Hubert Szymon Gabryś

Objectives: We evaluated the performance of mammography-based radiomics in prediction of clinical nodal status, clinical tumor stage, clinical disease stage, status of the PIK3CA mutation and concordance with the radiologist-assessed BI-RADS category. Mammography is often the first imaging modality performed in the screening or diagnostic workup of breast cancer. Materials and Methods: In this single-center retrospective study, we included 102 histopathologically confirmed cases of breast cancer from 100 patients. The tumor region and whole-breast parenchyma were contoured on 368 craniocaudal and mediolateral-oblique mammograms. Cases were temporally split into training and test sets by histopathological diagnosis date (training: 70%, held-out test: 30%). Six linear models were tuned by 5-fold x 3-repeat stratified cross-validation. The winning model per endpoint was applied once to the test set. We report AUCs with 95% confidence intervals, permutation p-values and Benjamini–Hochberg false discovery rate (BH-FDR) correction across five endpoints. Results: Three of the studied endpoints reached significance after BH-FDR correction: clinical nodal status (cN0 vs. cN-positive; AUC: 0.726 (95% CI: 0.501–0.886)), clinical tumor stage (cT1-2 vs. cT3-4; AUC: 0.792 (95% CI: 0.575–0.929)) and overall disease stage (I–II vs. III–IV; AUC: 0.778 (95% CI: 0.572–0.917)). Mammography-based radiomics failed to predict the presence of PIK3CA mutation and concordance with radiologist-assessed BI-RADS category. Conclusions: Mammography-based radiomics has shown a hypothesis-generating discriminative value for differentiation between cN-negative and cN-positive disease, early from advanced clinical tumor stage, and early from advanced overall disease stage in breast cancer. Mammography-based radiomics did not predict PIK3CA status and did not discriminate between radiologist-assessed BI-RADS 4 and 5 groups.

Background and purpose: MR-guided radiotherapy (MRgRT) offers superior soft-tissue contrast and enables adaptive treatments which improve tumor targeting and organ-at-risk sparing. However, integrated MR-Linac systems remain financially and logistically inaccessible for many hospitals. To address this gap, we implemented a modular MRgRT workflow — termed the “Virtual MR-Linac”— which decouples MR imaging and treatment delivery, enabling daily adaptive radiotherapy (ART) using diagnostic-quality MRI. Materials and methods: Between November 2023 and January 2025, 31 male patients with pelvic lymph node metastases were treated using this workflow. Daily MR scans were acquired in treatment position using a 1.5T MR scanner. Contours were adapted on 2D T2 turbo-spin-echo images, while synthetic CTs generated from T1 DIXON sequences were used for dose calculation. Adapted plans were delivered on a C-arm Linac using CBCT imaging for positioning. Quality assurance included independent Monte Carlo dose recalculations, automated plan parameters verification and a plan complexity check using in-house scripting. Results: Across 99 adapted fractions, plan adaptation improved dosimetric outcomes, with a median 14% increase in Planning Target Volume (PTV) coverage and a 7% reduction in bowel D1cc (dose to 1 cc). Median times for MR imaging, contouring, and planning were 12, 12, and 20 min, respectively. Conclusions: The Virtual MR-Linac approach enables high-quality ART to compensate for inter-fractional motion without the costs and complexity of integrated systems. It improves resource utilization, maintains patient comfort, and offers a flexible, scalable solution for clinical implementation.

Objectives We evaluated the performance of mammography-based radiomics in prediction of clinical nodal status, clinical tumor stage, clinical disease stage, status of the PIK3CA mutation and concordance with the radiologist-assessed BI-RADS category. Mammography is often the first imaging modality performed in the screening or diagnostic workup of breast cancer. Materials & Methods In this single-center retrospective study, we included 102 histopathologically-confirmed cases of breast cancer from 100 patients. The tumor region and whole-breast parenchyma were contoured on 368 craniocaudal and mediolateral-oblique mammograms. Cases were temporally split into training and test sets by histopathological diagnosis date (training 70%, held-out test 30%). Six linear models were tuned by 5-fold x 3-repeat stratified cross-validation. The winning model per endpoint was applied once to the test set. We report AUC with 95% confidence intervals, permutation p-values and Benjamini-Hochberg false discovery rate (BH-FDR) correction across five endpoints. Results Three of the studied endpoints reached significance after BH-FDR correction: clinical nodal status (cN0 vs cN-positive; AUC 0.726 (95% CI: 0.501–0.886)), clinical tumor stage (cT1-2 vs cT3-4; AUC 0.792 (95% CI: 0.575–0.929)) and overall disease stage (I-II vs III-IV; AUC 0.778 (95% CI: 0.572–0.917)). Mammography-based radiomics failed to predict the presence of PIK3CA mutation and concordance with radiologist-assessed BI-RADS category. Conclusion Mammography-based radiomics can be used to discriminate between cN-negative and cN-positive disease, early from advanced clinical tumor stage and early from advanced overall disease stage in breast cancer. Mammography-based radiomics did not predict PIK3CA status and did not discriminate between radiologist-assessed BI-RADS 4 and 5 groups.

