Artificial intelligence (AI) is rapidly transforming radiology, with applications spanning disease detection, lesion segmentation, workflow optimization, and report generation. As these tools become more integrated into clinical practice, new concerns have emerged regarding their vulnerability to adversarial attacks. This review provides an in-depth overview of adversarial AI in radiology, a topic of growing relevance in both research and clinical domains. It begins by outlining the foundational concepts and model characteristics that make machine learning systems particularly susceptible to adversarial manipulation. A structured taxonomy of attack types is presented, including distinctions based on attacker knowledge, goals, timing, and computational frequency. The clinical implications of these attacks are then examined across key radiology tasks, with literature highlighting risks to disease classification, image segmentation and reconstruction, and report generation. Potential downstream consequences such as patient harm, operational disruption, and loss of trust are discussed. Current mitigation strategies are reviewed, spanning input-level defenses, model training modifications, and certified robustness approaches. In parallel, the role of broader lifecycle and safeguard strategies are considered. By consolidating current knowledge across technical and clinical domains, this review helps identify gaps, inform future research priorities, and guide the development of robust, trustworthy AI systems in radiology.

The development of larger models for medical image analysis has led to increased performance. However, it also affected our ability to explain and validate model decisions. Models can use non-relevant parts of images, also called spurious correlations or shortcuts, to obtain high performance on benchmark datasets but fail in real-world scenarios. In this work, we challenge the capacity of convolutional neural networks (CNN) to classify chest X-rays and eye fundus images while masking out clinically relevant parts of the image. We show that all models trained on the PadChest dataset, irrespective of the masking strategy, are able to obtain an area under the curve (AUC) above random. Moreover, the models trained on full images obtain good performance on images without the region of interest (ROI), even superior to the one obtained on images only containing the ROI. We also reveal a possible spurious correlation in the Chákṣu dataset while the performances are more aligned with the expectation of an unbiased model. We go beyond the performance analysis with the usage of the explainability method SHAP and the analysis of embeddings. We asked a radiology resident to interpret chest X-rays under different masking to complement our findings with clinical knowledge.

Breast cancer represents a significant global health challenge, which makes it essential to detect breast cancer early and accurately to improve patient prognosis and reduce mortality rates. However, traditional diagnostic processes relying on manual analysis of medical images are inherently complex and subject to variability between observers, highlighting the urgent need for robust automated breast cancer detection systems. While deep learning has demonstrated potential, many current models struggle with limited accuracy and lack of interpretability. This research introduces the Deep Neural Breast Cancer Detection (DNBCD) model, an explainable AI-based framework that utilizes deep learning methods for classifying breast cancer using histopathological and ultrasound images. The proposed model employs Densenet121 as a foundation, integrating customized Convolutional Neural Network (CNN) layers including GlobalAveragePooling2D, Dense, and Dropout layers along with transfer learning to achieve both high accuracy and interpretability for breast cancer diagnosis. The proposed DNBCD model integrates several preprocessing techniques, including image normalization and resizing, and augmentation techniques to enhance the model's robustness and address class imbalances using class weight. It employs Grad-CAM (Gradient-weighted Class Activation Mapping) to offer visual justifications for its predictions, increasing trust and transparency among healthcare providers. The model was assessed using two benchmark datasets: Breakhis-400x (B-400x) and Breast Ultrasound Images Dataset (BUSI) containing 1820 and 1578 images, respectively. We systematically divided the datasets into training (70%), testing (20%,) and validation (10%) sets, ensuring efficient model training and evaluation obtaining accuracies of 93.97% for B-400x dataset having benign and malignant classes and 89.87% for BUSI dataset having benign, malignant, and normal classes for breast cancer detection. Experimental results demonstrate that the proposed DNBCD model significantly outperforms existing state-of-the-art approaches with potential uses in clinical environments. We also made all the materials publicly accessible for the research community at: https://github.com/romzanalom/XAI-Based-Deep-Neural-Breast-Cancer-Detection .

