LIVIVO - The Search Portal for Life Sciences

zur deutschen Oberfläche wechseln
Advanced search

Search results

Result 1 - 10 of total 222

Search options

  1. Article ; Online: Using adaptive learning rate to generate adversarial images

    Ding Ning / Möller Knut

    Current Directions in Biomedical Engineering, Vol 9, Iss 1, Pp 359-

    2023  Volume 362

    Abstract: Convolutional neural networks (CNNs) have proved their efficiency in performing image classification tasks, as they can automatically extract the image features and make the corresponding prediction. Meanwhile, the CNNs application is highly challenged ... ...

    Abstract Convolutional neural networks (CNNs) have proved their efficiency in performing image classification tasks, as they can automatically extract the image features and make the corresponding prediction. Meanwhile, the CNNs application is highly challenged by their vulnerability to adversarial samples. These samples are slightly different from the legitimate samples, but the CNN gives wrong classification. There are various ways to find the adversarial samples. The most common method is using backpropagation to generate gradients as the directed perturbation. Contrarily to set a constrained limitation, in this paper, we use iterative fast gradient sign method to generate adversarial images with the minimum perturbation. The CNNs were trained to perform surgical tool recognition as a configuration for the modern operation room. The coefficient or the learning rate which influenced the modification per iteration, was set to be adaptive instead of a fixed number. A few functions were utilized to perform the learning rate decay to compare the performance. Especially, we propose a new adaptive learning rate algorithm that consider the loss as a part of influence factor constitute the learning rate for the rest iterations. According to the experiments, our loss adaptive learning rate method was proved to be efficient to get the minimal perturbations for adversarial attack.
    Keywords convolutional neural network ; adversarial attack ; surgical tool recognition ; Medicine ; R
    Subject code 006
    Language English
    Publishing date 2023-09-01T00:00:00Z
    Publisher De Gruyter
    Document type Article ; Online
    Database BASE - Bielefeld Academic Search Engine (life sciences selection)

    Full text online

    More links

    Kategorien

    Inter-library loan at ZB MED

    Your chosen title can be delivered directly to ZB MED Cologne location if you are registered as a user at ZB MED Cologne.

  2. Article ; Online: Robustness evaluation on different training state of a CNN model

    Ding Ning / Möller Knut

    Current Directions in Biomedical Engineering, Vol 8, Iss 2, Pp 497-

    2022  Volume 500

    Abstract: Convolutional neural networks (CNNs) have proved to be successful in many applications such as image processing. However, even imperceptible perturbations applied to the images can make the neural network performance unreliable. To guarantee an accurate ... ...

    Abstract Convolutional neural networks (CNNs) have proved to be successful in many applications such as image processing. However, even imperceptible perturbations applied to the images can make the neural network performance unreliable. To guarantee an accurate performance in safety critical fields, it is necessary to assess the robustness of CNN solutions before launching. Adversarial attack is a machine learning approach to generate perturbations on real samples to detect the vulnerability of CNN. In this paper, we will use an adversarial attack technique to evaluate a CNN at different training states. The model was trained to perform surgical tool classification task, which was applied to recognize surgical tool in Cholecystectomy to further analyze surgical process. The experiments demonstrate the relation between training states and robustness, i.e. the robustness improved at higher training states, especially for some particular classes. In future work, additional training with generated adversarial images may improve the robustness of the model.
    Keywords convolutional neural network ; adversarial attack ; surgical tool recognition ; Medicine ; R
    Subject code 006
    Language English
    Publishing date 2022-09-01T00:00:00Z
    Publisher De Gruyter
    Document type Article ; Online
    Database BASE - Bielefeld Academic Search Engine (life sciences selection)

    Full text online

    More links

    Kategorien

    Inter-library loan at ZB MED

    Your chosen title can be delivered directly to ZB MED Cologne location if you are registered as a user at ZB MED Cologne.

  3. Article ; Online: Challenging requirements and optical depth estimation techniques in laparoscopy

    Gobel Birthe / Möller Knut

    Current Directions in Biomedical Engineering, Vol 8, Iss 2, Pp 687-

    2022  Volume 690

    Keywords Medicine ; R
    Language English
    Publishing date 2022-09-01T00:00:00Z
    Publisher De Gruyter
    Document type Article ; Online
    Database BASE - Bielefeld Academic Search Engine (life sciences selection)

    Full text online

    More links

    Kategorien

    Inter-library loan at ZB MED

    Your chosen title can be delivered directly to ZB MED Cologne location if you are registered as a user at ZB MED Cologne.

  4. Article ; Online: Determination of the water content of an ex vivo porcine liver

    Busch Christoph / Rupitsch Stefan J. / Moeller Knut

    Current Directions in Biomedical Engineering, Vol 9, Iss 1, Pp 222-

    2023  Volume 225

    Abstract: When modeling High-Frequency (HF) surgical applications the choice of the right dielectric tissue parameters is a key factor for success. The water content of biological tissue is of particular importance since the dielectric tissue properties highly ... ...

