Johannes B. Thalhammer

I am a doctoral candidate at the Chair of Biomedical Physics at the Technical University of Munich (TUM) and the Klinikum Rechts der Isar.

My research focuses on deep learning for medical X-ray imaging, in particular on reducing radiation dose in computed tomography: artifact reduction for sparse-view CT and self-supervised denoising for low-dose phase-contrast CT.

Email/LinkedIn/GitHub

Portrait of Johannes B. Thalhammer

Research

First-author publications

Low-dose (15 ms) CT slice before denoising The same slice denoised with Neighbor2Inverse Denoised Low dose
Neighbor2Inverse: Self-Supervised Denoising for Low-Dose ROI Phase Contrast CT

Johannes B. Thalhammer, Lorenzo D'Amico, Lucy Costello, Sebastian Peterhansl, Tina Dorosti, Florian Schaff, Jannis Ahlers, Ronan Smith, Marcus Kitchen, Franz Pfeiffer, Martin Donnelley, Daniela Pfeiffer, Kaye S. Morgan

Under submission

project page/code

Each noisy projection is split into two subsampled copies with independent noise. Both are reconstructed, and a denoiser is trained on the resulting slice pairs, with no high-dose reference scans required.

arXiv 2026
Artifact Reduction in Undersampled 3D Cone-Beam CTs using a Hybrid 2D-3D CNN Framework

Johannes Thalhammer, Tina Dorosti, Sebastian Peterhansl, Daniela Pfeiffer, Franz Pfeiffer, Florian Schaff

arXiv preprint, 2026

arXiv/code

A 2D U-Net extracts per-slice features and a 3D decoder fuses them across the volume, giving slice-consistent artifact reduction at low computational cost.

J. Imaging Inform. Med. 2025
Ultra-Low-Dose CTPA Using Sparse Sampling CT Combined with the U-Net for Deep Learning-Based Artifact Reduction: An Exploratory Study

Andreas Philipp Sauter*, Johannes Thalhammer*, Felix Meurer, Tina Dorosti, Daniel Sasse, Jessica Ritter, Yannik Leonhardt, Franz Pfeiffer, Florian Schaff, Daniela Pfeiffer (* equal contribution)

Journal of Imaging Informatics in Medicine, 2025

paper

A dual-frame U-Net removes sparse-sampling artifacts from CT pulmonary angiograms, improving image quality, reader confidence, and diagnostic performance, which supports substantial dose reduction.

Plot of hemorrhage detection AUC-ROC versus number of views: U-Net stays near full-view performance while FBP and TV drop at sparse views
Improving Automated Hemorrhage Detection at Sparse-View CT via U-Net–Based Artifact Reduction

Johannes Thalhammer, Manuel Schultheiß, Tina Dorosti, Tobias Lasser, Franz Pfeiffer, Daniela Pfeiffer, Florian Schaff

Radiology: Artificial Intelligence, 2024

paper/code

U-Net post-processing of sparse-view cranial CT keeps automated hemorrhage detection close to full-view performance while using 16× fewer projection views.