Background and purpose: Magnetic resonance-guided radiotherapy (MRgRT) offers high-quality soft tissue contrast and treatment monitoring that can be leveraged for longitudinal radiomics analysis. We investigated treatment-related changes in MRgRT-based features during stereotactic body radiotherapy (SBRT) for liver metastases and surrounding healthy liver.Materials and methods: Features were extracted from the gross tumor volume (GTV) and five 2-cm concentric liver rings around the planning target volume (PTV) in 34 patients treated with five-fraction MRgSBRT on the ViewRay MRIdian system. Features below baseline variability and highly correlated (Spearman’s |r|>0.9) were excluded, with pre-treatment values averaged into a robust baseline. Temporal evolution, quantified using linear regression slopes, informed UMAP and HDBSCAN clustering for patient stratification, with false discovery rate-corrected Kruskal–Wallis testing identifying representative features. Correlations with GTV volume and mean cumulative liver-ring dose were evaluated.Results: Of 160 GTV features, 55 uncorrelated features were retained. Three patient groups with distinct longitudinal patterns (n=4, 10, and 20 patients) were identified. Intensity-based energy showed significant longitudinal differences across patient groups (adjusted p=0.001), weakly correlated (|r|<0.7) with volume in the two larger groups. In the innermost liver ring, 49 uncorrelated features were analyzed. Two patient groups (n=19 and 15 patients) were identified, with intensity-based energy, mean intensity, and maximum intensity showing significant longitudinal differences between the groups, weakly correlated with the mean cumulative dose.Conclusions: This longitudinal MR-based radiomics framework identified GTV- and healthy-liver-based patient groups, with distinct treatment-related feature changes. This pipeline can be extended to integrate treatment response assessment in future MRgRT radiomic studies.

Background and Purpose Accurate image-based cancer detection and segmentation remain challenging for artificial intelligence models. Most existing solutions are optimized for specific tumor types or anatomical regions, showing limited generalizability beyond their original domains. In this study, we evaluated whether the winning model of the AutoPET III challenge, trained on metastatic datasets, could demonstrate broader applicability.Materials and MethodsThe AutoPET winning PET/CT cancer segmentation model, trained on images of lymphoma, metastatic melanoma, lung cancer, and healthy control datasets, was validated across internal and external datasets, including melanoma, lung cancer, head and neck cancer, soft tissue sarcoma, breast cancer, and healthy controls. Validation focused on detection and segmentation performance, as well as lesion-wise analyses of volume, glycolysis, and total tumor burden.ResultsModel performance varied depending on cancer type and lesion size. Overall detection rates were highest in metastatic melanoma (recall=0.84, precision=0.75). False-positive findings occurred across all datasets, predominantly in small lesions (1cc) across melanoma, lung cancer, and head and neck cancer (ICC(3,1)≥0.94). Total tumor burden estimation performed best in melanoma (ICC(3,1)=0.86). Only 29% of scans were correctly classified as negative.ConclusionsThe analyzed model demonstrated strong performance in melanoma and moderate performance in lung cancer, while performance in other tumor types was limited. The 71% misclassification rate among healthy controls represents a major limitation.