    Abstract When modeling High-Frequency (HF) surgical applications the choice of the right dielectric tissue parameters is a key factor for success. The water content of biological tissue is of particular importance since the dielectric tissue properties highly depend on it. Therefore, we investigate a simple and fast method for the determination of relative tissue water content using small samples of an ex vivo porcine liver. To identify the water content, we utilized a desiccation and weighing experiment. It is assumed that the weight loss can be directly attributed to the loss of water in the case of desiccation. Based on the ratio between the initial and dry weight of a sample, the relative water content can be determined. The drying procedure lasted 28 hours, with 16 small samples taken from an ex vivo porcine liver. At the end of the drying procedure, a relative tissue water content of 69.6 % ± 1.1 % was observed. Using a curve fit to the measured data, it can be assumed that approximately 99.3 % of the tissue water content had evaporated by the end of the drying process. With the drying and weighing method used, we were able to determine the water content of the ex vivo porcine liver relatively easily. However, our results also suggest that optimizations of the applied method should be considered for future measurements. Especially, if a faster determination of the water content is required, which in a broader sense should afterward serve for the validation of a simulation model.
    Keywords tissue water content ; porcine liver ; desiccation ; Medicine ; R
    Language English
    Publishing date 2023-09-01T00:00:00Z
    Publisher De Gruyter
    Document type Article ; Online
    Database BASE - Bielefeld Academic Search Engine (life sciences selection)

    Full text online

    More links

    Kategorien

    Inter-library loan at ZB MED

    Your chosen title can be delivered directly to ZB MED Cologne location if you are registered as a user at ZB MED Cologne.

  5. Article ; Online: Analysis of stretch distribution of high compliant elastomers within folded lumen vessels

    Bhave Ashish / Rupitsch Stefan J. / Moeller Knut

    Current Directions in Biomedical Engineering, Vol 9, Iss 1, Pp 686-

    2023  Volume 689

    Abstract: Elastomer based high compliant balloons containing strain sensing element(s) (SEs) is currently under development intended for in-vivo biomechanical diagnostics of vessels. It could potentially reveal local lumen features based on patterns derived from ... ...

    Abstract Elastomer based high compliant balloons containing strain sensing element(s) (SEs) is currently under development intended for in-vivo biomechanical diagnostics of vessels. It could potentially reveal local lumen features based on patterns derived from the sensing elements. A Finite Element based, simulation study in COMSOL® (v5.6) focusses on in-vivo inflation behavior of an elastomeric balloon being equipped with SE, whose compliance is ideally magnitudes higher than the surrounding tissue in an idealized 2D setup. We hypothesized the vessel’s inner wall as a closed convexconcave 4-fold structure correlated to surface structures found in urethrae and parameterized the fold depth. A set of SEs consisted of one SE over the inner surface of balloon while the other over the outer wall. Out of the three adjacent placed sets, The first set was closer to the tissue lumen while the third set the farthest. We assessed the stretch of balloon over its inner circumference through SEs. At conformal contact with the tissue wall, The first SE shows a higher value, while the third element undergoes the least stretch within the three sensing elements. The SE’s over the outer circumference show the exact opposite relation. The differences between the sensing elements over the inner and outer circumferences show a correlation with the curvature of the tissue it conforms to. With use of balloon that was 10x thicker (10μm vs 1μm), around 10x larger stretch differences were captured suggesting possibility to use less sensitive measuring system. For the ideal situations performed, the curvature and depth information may be comprehended by observing differences between inner and outer SEs while thicker balloons prove to be more useful to generate differences that are easier to capture in practical situations. This semiquantitative study suggests that simulation studies are powerful tools to obtain new analytical techniques for shape characterization.
    Keywords elastomers ; strain sensing ; shape characterization ; biomechanics ; lumen ; Medicine ; R
    Subject code 500
    Language English
    Publishing date 2023-09-01T00:00:00Z
    Publisher De Gruyter
    Document type Article ; Online
    Database BASE - Bielefeld Academic Search Engine (life sciences selection)

    Full text online

    More links

    Kategorien

    Inter-library loan at ZB MED

    Your chosen title can be delivered directly to ZB MED Cologne location if you are registered as a user at ZB MED Cologne.

  6. Article ; Online: Detection of Outdated Structural Priors in the Discrete Cosine Transformation-based Electrical Impedance Tomography Algorithm

    Chen Rongqing / Moeller Knut

    Current Directions in Biomedical Engineering, Vol 7, Iss 2, Pp 676-

    2021  Volume 679

    Abstract: Morphological prior information incorporated with the discrete cosine transformation (DCT) based electrical impedance tomography (EIT) algorithm can improve the interpretability of EIT reconstructions in clinical applications. However, an outdated ... ...