Deep learning (DL) models for accurate renal tumor characterization may benefit from multi-center datasets for improved generalizability; however, data-sharing constraints necessitate privacy-preserving solutions like federated learning (FL). To assess the performance and reliability of FL for renal tumor segmentation and classification in multi-institutional MRI datasets. Retrospective multi-center study. A total of 987 patients (403 female) from six hospitals were included for analysis. 73% (723/987) had malignant renal tumors, primarily clear cell carcinoma (n = 509). Patients were split into training (n = 785), validation (n = 104), and test (n = 99) sets, stratified across three simulated institutions. MRI was performed at 1.5 T and 3 T using T2-weighted imaging (T2WI) and contrast-enhanced T1-weighted imaging (CE-T1WI) sequences. FL and non-FL approaches used nnU-Net for tumor segmentation and ResNet for its classification. FL-trained models across three simulated institutional clients with central weight aggregation, while the non-FL approach used centralized training on the full dataset. Segmentation was evaluated using Dice coefficients, and classification between malignant and benign lesions was assessed using accuracy, sensitivity, specificity, and area under the curves (AUCs). FL and non-FL performance was compared using the Wilcoxon test for segmentation Dice and Delong's test for AUC (p < 0.05). No significant difference was observed between FL and non-FL models in segmentation (Dice: 0.43 vs. 0.45, p = 0.202) or classification (AUC: 0.69 vs. 0.64, p = 0.959) on the test set. For classification, no significant difference was observed between the models in accuracy (p = 0.912), sensitivity (p = 0.862), or specificity (p = 0.847) on the test set. FL demonstrated comparable performance to non-FL approaches in renal tumor segmentation and classification, supporting its potential as a privacy-preserving alternative for multi-institutional DL models. 4. Stage 2.

Multi-modal medical image fusion (MMIF) is increasingly recognized as an essential technique for enhancing diagnostic precision and facilitating effective clinical decision-making within computer-aided diagnosis systems. MMIF combines data from X-ray, MRI, CT, PET, SPECT, and ultrasound to create detailed, clinically useful images of patient anatomy and pathology. These integrated representations significantly advance diagnostic accuracy, lesion detection, and segmentation. This comprehensive review meticulously surveys the evolution, methodologies, algorithms, current advancements, and clinical applications of MMIF. We present a critical comparative analysis of traditional fusion approaches, including pixel-, feature-, and decision-level methods, and delves into recent advancements driven by deep learning, generative models, and transformer-based architectures. A critical comparative analysis is presented between these conventional methods and contemporary techniques, highlighting differences in robustness, computational efficiency, and interpretability. The article addresses extensive clinical applications across oncology, neurology, and cardiology, demonstrating MMIF's vital role in precision medicine through improved patient-specific therapeutic outcomes. Moreover, the review thoroughly investigates the persistent challenges affecting MMIF's broad adoption, including issues related to data privacy, heterogeneity, computational complexity, interpretability of AI-driven algorithms, and integration within clinical workflows. It also identifies significant future research avenues, such as the integration of explainable AI, adoption of privacy-preserving federated learning frameworks, development of real-time fusion systems, and standardization efforts for regulatory compliance.

Breast cancer remains a significant global health issue. Despite advances in detection and treatment, its complexity is driven by genetic, environmental, and structural factors. Computational methods like Finite Element Modeling (FEM) have transformed our understanding of breast cancer risk and progression. Advanced computational approaches in breast cancer research are the focus, with an emphasis on FEM's role in simulating breast tissue mechanics and enhancing precision in therapies such as radiofrequency ablation (RFA). Machine learning (ML), particularly Convolutional Neural Networks (CNNs), has revolutionized imaging modalities like mammograms and MRIs, improving diagnostic accuracy and early detection. AI applications in analyzing histopathological images have advanced tumor classification and grading, offering consistency and reducing inter-observer variability. Explainability tools like Grad-CAM, SHAP, and LIME enhance the transparency of AI-driven models, facilitating their integration into clinical workflows. Integrating FEM and ML represents a paradigm shift in breast cancer management. FEM offers precise modeling of tissue mechanics, while ML excels in predictive analytics and image analysis. Despite challenges such as data variability and limited standardization, synergizing these approaches promises adaptive, personalized care. These computational methods have the potential to redefine diagnostics, optimize treatment, and improve patient outcomes.

