Jakub Ceranka | Searcharxiv

arXiv2026

Topogram-Gated Multi-Modal Pseudo-CT Synthesis for PET/MR Attenuation Correction

As part of the Big Cross-Modal Attenuation Correction Challenge (BIC-MAC), we aimed to synthesise pseudo-CT images for PET/MR attenuation correction from whole-body non-attenuation-corrected PET (NAC-PET), Dixon MRI, and a two-dimensional topogram image. We developed a residual 3D U-Net in which a dedicated 2D topogram encoder provides bounded, multi-scale feature gating. The network was supervised jointly in CT and 511-keV attenuation-map space. Two complementary models, one emphasizing activity and cranial regions and one including a projection-domain attenuation-correction-factor regulariser, were combined by equal averaging in CT space. In controlled development experiments, the two training objectives yielded complementary error profiles. On the online BIC-MAC validation leaderboard, the ensemble achieved a mumap MAE of 0.005656, whole-body SUV MAE of 0.0356, organ bias of 2.65%, and brain outlier score of 0.0294. Topogram-gated multi-modal learning combines local volumetric information with global projected anatomy and provides a practical, physically informed approach to pseudo-CT synthesis for PET/MR.

physics.med-ph↗

arXiv2025

Learning from Anatomy: Supervised Anatomical Pretraining (SAP) for Improved Metastatic Bone Disease Segmentation in Whole-Body MRI

The segmentation of metastatic bone disease (MBD) in whole-body MRI (WB-MRI) is a challenging problem. Due to varying appearances and anatomical locations of lesions, ambiguous boundaries, and severe class imbalance, obtaining reliable segmentations requires large, well-annotated datasets capturing lesion variability. Generating such datasets requires substantial time and expertise, and is prone to error. While self-supervised learning (SSL) can leverage large unlabeled datasets, learned generic representations often fail to capture the nuanced features needed for accurate lesion detection. In this work, we propose a Supervised Anatomical Pretraining (SAP) method that learns from a limited dataset of anatomical labels. First, an MRI-based skeletal segmentation model is developed and trained on WB-MRI scans from healthy individuals for high-quality skeletal delineation. Then, we compare its downstream efficacy in segmenting MBD on a cohort of 44 patients with metastatic prostate cancer, against both a baseline random initialization and a state-of-the-art SSL method. SAP significantly outperforms both the baseline and SSL-pretrained models, achieving a normalized surface Dice of 0.76 and a Dice coefficient of 0.64. The method achieved a lesion detection F2 score of 0.44, improving on 0.24 (baseline) and 0.31 (SSL). When considering only clinically relevant lesions larger than 1~ml, SAP achieves a detection sensitivity of 100% in 28 out of 32 patients. Learning bone morphology from anatomy yields an effective and domain-relevant inductive bias that can be leveraged for the downstream segmentation task of bone lesions. All code and models are made publicly available.

eess.IV↗