    Abstract Morphological prior information incorporated with the discrete cosine transformation (DCT) based electrical impedance tomography (EIT) algorithm can improve the interpretability of EIT reconstructions in clinical applications. However, an outdated structural prior can yield a misleading reconstruction compromising the accuracy of the clinical diagnosis and the appropriate treatment decision. In this contribution, we propose a redistribution index scaled between 0 and 1 to quantify the possible error in a DCT-based EIT reconstruction influenced by structural prior information. Two simulation models of different tissue atelectasis and collapsed ratios were investigated. Outdated and updated structural prior information were applied to obtain different EIT reconstructions using this simulated data, with which the redistribution index was calculated and compared. When the difference between prior and reality (the redistribution index) became larger and exceeded a threshold, this was considered as an indicator of an outdated prior information. The evaluation result shows the potential of the redistribution index to detect outdated prior information in a DCT-based EIT algorithm.
    Keywords electrical impedance tomography ; simulation ; prior information ; redistribution index ; Medicine ; R
    Subject code 518
    Language English
    Publishing date 2021-10-01T00:00:00Z
    Publisher De Gruyter
    Document type Article ; Online
    Database BASE - Bielefeld Academic Search Engine (life sciences selection)

    Full text online

    More links

    Kategorien

    Inter-library loan at ZB MED

    Your chosen title can be delivered directly to ZB MED Cologne location if you are registered as a user at ZB MED Cologne.

  7. Article ; Online: Comparison of a histology based multi layer artery model to its simplified axisymmetric model.

    Bhave Ashish / Möller Knut

    Current Directions in Biomedical Engineering, Vol 7, Iss 2, Pp 590-

    2021  Volume 593

    Abstract: Arteries are vessel structures that serve vital function of transportation of blood to different parts of the body. Researchers have experimented with some approaches to model the arterial behaviour and to analyse its biomechanical properties. To analyse ...

    Abstract Arteries are vessel structures that serve vital function of transportation of blood to different parts of the body. Researchers have experimented with some approaches to model the arterial behaviour and to analyse its biomechanical properties. To analyse the in-vivo arterial properties, at Furtwangen University an inflatable sensoractuator system is being developed, which provides the basis for a decision support system for vascular surgeons. The capabilities of this sensor shall be evaluated in simulations which requires appropriate modelling of the arteries. The inverse problem, i.e. how to efficiently identify arterial wall properties from sensor readings is targeted. A histology motivated 3D artery model was implemented in FEM using COMSOL (v5.5). The geometry of one model was based on a cross section of a real artery. The second model was axisymmetric and of equal dimensions with respect to volume, layer thickness etc. A biomechanical pressure-stretch analysis was performed applying an inflating pressure inside the walls of the vessels. Stretch in different areas of the first model was evaluated and the circumferential strain was compared to the axisymmetric model. The results show variation of strains within the segments of the first model of upto 10 percent. In addition, its outer wall circumferential stretch was found to be 10 percent lower compared to the axisymmetric setup. This comparison sheds light upon whether a simplification of arterial models is possible, without loss of accuracy in the context of the novel sensor evaluation. It provides useful information whether e.g. standardizing vessel structures to axisymmetric models will still provide results within allowable tolerance limits. Simulations proved useful to evaluate different vessel model formulations in the context of arterial diagnostics.
    Keywords modelling ; artery ; identification ; simulation ; Medicine ; R
    Subject code 600
    Language English
    Publishing date 2021-10-01T00:00:00Z
    Publisher De Gruyter
    Document type Article ; Online
    Database BASE - Bielefeld Academic Search Engine (life sciences selection)

    Full text online

    More links

    Kategorien

    Inter-library loan at ZB MED

    Your chosen title can be delivered directly to ZB MED Cologne location if you are registered as a user at ZB MED Cologne.

  8. Article ; Online: Analysis of expansion within a pressure inflated section of an urethral stricture model

    Bhave Ashish / Möller Knut

    Current Directions in Biomedical Engineering, Vol 7, Iss 2, Pp 578-

    2021  Volume 581

    Abstract: The Urethra is a long tubular structure in the genitourinary tract and serves important functions. Researchers have experimented with some approaches to model the urethra and to analyse its biomechanical properties. However, experiments to model the in- ... ...

