Accurate assessment of surgical resection margins is critical for improving oncological outcomes, as incomplete tumor removal is strongly associated with increased risk of recurrence. Conventional intraoperative techniques, such as frozen-section histopathology, are limited by time constraints, sampling bias, and dependence on expert interpretation. X-ray phase-contrast tomography (XPCT) has recently been demonstrated by our group as a powerful imaging modality capable of providing high-resolution, three-dimensional visualization of soft tissue with enhanced contrast, enabling detailed characterization of tumor morphology [1]. However, the large volumetric datasets generated by XPCT pose significant challenges for rapid and objective margin assessment. This project aims to develop an automated computational framework for quantitative resection margin measurement in XPCT data. The objectives are to (i) segment tumor and surrounding healthy tissue in high-resolution 3D volumes, (ii) compute spatially resolved margin distances across complex tumor geometries, and (iii) establish robust metrics for identifying positive or close margins. The project will integrate advanced image processing and machine learning approaches, including convolutional neural networks and hybrid model-based methods, to enable accurate and reproducible tissue classification and boundary detection. The computational workload associated with this project is substantial. XPCT datasets are typically large-scale volumetric images with high spatial resolution, requiring intensive preprocessing, segmentation, and geometric analysis. Training and deploying deep learning models on such data demands significant computational resources, particularly for 3D architectures and large batch processing. Furthermore, efficient margin computation involves voxel-wise distance calculations and surface reconstruction, which benefit significantly from parallelization. Access to GPU-enabled resources within the KTH High-Performance Computing (HPC) infrastructure is essential to accelerate both model training and large-scale data analysis. GPU acceleration will enable efficient handling of volumetric datasets, reduce training times for deep learning models, and support iterative optimization of algorithms. It will also facilitate the development of scalable pipelines for processing multiple datasets and performing parameter sweeps. The expected outcome of this project is a validated, automated pipeline for resection margin assessment in XPCT data, providing high-resolution, quantitative, and reproducible measurements. This approach has the potential to complement or enhance current pathology workflows and contribute to the development of future intraoperative or rapid ex vivo imaging solutions. By combining advanced imaging with GPU-accelerated analysis, this work aims to advance precision surgery and improve clinical decision-making in oncological treatment. Reference 1. William Twengström, Carlos F. Moro, Jenny Romell, Jakob C. Larsson, Ernesto Sparrelid, Mikael Björnstedt, Hans M. Hertz, "Can laboratory x-ray virtual histology provide intraoperative 3D tumor resection margin assessment?," J. Med. Imag. 9(3) 031503 (7 February 2022) https://doi.org/10.1117/1.JMI.9.3.031503