The Challenge: Fighting Invisible Threats with Computer Science
Every year, over 4.3 million patients in the EU acquire Healthcare-Associated Infections (HAIs). A major cause is invisible airborne pathogens disrupting the sterile airflow in Operating Rooms (ORs) during surgery. While Computational Fluid Dynamics (CFD) can simulate these risks with high fidelity, it takes supercomputers days to process millions of data points. Surgeons cannot see this data, meaning they often unknowingly break the protective air curtain.
Our Mission: The Delivery Truck for Real-Time Physics at NTNU’s Department of Computer Science (IDI) and St. Olav’s Hospital, we are building an Extended Reality (XR) framework to solve this. As part of the EU Horizon Europe HumanIC project, we are developing the software pipeline that translates heavy engineering simulations into a real-time, interactive "sixth sense" for surgeons, aiming to reduce infection risks by 30% and increasing energy efficiency by 10%.
The Tech Stack You Will Work With: If you join our research group, you will tackle some of the hardest problems in modern computer graphics and AI:
•Next-Gen Digital Twins: Traditional photogrammetry fails on the shiny, reflective stainless steel and glass of medical equipment. We are pushing the boundaries of 3D Gaussian Splatting (3DGS) to capture hyper-realistic, geometrically accurate OR environments instantly using consumer cameras and smartphones.
•Lagrangian-Based Middleware: CFD software (Ansys) and game engines (Unreal) do not speak the same language. We are building custom Python and Blender middleware that extracts raw particle tracking data and "bakes" the physics (velocity, particle age, temperature) directly into mesh vertex attributes.
•High-Performance XR Rendering: We are developing hybrid rendering pipelines that combine lightweight 3D meshes for room architecture with photorealistic 3DGS for medical tools, achieving a motion-sickness-free 72+ FPS on PC-VR (Meta Quest 3 + RTX 4090).
•Real-Time Edge AI: To make airflow predictions instant, we are bypassing traditional physics solvers. We train AI Surrogate Models (Neural Operators) to act as a "Smart
Inbetweener," predicting chaotic, non-linear turbulence in just 5-10 milliseconds directly on the GPU’s Tensor Cores.
Suggested research topics :
1.Tackling "Phantom Geometry" from Reflective OR Equipment in 3DGS (Computer Vision)
2.Enhancing SfM Point Could initialization in Texture-less Surgical Environments (Image Processing)
3.Cognitive Ergonomics and UI/UX Evaluation for Airflow Visualization in AR (HCI)
4.Synthetic Dataset Generation for Medical Digital Twins (Data Engineering)
5.Hardware-Aware AI Surrogate Modeling: Unlocking Asynchronous Compute and FP8 Inference for Real-Time CFD (ML-Computer Architecture)
Supervisors: Gabriel Kiss kiss@ntnu.no Rahmat Rizal Andhi rahmatra@stud.ntnu.no Frank Lindseth frankl@ntnu.no (COMP/IDI)
[ Skjul beskrivelse ]
