Intraoperative Multi-Sensor Tissue Differentiation in Oncology

  • Funding: German Research Foundation, project ID 40947457.
  • PIs: Oliver Sawodny (University of Stuttgart), Arnulf Stenzl (University Hospital Tübingen)

    🔗 Related Publications
    See Biomedical Engineering Publications

In oncology, the differentiation between healthy and tumorous tissue is crucial for improved diagnostics, complete tumor removal, and better patient outcomes. The current gold standard, the histopathological instantaneous section, comes with a time delay due to the need to remove tissue, transport it to and examine it in a laboratory. Since tumors often differ physically from surrounding tissue, introducing real-time sensor systems in surgery can help guide decisions directly in the OR. A short review of this topic is available here.

My PhD topic was impedance-based tissue differentiation for bladder tumor detection, and most of my works focus on this topic. After finishing my PhD, I have been a group leader within the overall project and supervise projects dealing with impedance sensor arrays, data-driven sensor fusion, and intraoperative navigation and localisation based on endoscopic videos.

Impedance-based tissue classification

Bladder tumor recurrence remains a major challenge in uro-oncology, increasing both risk and cost of patient care. Bladder tumors do not only differ visually from healthy tissue, but also in their electrical properties due to an altered physiology. A possible base for intraoperative tissue differentiation are impedance measurements, which are, however, difficult to obtain passing through the limited space of the urethra.

Highlights of my work are:

  • Developed and validated a minimally invasive impedance sensor (Ø 2 mm) suitable for transurethral insertion during bladder endoscopy (📄 key publication)
  • Created multi-physical tissue models, especially to connect the intertwined mechanical and electrical domains in the context of sensor-based tissue differentiation (📄 key publication).
  • Designed and evaluated a patient-specific machine learning pipeline for tissue classification robust to deformation, tumor types, and radiotherapy artifacts (📄 key publication)

Recent directions within our group extend this work to multi-electrode sensor arrays provinding a spatial resolution of the measurement area.

Sensor fusion

Combining sensors from different physical domains increases prediction robustness and offers redundancy — but comes with challenges such as variable resolution, deformation, or differing acquisition rates.

Key developments:

  • Cross-modal translation and feature alignment, aiming for knowledge transfer from a strong to a weak modality during training to enhance unimodal performance of the weak modality at inference (📄 publication).
  • Entropy-based sensor placement and fusion strategies to combine pointwise impedance measurements with wide-field hyperspectral imaging (submitted to IEEE Sensors 2025)
  • Multi-physical tumor modeling for better understanding and model-based investigations (to be presented at IEEE EMBC 2025)

Intraoperative navigation and localization

Bladder surgery is constrained by limited visibility through endoscopes. Reliable feature extraction is essential for orientation and navigation within the hollow organ. Blood vessels offer a robust and unique structure for geometric mapping.

Key developments:

  • Sparse feature detection and localization in the deformable bladder environment (📄 publication)
  • Improved datasets and segmentation algorithms for blood vessel recognition (to appear at IEEE EMBC 2025)

Last updated: June 2025