Objective. We propose a method called N-Peaks that normalizes intensity units in structural magnetic resonance (MR) images by harmonizing the intensities of certain reference tissues. Approach. The N-Peaks normalization requires an image and a number of normal tissue reference contours as input. A map of local intensity change is calculated, which is used to isolate homogeneous regions in each contour. An intensity histogram of the homogeneous voxels is calculated and the peak intensity on that histogram is identified to serve as a landmark. The image intensities are then transformed in a piecewise linear fashion to map the landmark intensities of the normal tissue to predefined target values. The approach was tested for a data set of 194 abdomen images acquired on a 0.35 T MR-Linac. The images were normalized with N-Peaks using the background, manually contoured liver and semi-automatically contoured fat as reference tissues. For comparison, a Nyul normalization and Z -Score normalizations based on the body and liver were applied. The heterogeneity of intensity histograms after each normalization was quantified using the Jensen–Shannon distance (JSD) of individual intensity histograms to the average intensity histogram for the body, liver and fat contours. Main result. The N-Peaks normalization resulted in consistently low JSD in all three tissues. In the body contour, Nyul normalization achieved the lowest JSD, but it resulted in a distorted histogram shape for fat. Z -Score based on the body showed high JSD in fat and liver, and Z -Score based on the liver resulted in the lowest JSD in the liver but high JSD in body and fat. Significance. We demonstrate that N-Peaks successfully normalizes MR image intensities of selected normal tissues, improving the consistency of the intensities regarding tissue types. An implementation of the method is provided, offering potential applications for different quantitative MR image analyses.

Postoperative radiotherapy improves local control (LC) after resection of brain metastases. In comparison to whole brain radiotherapy (WBRT) stereotactic radiosurgery (SRS) to resection cavity significantly reduces cognitive side effects. However, two phase-III trials have reported suboptimal LC with SRS, leading to increased interest in hypofractionated stereotactic radiotherapy (SRT) as an alternative to improve outcomes. This single-centre study, based on a prospective quality assurance protocol, included 161 patients with 185 resected brain metastases treated with either SRS or SRT between February 2018 and June 2023. Patients were assigned to treatment based on cavity size, with SRS typically used for cavities < 10 cc and SRT for larger volumes. Primary and secondary endpoints were LC and radiation necrosis (RN), respectively. Data analysis was conducted retrospectively. Median cavity size was 13.3 cc, with 20% of cavities receiving SRS and 80% SRT. 12-month LC was 92.6% (95-CI: 88.2 – 97.3%), 12-month RN incidence was 9% (95-CI: 3–14%), with RN limited to CTCAE v5 ≤ 2. In cavities < 10 cc, no significant difference in LC was found between SRS and SRT. For cavities > 10 cc, PTV volume was the only significant predictor of LC, while fractionation and dose did not significantly impact outcomes. SRS and SRT both offer excellent LC for resection cavities < 10 cc with low rates of RN, suggesting SRS as the preferred treatment in this collective, in consideration of patient comfort and resource allocation. In larger cavities, PTV volume significantly influences LC. Dose escalation might be beneficial in improving outcomes in these cases.

Antifibrotic therapy with nintedanib is the clinical mainstay in the treatment of progressive fibrosing interstitial lung disease (ILD). High-dimensional medical image analysis, known as radiomics, provides quantitative insights into organ-scale pathophysiology, generating digital disease fingerprints. Here, we performed an integrative analysis of radiomic and proteomic profiles (radioproteomics) to assess whether changes in radiomic signatures can stratify the degree of antifibrotic response to nintedanib in (experimental) fibrosing ILD. Unsupervised clustering of delta radiomic profiles revealed 2 distinct imaging phenotypes in mice treated with nintedanib, contrary to conventional densitometry readouts, which showed a more uniform response. Integrative analysis of delta radiomics and proteomics demonstrated that these phenotypes reflected different treatment response states, as further evidenced on transcriptional and cellular levels. Importantly, radioproteomics signatures paralleled disease- and drug-related biological pathway activity with high specificity, including extracellular matrix (ECM) remodeling, cell cycle activity, wound healing, and metabolic activity. Evaluation of the preclinical molecular response–defining features, particularly those linked to ECM remodeling, in a cohort of nintedanib-treated fibrosing patients with ILD, accurately stratified patients based on their extent of lung function decline. In conclusion, delta radiomics has great potential to serve as a noninvasive and readily accessible surrogate of molecular response phenotypes in fibrosing ILD. This could pave the way for personalized treatment strategies and improved patient outcomes.