The application of AI for predicting critical heart failure endpoints using echocardiography is a promising avenue to improve patient care and treatment planning. However, fully supervised training of deep learning models in medical imaging requires a substantial amount of labelled data, posing significant challenges due to the need for skilled medical professionals to annotate image sequences. Our study addresses this limitation by exploring the potential of self-supervised learning, emphasising interpretability, robustness, and safety as crucial factors in cardiac imaging analysis. We leverage self-supervised learning on a large unlabelled dataset, facilitating the discovery of features applicable to a various downstream tasks. The backbone model not only generates informative features for training smaller models using simple techniques but also produces features that are interpretable by humans. The study employs a modified Self-supervised Transformer with Energy-based Graph Optimisation (STEGO) network on top of self-DIstillation with NO labels (DINO) as a backbone model, pre-trained on diverse medical and non-medical data. This approach facilitates the generation of self-segmented outputs, termed "parcels", which identify distinct anatomical sub-regions of the heart. Our findings highlight the robustness of these self-learned parcels across diverse patient profiles and phases of the cardiac cycle phases. Moreover, these parcels offer high interpretability and effectively encapsulate clinically relevant cardiac substructures. We conduct a comprehensive evaluation of the proposed self-supervised approach on publicly available datasets, demonstrating its adaptability to a wide range of requirements. Our results underscore the potential of self-supervised learning to address labelled data scarcity in medical imaging, offering a path to improve cardiac imaging analysis and enhance the efficiency and interpretability of diagnostic procedures, thus positively impacting patient care and clinical decision-making.

Deep learning techniques have significantly enhanced the convenience and precision of ultrasound image diagnosis, particularly in the crucial step of lesion segmentation. However, recent studies reveal that both train-from-scratch models and pre-trained models often exhibit performance disparities across sex and age attributes, leading to biased diagnoses for different subgroups. In this paper, we propose APPLE, a novel approach designed to mitigate unfairness without altering the parameters of the base model. APPLE achieves this by learning fair perturbations in the latent space through a generative adversarial network. Extensive experiments on both a publicly available dataset and an in-house ultrasound image dataset demonstrate that our method improves segmentation and diagnostic fairness across all sensitive attributes and various backbone architectures compared to the base models. Through this study, we aim to highlight the critical importance of fairness in medical segmentation and contribute to the development of a more equitable healthcare system.

Deep neural networks excel in medical imaging but remain prone to biases, leading to fairness gaps across demographic groups. We provide the first systematic exploration of Human-AI alignment and fairness in this domain. Our results show that incorporating human insights consistently reduces fairness gaps and enhances out-of-domain generalization, though excessive alignment can introduce performance trade-offs, emphasizing the need for calibrated strategies. These findings highlight Human-AI alignment as a promising approach for developing fair, robust, and generalizable medical AI systems, striking a balance between expert guidance and automated efficiency. Our code is available at https://github.com/Roypic/Aligner.

Integrating deep learning into clinical workflows for medical image analysis holds promise for improving diagnostic accuracy. However, strict data privacy regulations and the sensitivity of clinical IT infrastructure limit the deployment of cloud-based solutions. This paper introduces WebIPred, a web-based application that loads deep learning models directly within the client's web browser, protecting patient privacy while maintaining compatibility with clinical IT environments. WebIPred supports the application of pre-trained models published on Zenodo and other repositories, allowing clinicians to apply these models to real patient data without the need for extensive technical knowledge. This paper outlines WebIPred's model integration system, prediction workflow, and privacy features. Our results show that WebIPred offers a privacy-protecting and flexible application for image classification, only relying on client-side processing. WebIPred combines its strong commitment to data privacy and security with a user-friendly interface that makes it easy for clinicians to integrate AI into their workflows.