In: IEEE Xplore digital library, ISSN 2473-2001, (2023), insges. 4 S.
Well-established forward modeling methods in electrocardiography (ECG) require fine meshes to calculate the electric scalar potential at the body surface with high accuracy. We introduce a fast fictitious surface charge method (FSCM) with local mesh refinement and smart calculations of elements interactions which improves the accuracy of the calculations and, at the same time, preserves the performance speed.
https://doi.org/10.1109/COMPUMAG56388.2023.10411804
Linear versus quadratic detrending in analyzing simultaneous changes in DC-EEG and transcutaneous pCO2. - In: 2023 45th Annual International Conference of the IEEE Engineering in Medicine & Biology Conference (EMBC), (2023), insges. 4 S.
Physiological direct current (DC) potential shifts in electroencephalography (EEG) can be masked by artifacts such as slow electrode drifts. To reduce the influence of these artifacts, linear detrending has been proposed as a pre-processing step. We considered quadratic detrending, which has hardly been addressed for ultralow frequency components in EEG. We compared the performance of linear and quadratic detrending in simultaneously acquired DC-EEG and transcutaneous partial pressure of carbon dioxide during two activation methods: hyperventilation (HV) and apnea (AP). Quadratic detrending performed significantly better than linear detrending in HV, while for AP, our analysis was inconclusive with no statistical significance. We conclude that quadratic detrending should be considered for DC-EEG preprocessing.
https://doi.org/10.1109/EMBC40787.2023.10340855
Entrainment and resonance effects with a new mobile audio-visual stimulation device. - In: 2023 45th Annual International Conference of the IEEE Engineering in Medicine & Biology Conference (EMBC), (2023), insges. 4 S.
Entrainment and photic driving effects triggered by repetitive visual stimulation are long-established in clinical and therapeutic scenarios. Nonetheless, such stimulation patterns are typically bound to stationary clinical and laboratory applications. We investigated the effects of repetitive stimulation with a new dynamic auditory-visual stimulation pattern using a novel consumer-grade stimulation device for home application. Fourteen volunteers (study group) received 16 sessions of combined auditory-visual stimulation during four weeks. An additional control group (seven volunteers) received auditory-only stimulation for two sessions. From 64-channel electroencephalography recordings, we compared individual alpha peak frequencies (iAPF) between week one and week four as well as power values from the time-frequency analysis. The novel stimulation device yielded stable entrainment and resonance effects for all investigated stimulation frequencies. Both groups showed no differences in their iAPFs between weeks one and four. The power comparison suggests that there are similar entrainment and resonance effects in week one and week four within the study group. We conclude that the novel portable consumer-grade stimulation device is suitable for home-based auditory-visual stimulation leading to consistent entrainment and resonance effects over the course of 16 stimulation sessions over four weeks.
https://doi.org/10.1109/EMBC40787.2023.10341051
Dropout is NOT all you need to prevent gradient leakage. - In: 37th AAAI Conference on Artificial Intelligence (AAAI-23), (2023), S. 9733-9741
Gradient inversion attacks on federated learning systems reconstruct client training data from exchanged gradient information. To defend against such attacks, a variety of defense mechanisms were proposed. However, they usually lead to an unacceptable trade-off between privacy and model utility. Recent observations suggest that dropout could mitigate gradient leakage and improve model utility if added to neural networks. Unfortunately, this phenomenon has not been systematically researched yet. In this work, we thoroughly analyze the effect of dropout on iterative gradient inversion attacks. We find that state of the art attacks are not able to reconstruct the client data due to the stochasticity induced by dropout during model training. Nonetheless, we argue that dropout does not offer reliable protection if the dropout induced stochasticity is adequately modeled during attack optimization. Consequently, we propose a novel Dropout Inversion Attack (DIA) that jointly optimizes for client data and dropout masks to approximate the stochastic client model. We conduct an extensive systematic evaluation of our attack on four seminal model architectures and three image classification datasets of increasing complexity. We find that our proposed attack bypasses the protection seemingly induced by dropout and reconstructs client data with high fidelity. Our work demonstrates that privacy inducing changes to model architectures alone cannot be assumed to reliably protect from gradient leakage and therefore should be combined with complementary defense mechanisms.
On learning of inverse kinematics for highly redundant robots with neural networks. - In: IEEE Xplore digital library, ISSN 2473-2001, (2023), S. 402-408
The inverse kinematic problem for redundant robots is still difficult to solve. One approach is learning the inverse kinematic model with artificial neural networks, while the key challenge is the ambiguity of solutions. Due to the redundancy in the robot's degrees of freedom, there are multiple or even unlimited valid joint states bringing the end effector to a desired position. We show to what extent this problem influences the achievable accuracy of supervised training approaches depending on the number of degrees of freedom. To overcome the difficulties, a new training scheme is proposed, which uses the analytically solvable forward kinematics model. The new unsupervised training method uses random sampling in the joint state space and is not dependent on ambiguous tuples of joint values and end effector poses. We analyze the effect of the sample density on the remaining position error and show that additional soft constraints can easily be integrated in the training scheme, which offers the possibility to consider obstacle avoidance directly in the inverse kinematic model. Evaluations have been done using different robot models with up to 20 degrees of freedom, while not only position, but also the end effector's orientation at the goal point is considered.
https://doi.org/10.1109/ICAR58858.2023.10406939