Francesca Lizzi

Preoperative radiotherapy and surgery are standard treatments for high-risk soft tissue sarcomas (STS), yet distant metastasis remains a significant clinical challenge. This study aims to develop a robust and interpretable multi-omics predictive model for metastasis, specifically addressing the limitations of small datasets. A retrospective analysis included 40 STS patients (23 with metastasis) treated with preoperative radiotherapy and surgery. Using PyRadiomics, 107 radiomic and dosiomic features were extracted from CT scans, dose distributions, and post-treatment contrast-enhanced T1-weighted MRI. A Logistic Regression classifier with L1 regularization was identified as optimal and rigorously evaluated via Stratified 5-Fold Cross-Validation repeated 100 times. Single-modality models demonstrated limited predictive performance, yielding mean AUC-ROC values of 0.52 for CT, 0.59 for dose distribution, and 0.62 for normalized MRI. Integrating these features into a combined multi-omics signature substantially improved the mean AUC-ROC to 0.75. Furthermore, model explainability was ensured using Log-Odds to identify key predictive features. Combining anatomical, dosiomic, and MRI-derived features significantly enhances the prediction of metastatic relapse in STS patients. This interpretable approach shows strong potential for guiding personalized clinical treatments.

We propose an ensemble learning approach to classify malignant versus benign breast lesions leveraging morphological and dynamic features derived from Magnetic Resonance Images (MRI). The analysis has been performed on 164 breast lesions of the publicly available “Advanced MRI Breast Lesions” dataset from The Cancer Imaging Archive, containing T2-weighted and Dynamic Contrast-Enhanced (DCE)-MRI sequences, along with the segmentation masks of suspicious lesions. After extracting radiomic features using Pyradiomics Python package, we computed dynamic features from DCE-MRI kinetic curves, which describe the contrast agent wash-in and wash-out. These features have been defined as the derivatives of image intensity measures, like mean and standard deviation, computed inside the masks on the 5 DCE-MRI time steps. We trained and evaluated an eXtreme Gradient Boosting (XGBoost) classifier, experimenting with different combinations of features and segmentation masks in a stratified 5-fold cross-validation. The best model trained on T2-weighted MRI morphological features achieved an Area Under the Curve (AUC) score of 0.83±0.04 on the independent test set consisting of 20 lesions, while the model using only dynamic features performed an AUC of 0.91±0.03. Despite being obtained on a small test sample, these results show the potential of features derived from DCE images in breast lesions classification.

Accurate segmentation of the pancreas in Computed Tomography (CT) scans is a challenging task, which may be crucial for the diagnosis and treatment planning of pancreatic cancer. The irregular shape of the pancreas, its low contrast relative to surrounding tissues, and its close proximity to other complex anatomical structures make it difficult to accurately delineate its contours. Furthermore, a significant class imbalance between foreground (pancreas) and background voxels in an abdominal CT series represents an additional challenge for deep-learning-based approaches. In this study, we developed a deep learning model for automated pancreas segmentation based on a 3D U-Net architecture enhanced with an attention mechanism to improve the model capability to focus on relevant anatomical features of the pancreas. The model was trained and evaluated on two widely used benchmark datasets for volumetric segmentation, the NIH Healthy Pancreas-dataset and the Medical Segmentation Decathlon (MSD) pancreas dataset. The proposed attention-guided 3D U-Net achieved a Dice score of 80.8 ± 2.1%, ASSD of 2.1 ± 0.3 mm, and HD₉₅ of 8.1 ± 1.6 mm on the NIH dataset, and the values of 78.1 ± 1.1%, 3.3 ± 0.3 mm, and 12.3 ± 1.5 mm for the same metrics on the MSD dataset, demonstrating the value of attention mechanisms in improving segmentation performance in complex and low-contrast anatomical regions.

Artificial intelligence (AI) is widely expected to transform healthcare, yet its adoption in clinical practice remains limited. This paper examines the perspectives of Italian clinicians and medical physicists on the drivers of and barriers to AI use. Using an online survey of healthcare professionals across different domains, we find that efficiency gains—such as reducing repetitive tasks and accelerating diagnostics—are the strongest incentives for adoption. However, trust in AI systems, explainability, and the limited availability of AI tools are major obstacles. Respondents emphasized that AI should augment, not replace, medical expertise, calling for participatory development processes where clinicians are actively involved in the design and validation of decision support tools. At the organizational level, the adoption of AI tools is facilitated by innovation-oriented leadership and sufficient resources, while conservative management and economic constraints hinder implementation. The awareness of regulatory frameworks, including the EU AI Act, is moderate, and many clinicians express the need for targeted training to support safe integration. Our findings suggest that the successful adoption of AI in healthcare will depend on building trust through transparency, clarifying legal responsibilities, and fostering organizational cultures that support collaboration between humans and AI. The role of AI in medicine is therefore best understood as a complement to clinical judgment, rather than a replacement.

Background/Objectives: Glioblastoma (GBM) is the most malignant subtype of glioma and shows the poorest prognosis with a median survival time of 15 months. The methylation status of the Methylguanine-DNA Methyltransferase (MGMT) was proven to be a crucial factor in selecting the most appropriate therapy. Currently, it is assessed through brain biopsy, which is a highly invasive and very expensive technique. For these reasons, in recent years, the possibility of inferring this information from multi-parametric Magnetic Resonance Imaging (mpMRI) has been widely explored. However, substantial differences in performance are reported in the literature. Methods: In this study, we developed several models based on either radiomic or deep learning approaches and a mixture of them using mpMRI for the MGMT status assessment using the public dataset UPENN-GBM, available on The Cancer Imaging Archive. Despite the tests performed using all MRI acquisitions and different methodological approaches, we did not obtain sufficiently reliable performance to direct the therapeutic path of patients. We thus investigated the impact of segmentation quality on MGMT status prediction since the UPENN-GBM dataset contains both automatic and manual refined segmentation masks. Results: We found that performance obtained through radiomic features computed on manually segmented tumors was significantly higher compared to that obtained using automatic segmentation, even when the differences between segmentation masks, measured in terms of Dice Similarity Coefficient (DSC), is not significantly different. Conclusion: This could be the reason why very different MGMT classification performance is typically reported and suggests the creation of a benchmark dataset, with high-quality segmentation masks.

Although Artificial Intelligence (AI) has proven to be valuable in healthcare, its application in clinical practice is still not widespread. We conducted an online survey among clinicians and medical physicists from Italian professional associations to investigate the perceived drivers and barriers to the use of AI in medical practice. Increased efficiency is the main driver for physicians to use AI systems, while the main barriers are related to trust and availability in clinical settings. Efficiency is linked to saving time in task management, but this requires trust in the technology. However, doctors believe that AI cannot replace their knowledge and ask to be involved in the development of the socio-technical system. At the organisational level, the slow adoption revolves around two main criticalities: their economic resources and the attitude of their management. Large centres led by innovation-driven management tend to favour the introduction of AI, while small and less wealthy centres will find it more difficult to integrate AI.