Refine
Year of publication
Document Type
- conference proceeding (presentation, abstract) (39) (remove)
Is part of the Bibliography
- no (39)
Keywords
- Deep Learning (3)
- Altersgerechte Assistenzsysteme (2)
- Artificial Intelligence (2)
- Endoscopy (2)
- Künstliche Intelligenz (2)
- AAL-Technik (1)
- Aerodynamics (1)
- Altenpflege (1)
- Ambulanter Hilfs- und Pflegedienst (1)
- Assistenzsystem (1)
Institute
- Regensburg Center of Health Sciences and Technology - RCHST (39) (remove)
Begutachtungsstatus
- peer-reviewed (8)
- begutachtet (5)
Aims
VA is an endoscopic finding of celiac disease (CD), which can easily be missed if pretest probability is low. In this study, we aimed to develop an artificial intelligence (AI) algorithm for the detection of villous atrophy on endoscopic images.
Methods
858 images from 182 patients with VA and 846 images from 323 patients with normal duodenal mucosa were used for training and internal validation of an AI algorithm (ResNet18). A separate dataset was used for external validation, as well as determination of detection performance of experts, trainees and trainees with AI support. According to the AI consultation distribution, images were stratified into “easy” and “difficult”.
Results
Internal validation showed 82%, 85% and 84% for sensitivity, specificity and accuracy. External validation showed 90%, 76% and 84%. The algorithm was significantly more sensitive and accurate than trainees, trainees with AI support and experts in endoscopy. AI support in trainees was associated with significantly improved performance. While all endoscopists showed significantly lower detection for “difficult” images, AI performance remained stable.
Conclusions
The algorithm outperformed trainees and experts in sensitivity and accuracy for VA detection. The significant improvement with AI support suggests a potential clinical benefit. Stable performance of the algorithm in “easy” and “difficult” test images may indicate an advantage in macroscopically challenging cases.
Aims
Evaluation of the add-on effect an artificial intelligence (AI) based clinical decision support system has on the performance of endoscopists with different degrees of expertise in the field of Barrett's esophagus (BE) and Barrett's esophagus-related neoplasia (BERN).
Methods
The support system is based on a multi-task deep learning model trained to solve a segmentation and several classification tasks. The training approach represents an extension of the ECMT semi-supervised learning algorithm. The complete system evaluates a decision tree between estimated motion, classification, segmentation, and temporal constraints, to decide when and how the prediction is highlighted to the observer. In our current study, ninety-six video cases of patients with BE and BERN were prospectively collected and assessed by Barrett's specialists and non-specialists. All video cases were evaluated twice – with and without AI assistance. The order of appearance, either with or without AI support, was assigned randomly. Participants were asked to detect and characterize regions of dysplasia or early neoplasia within the video sequences.
Results
Standalone sensitivity, specificity, and accuracy of the AI system were 92.16%, 68.89%, and 81.25%, respectively. Mean sensitivity, specificity, and accuracy of expert endoscopists without AI support were 83,33%, 58,20%, and 71,48 %, respectively. Gastroenterologists without Barrett's expertise but with AI support had a comparable performance with a mean sensitivity, specificity, and accuracy of 76,63%, 65,35%, and 71,36%, respectively.
Conclusions
Non-Barrett's experts with AI support had a similar performance as experts in a video-based study.
Aims
AI has proven great potential in assisting endoscopists in diagnostics, however its role in therapeutic endoscopy remains unclear. Endoscopic submucosal dissection (ESD) is a technically demanding intervention with a slow learning curve and relevant risks like bleeding and perforation. Therefore, we aimed to develop an algorithm for the real-time detection and delineation of relevant structures during third-space endoscopy.
Methods
5470 still images from 59 full length videos (47 ESD, 12 POEM) were annotated. 179681 additional unlabeled images were added to the training dataset. Consequently, a DeepLabv3+ neural network architecture was trained with the ECMT semi-supervised algorithm (under review elsewhere). Evaluation of vessel detection was performed on a dataset of 101 standardized video clips from 15 separate third-space endoscopy videos with 200 predefined blood vessels.
