KI-Glossar
Das vollständige Wörterbuch der Künstlichen Intelligenz
Score Matching Denoising
Fundamental technique underlying diffusion models, where the neural network learns to predict the noise added to an image at each step, thereby enabling reconstruction of the original image by reversing the noising process.
Latent Diffusion Model (LDM)
Diffusion model architecture that operates in a lower-dimensional latent space, obtained via an auto-encoder, in order to significantly reduce computational costs while preserving generation quality.
Medical Image Synthesis
Application of diffusion models to create realistic new medical images (MRI, CT scans, etc.), used for augmenting training datasets, protecting patient privacy, or simulating rare pathologies.
Medical Image Super-Resolution
Process of enhancing the spatial resolution of low-quality medical images using diffusion models, allowing fine details essential for more accurate diagnosis to be revealed.
Medical Image Denoising
Use of diffusion models to eliminate specific noise (e.g., Poisson noise, Rician noise) present in medical imaging acquisitions such as MRI or CT scans, thereby improving the signal-to-noise ratio.
Medical Inpainting
Technique consisting of reconstructing or replacing missing or corrupted regions in a medical image, using a diffusion model to fill in gaps in a manner consistent with the surrounding anatomical context.
Continuous-Time Diffusion Model
Advanced approach that treats the diffusion process as a continuous stochastic process in time, offering greater flexibility and better sampling efficiency compared to discrete-step models.
U-Net Architecture for Diffusion
U-shaped neural network structure, composed of an encoder path and a decoder path with skip connections, commonly used as the backbone of diffusion models to efficiently predict noise at each step.
Diagnostic Fidelity
Specific evaluation criterion for medical imaging that measures the ability of a synthesized or diffusion model-enhanced image to preserve clinically relevant features for correct diagnosis.
Anatomical Consistency
Principle according to which medical images generated or modified by a diffusion model must respect the known anatomical constraints and relationships of the human body to be clinically valid.
Domain Adaptation for Diffusion
Technique aiming to adapt a diffusion model pre-trained on one image domain (e.g., natural photos) to a specific target domain such as medical imaging, often with limited additional training data.