MIBE Seminar: Self-supervised noise reduction for low dose full interior propagation-based phase-contrast CTs

Titel: Self-supervised noise reduction for low dose full interior propagation-based phase-contrast CTs

Sprecher: Johannes Thalhammer (Lehrstuhl Biomedizinische Physik Prof. Pfeiffer)

Datum und Uhrzeit: Dienstag, 6. Mai 2025, 13:00 Uhr

Abstract: Neural network-based denoising shows great potential for enhancing image quality in low-dose computed tomography (CT). However, training these models usually requires clean target datasets, which are often not available. To tackle this issue, we explore a self-supervised approach inspired by Neighbor2Neighbor, which we adapt for the inverse problem of computed tomography. Our method is applied to full interior propagation-based phase-contrast CT images of excised calf lungs, demonstrating effective denoising without the need for ground truth data.