Computational Patient Models

headed by Dr. Kristina Giske

We are physicists, medical informaticians, computer scientists, mathematicians, and radiooncologists working together for the benefit of the patient.

You also want to join our team within your →thesis project? Get in touch

focussing to

Our research is dedicated to the in-silico radiotherapy simulation of the cancer patient receiving curative treatment by irradiation. The challenge for technically precise radiation therapy is the adaptation of dose delivery to deformations in the patient’s anatomy, which are induced by motion and physiological changes. Our vision is to design a virtual patient model or a digital twin, which is serving as a proxy for treatment simulation accumulating the physical dose and predicting the therapeutic effect of the planned treatment. Thus, the treating physicians are empowered to tailor the treatment to the specific needs of the patient prior to irradiation. Algorithm development, emerging computer technology, and our passion for unusual solutions are our tools in the fight against cancer.

working on

• Utilzing CBCTs for dose computation (phantom measurements, CBCT reconstruction, dose re-optimization, image registration, CNN style transfer, cycleGANs)
  
• Clinical Target Volume definition (image segmentation, inter-observer variability, expert guidelines, expert conformance, ANNs, nn-Unet, neuro-symbolic AI) 
• MR-guided treatment and MR-only treatment planning (MR-DECT imaging studies, CNN style transfer, segmentation & regression, deformable image regsitration)
• Biomechanical motion modelling (Multi-body phsics, Finite Element models)
• CT generation and extrapolation of anatomy (Statistical models, Data generators, Diffusion models)
• GPGPU parallelisation for image processing algorithms (CUDA kernels, multi-GPU scheduling)
• Voxelized, tessellated, and analytical patient representation (Dosimetric impact for small structures)

looking forward to

We are proud to co-organize the 5^th Summer School in Medical Physics 2023: Data Science and Machine Learning in Radiotherapy. Lecture contributions by Kristna Giske and young investigators Alexandra Walter and Pedro Rodrigues. follow me for more details...

Accurate motion estimation between medical scans of different modalities (e.g. CT, MRI) is crucial for adaptive radiation therapy. Broadly used intesity-based deformable image registration is limited by missing physical constraints during the deformation. The KinematicDIR project studies the use of a biomechanical model for biofidelic image registration. For head and neck area, it is focussing on a kinematically articulated skeleton model as driving motion model.

In this BMBF*-funded doctoral project Cornelius Bauer investigates the best approach to incorporate the body model into the registration task and assesses the accuracy and robustness - but also limitations - of the model-based approach. follow me for more details...

*ARTEMIS project in collaboration with UKHD & HIT (Jürgen Debus, Oliver Jäkel)

based on PuppetMaster preceding projects by Hendrik Teske and Kathrin Bartelheimer

Stopping power ratio (SPR) maps are needed for dose deposition calculations in ion beam cancer treatments and are typically estimated from single energy CT (SECT) in clinical routine. Dual-energy CT (DECT) involving the acquisition of two energy spectra captures both material-specific information and tissue characterization which is promising to improve patient-tailored SPR map conversion.

In this BMBF*-funded doctoral project Ama Katseena Yawson scrutinizes the capability of deep learning models to convert MRI scans to pseudo-CTs or pseudo-SPR maps. follow me for more details..

*ARTEMIS project in collaboration with UKHD & HIT (Jürgen Debus, Oliver Jäkel)

associated project by Nora Wolf

Patient-tailored contours of target volumes are fundamental for radiation treatment planning and thus, the outcome of the cancer treatment. Clinical target volume delineation on planning CT scans is challenging for human experts, extremely time consuming, and shows large variation between observers. State-of-the-art machine learning algorithms reach high accuracy on the automatic segmentation of anatomical structures which is not transferable to target volumes without additional constraints. The translation of the expert guidelines into the machine learning realm can advance automated target volume delineation to facilitate its guideline conformance and its clinical use.

In this HIDSS4Health-funded doctoral project Alexandra Walter tackles the implications of supervised learning on data prone to inter-observer variabilities and patient-individual differences utilizing mathematical rules given by human experts. follow me for more details...

collaboration with Martin Frank (KIT)

associated project by Jakob Kreft

CBCT imaging has become an unavoidable part of photon radiotherapy devices. However, its image quality is limited in its capabilities to patient positioning. The main aim of the project is bridging the gap towards the image quality of planning CTs, thereby allowing for on-couch treatment plan adaptation based on CBCTs only.

In this Varian-funded doctoral project Goran Stanic investigates the impact of typical image quality shortcoming of CBCT on plan quality degradation. Additionally, generative deep learning models are used to generate synthetic CTs from the lower quality CBCT images. follow me for more details..

collaboration with Niklas Wahl (E0404) (PI: Oliver Jäkel)

associated project by Mark Arndt

Reseach progress in medicine is grounded on patient-sensitive data. Especially, radiation therapy combines most diverse and most frequent imaging necessity for each patient within its precise image-guided adaptive treatment options. Sharing of such rich connected data complying with ethical and legal requirements needs involved preparations in long-term projects. Technological short-term projects cannot share in-house data such that competing algorithm designs cannot be directly compared. A data cohort of non-existent but patient-representative radiotherapy-relevant scans could be made accessible openly for radiotherapy research.

In this AI Health Innovation Cluster-funded infrastructural project Pedro Rodrigues scrutinizes cutting-edge AI image generator technologies for their capability to mimic medical scans with consistent realistic patient anatomies. follow me for more details..

collaboration with Jens Fleckenstein (UMM)

associated project by Mikulas Bankovic