Multimodal Learning from Pre- and Postoperative Imaging for Orthopedic Surgery
Background:

Accurate planning and evaluation of orthopaedic procedures such as knee arthroplasty are essential for optimal surgical outcomes. Radiographs remain the most widely used and accessible imaging modality in orthopaedic diagnostics and follow-up. However, the interpretation of radiographs and the associated planning or validation steps often depend heavily on the experience of clinicians and can vary across institutions. In particular, the selection of appropriate implant size and the postoperative identification of implanted devices are tasks that require precision and consistency.

With the increasing availability of digital imaging and structured metadata in medical formats, modern machine learning algorithms offer new ways to support decision-making in clinical orthopaedics. These techniques can potentially assist clinicians in both preoperative planning and postoperative assessment while maintaining a high level of automation and reproducibility. In this project, both pre- and postoperative radiographic data will be available and shall be jointly analyzed to improve prediction and validation performance through integrated machine learning models.

Your tasks:
  • Developing machine learning models for orthopedic planning assessment 
  • Handling real-world multimodal medical data 
  • Dealing with uncertainty, bias, and variability in clinical datasets 
  • Visualizing, presenting, and interpreting model outputs in a clinical context
  • Further details to be discussed.
What we offer:
  • Unique medical data with high potential for publication 
  • Highly educated & interdisciplinary environment 
  • Top level hardware for scientific computing 
  • Regular feedback from medical and computer science experts
Project details:

Please send a short CV and a recent transcript of records with your inquiry to florian.hinterwimmer@tum.de .

References:
  • Wilhelm, N.J., et al. (2024). Multicentric development and validation of a multi-scale and multi-task deep learning model for comprehensive lower extremity alignment analysis. *Artificial Intelligence in Medicine, 150*, 102843. 
  • Rajpurkar P, Irvin J, Ball RL, et al. (2020). Deep Learning for Radiographic Detection of Osteoarthritis. *Radiology*. 
  • Tiulpin A, Klein S, Bierma-Zeinstra S, Thevenot J, Saarakkala S. (2019). Deep Learning for Predicting Knee Replacement from Radiographs. *Sci Rep*.
Florian Hinterwimmer
Florian Hinterwimmer