Results
Internal validation yielded an overall mean Dice score of 85% (68% for blood vessels, 86% for submucosal layer, 88% for muscle layer). On the video test data, the overall vessel detection rate (VDR) was 94% (96% for ESD, 74% for POEM). The median overall vessel detection time (VDT) was 0.32 sec (0.3 sec for ESD, 0.62 sec for POEM).
Conclusions
Evaluation of the developed algorithm on a video test dataset showed high VDR and quick VDT, especially for ESD. Further research will focus on a possible clinical benefit of the AI application for VDR and VDT during third-space endoscopy.
The special wing geometry of dragonflies consisting of veins and a membrane forming a corrugated profile leads to special aerodynamic characteristics. To capture the governing flow regimes of a dragonfly wing in detail, a realistic wing model has to be investigated. Therefore, this study aimed to analyze the aerodynamic characteristics of a 3D dragonfly wing reconstructed from a high-resolution micro-CT scan. Afterwards, a spatially high discretized mesh was generated using the mesh generator CENTAUR™ 14.5.0.2 (CentaurSoft, Austin, TX, US) to finally conduct Computational Fluid Dynamics (CFD) investigations in Fluent® 2020 R2 (ANSYS, Inc., Canonsburg, PA, US). Due to the small dimensions of the wing membrane, only the vein structure of a Camacinia Gigantea was captured at a micro-CT voxel size of 7 microns. The membrane was adapted and connected to the vein structure using a Boolean union operation. Occurring nconsistencies after combining the veins and the membrane were corrected using an adapted pymesh script [1]. As an initial study, only one quarter of the wing (outer wing section) was investigated to reduce the required computational effort. The resulting hybrid mesh consisting of 10 pseudo-structured prism layers along the wing surface and tetrahedra in the farfield area has 43 mio. nodes. The flow around the wing was considered to be incompressible and laminar using transient calculations. When the flow passes the vein structures, steady vortices occur in the corrugation valleys leading to recirculation zones. Therefore, the dragonfly wing resembles the profile of an airfoil. This leads to comparable lift coefficients of dragonfly wings and airfoil profiles at significantly reduced structural weight. The reconstructed geometry also included naturally occurring triangular prismlike serrated structures at the leading edge of the wing, which have comparable effects to micro vortex generators and might stabilize the recirculation zones. Further work aims to investigate the aerodynamic properties of a complete dragonfly wing during wing flapping.
High Spatial Resolution Tomo-PIV of the Nasopharynx Focussing on the Physiological Breathing Cycle
(2022)
Investigations of complex patient-specific flow in the nasopharynx requires high resolution numerical calculations validated by reliable experiments. When building the validation base and the benchmark
of computational fluid dynamics, an experimental setup of the nasal airways was developed. The applied optical measurement technique of tomo-PIV supplies information on the governing flow field in three dimensions. This paper presents tomo-PIV measurements of the highly complex patient-specific geometry of the human trachea. A computertomographic scan of a person’s head builds the basis of the experimental silicone model of the nasal airways. An optimised approach for precise refractive index matching avoids optical distortions even in highly complex non-free-of-sight 3D geometries. A linear-motor-driven pump generates breathing scenarios, based on measured breathing cycles. Adjusting of the CCD cameras‘ double-frame-rate PIV-Δt enables the detailed analysis of flow structures during different cycle phases. Merging regions of interest enables high spatial resolution acquisition of the flow field.
Surgical smoke has been a little discussed topic in the context of the current pandemic. Surgical smoke is generated during the cauterization of tissue with heat-generating devices and consists of 95% water
vapor and 5% cellular debris in the form of particulate matter. In-vivo investigations are performed during tracheotomies where surgical smoke is produced during tissue electrocautery. Furthermore, in-vitro parametric studies to investigate the particle number and size distribution and the spatial distribution of surgical smoke with laser light sheet technique are conducted. The higher the power of the high-frequency-device the larger the particles in size and the higher the resulting particle counts. The images taken show the densest smoke at 40W with artificial saliva. The resulting characteristic size distribution, which may include viruses and bacterial components, confirms that the risk arising from surgical smoke should be considered. Furthermore, the experiments will provide the database for further numerical investigations.