    Abstract The Urethra is a long tubular structure in the genitourinary tract and serves important functions. Researchers have experimented with some approaches to model the urethra and to analyse its biomechanical properties. However, experiments to model the in-vivo behaviour of urethra with strictures is not thoroughly explored. To analyse the in-vivo urethral properties and specifically for supporting treatment of strictures, a new inflatable sensor-actuator system is being developed. The capabilities of this sensor shall be evaluated in simulations which require appropriate modelling of the human male urethra with strictures. This forms a part of the identification procedure for a variety of urethra conditions and geometries, which in turn forms a basis for inverse modelling. As an initial simplified approach, an axisymmetric Finite Element model was generated that resembled the urethra incorporating a stricture region. An ideal actuator with sensor elements exerting a pressure on inner wall of this urethra was simulated. Three circumference measurement zones within the sensor height (top surface, centre and bottom surface) were implemented. The resulting pressure-extension (circumferential) responses were determined at these measurement zones. The sensor was placed at different lengths within this urethral tube and inflated and the pressure-extension responses were noted. It was found that depending on the position of the sensor-actuator, the extension of tissue can vary. The possible factors for this variation were the finite length of the actuator as well as the influence of tissue properties around the measurement zones. This is important information for the interpretation of sensor data to be gained by the current development. It was possible to generate datasets based on an ideal sensor model, that proved helpful in the evaluation of biomechanical tissue properties in healthy and stricture conditions. This indicates simulations are a versatile and prospective way to test new sensors prior to real experiments.
    Keywords modelling ; urethra ; strictures ; simulation ; Medicine ; R
    Subject code 620
    Language English
    Publishing date 2021-10-01T00:00:00Z
    Publisher De Gruyter
    Document type Article ; Online
    Database BASE - Bielefeld Academic Search Engine (life sciences selection)

    Full text online

    More links

    Kategorien

    Inter-library loan at ZB MED

    Your chosen title can be delivered directly to ZB MED Cologne location if you are registered as a user at ZB MED Cologne.

  9. Article ; Online: Generating adversarial images to monitor the training state of a CNN model

    Ding Ning / Möller Knut

    Current Directions in Biomedical Engineering, Vol 7, Iss 2, Pp 303-

    2021  Volume 306

    Abstract: Deep neural networks have shown effectiveness in many applications, however, in regulated applications like automotive or medicine, quality guarantees are required. Thus, it is important to understand the robustness of the solutions to perturbations in ... ...

    Abstract Deep neural networks have shown effectiveness in many applications, however, in regulated applications like automotive or medicine, quality guarantees are required. Thus, it is important to understand the robustness of the solutions to perturbations in the input space. In order to identify the vulnerability of a trained classification model and evaluate the effect of different perturbations in the input on the output class, two different methods to generate adversarial examples were implemented. The adversarial images created were developed into a robustness index to monitor the training state and safety of a convolutional neural network model. In the future work, some generated adversarial images will be included into the training phase to improve the model robustness.
    Keywords deep learning ; convolutional neural network ; adversarial attack ; Medicine ; R
    Language English
    Publishing date 2021-10-01T00:00:00Z
    Publisher De Gruyter
    Document type Article ; Online
    Database BASE - Bielefeld Academic Search Engine (life sciences selection)

    Full text online

    More links

    Kategorien

    Inter-library loan at ZB MED

    Your chosen title can be delivered directly to ZB MED Cologne location if you are registered as a user at ZB MED Cologne.

  10. Article ; Online: Nonlinearity of Magnetostrictive Torque Sensor under Varying External Magnetic Field Strength

    Wilkie Jack A. / Moeller Knut

    Current Directions in Biomedical Engineering, Vol 7, Iss 2, Pp 759-

    2021  Volume 762

    Abstract: Correctly torquing bone screws is an important factor in achieving positive patient outcomes during orthopaedic surgery. A torque-limiting smart screwdriver concept has been proposed, and ongoing work is being undertaken to model the screwing process and ...

    Abstract Correctly torquing bone screws is an important factor in achieving positive patient outcomes during orthopaedic surgery. A torque-limiting smart screwdriver concept has been proposed, and ongoing work is being undertaken to model the screwing process and allow the concept to work. These models require experimental validation, so a test rig was developed. The magnetostrictive torque sensor in this test rig was affected by magnetic parts of the test rig, which offset the zero-torque point; this raised concerns over the effects on linearity, which were tested here. The torque sensor was tested against a non-magnetostrictive reference under varying external magnetic conditions. While the magnetic field offset the torque, there was no notable change in linearity under the conditions tested, and the linearity was always within the datasheet specifications. Hence, we conclude that in the context of this test rig, there were no negative effects on linearity, although under higher loading or stronger magnetic conditions, this may not hold.
    Keywords magnetostrictive ; torque sensor ; torque measurement ; nonlinearity ; linearity ; magnetic field ; Medicine ; R
    Subject code 600
    Language English
    Publishing date 2021-10-01T00:00:00Z
    Publisher De Gruyter
    Document type Article ; Online
    Database BASE - Bielefeld Academic Search Engine (life sciences selection)

    Full text online

    More links

    Kategorien

    Inter-library loan at ZB MED

    Your chosen title can be delivered directly to ZB MED Cologne location if you are registered as a user at ZB MED Cologne.

To top