Studierfenster

Studierfenster provides client-server processing and analysis of (bio-)medical imaging data including CT and MRI for visualization, segmentation, landmark detection, cranial implant design, and aortic inpainting using convolutional neural networks (CNNs) and generative adversarial networks (GANs).

Key Features:

• Visualization: Visualizes medical data in two-dimensional (2D) and three-dimensional (3D) formats, including virtual reality (VR) environments and augmented reality (AR) applications such as facial reconstruction and registration.
• Manual annotation: Supports manual slice-by-slice outlining of anatomical structures and placement of landmarks within imaging data.
• Automated analysis: Implements a CNN for automatic cranial implant design, a GAN for inpainting aortic dissections, and a CNN-based method for automatic detection of aortic landmarks in CT angiography images.
• Metric calculation: Calculates segmentation accuracy metrics including the dice score and Hausdorff distance.
• Validation: Validates results by comparison to ground truth segmentations performed by a physician at the Medical University of Graz.
• Architecture: Employs a client-server architecture to process 3D volumetric medical data.

Scientific Applications:

• Diagnostic imaging: Applied to CT and MRI data for diagnostics, clinical studies, and treatment planning.
• Surgical planning and reconstruction: Used for cranial implant design and facial reconstruction/registration.
• Vascular analysis: Applied to aortic dissection inpainting and aortic landmark detection in CT angiography.
• Veterinary medicine: Supports processing of CT and MRI scans in animal diagnostics.
• Non-medical image processing: Applicable to optical measuring techniques, astronomy, and archaeology for image or volume processing tasks.

Methodology:

Client-server processing of 3D volumes; convolutional neural networks for cranial implant design and aortic landmark detection; a generative adversarial network for aortic dissection inpainting; computation of dice score and Hausdorff distance; validation by comparison to physician-derived ground truth segmentations.