Research in this area focuses on addressing critical challenges in prenatal diagnostics, specifically regarding the detection of Placenta Accreta Spectrum (PAS). PAS is a severe condition associated with a high mortality risk, where timely diagnosis is crucial for the outcome; however, studies indicate that approximately 30% of cases are currently missed during ultrasound screenings.
VR Ultrasound Simulation. We are developing a virtual reality (VR) system that simulates the ultrasound examination of pregnant women. The objective is to create a safe educational environment where gynecologists can gain experience with rare and complex cases, which are often difficult to access in standard clinical practice. This approach directly aims to reduce the rate of missed diagnoses by enhancing physician experience and awareness.
Generative AI for Medical Image Synthesis. To enable training on a diverse range of pathologies, we are developing innovative methods for the synthetic generation of realistic ultrasound images from Magnetic Resonance Imaging (MRI) data. Utilizing advanced diffusion models and Generative Adversarial Networks (GANs), our system learns to map detailed 3D information from MRI into 2D ultrasound views. This process maintains clinical accuracy and characteristic visual features, such as speckle noise. These synthetic datasets serve to enrich data repositories and train new diagnostic models.
Used Technologies
-
Virtual Reality (VR) for medical simulation
-
Ultrasound view generation from MRI modalities (Image-to-Image translation)
-
Generative models (Diffusion models, GANs)
-
Advanced medical image segmentation (Deep Learning)
-
Statistical modeling of ultrasound noise
-
Perinatal education and training systems
