Ultrasound Image Analysis for Diagnostic and Therapy Guidance

Ultrasound imaging offers several key advantages that make it highly attractive for both diagnostic and interventional applications: it is real-time capable, non-invasive, portable, and free from ionizing radiation. However, ultrasound data is also inherently challenging to interpret due to its limited field of view, strong speckle noise, and frequent artifacts. These factors significantly complicate automated image analysis and interpretation, especially compared to modalities like CT or MRI.

In our research group, we focus on advancing the field of ultrasound image analysis with a particular emphasis on 3D ultrasound. Our goal is to develop real-time capable, AI-driven algorithms that enable robotic-assisted therapy guidance. To this end, we aim to process and interpret ultrasound images in a way that supports precise and automated decision-making during both diagnostic and therapeutic procedures.

Our work includes developing methods for target tracking, automatic generation of standard views, and intelligent scanning protocols for diagnostic applications. We are particularly interested in challenges such as machine learning-based feature detection and description, representation learning, semantic segmentation, and anomaly detection. By leveraging state-of-the-art approaches in deep learning and real-time data processing, we strive to make ultrasound a more robust and intelligent modality for both clinical and interventional use.

Key Publications

Daniel Wulff, Floris Ernst. Real-time deformable structure tracking in 3D ultrasound sequences using deformable convolutional layers. Comput. Biol. Med. 186 109671. 2025. [web]

Daniel Wulff, Floris Ernst. Analysis of intra- and inter-observer variability in 4D liver ultrasound landmark labeling. J. Med. Imag. 12(5) 051807. 2025. [web]