The Information and Communication Technology sector accounted for approximately 1.4

As artificial intelligence (AI) becomes increasingly integrated into clinical environments, questions of trust, transparency, and shared decision-making come to the fore. This article examines how public perspectives can influence the ethical and technical development of AI tools in medicine, drawing on empirical insights from an interdisciplinary project focused on developing AI to support the diagnosis and treatment of skin cancer. Rather than treating ethical concerns as external to technical design, we argue that they must be addressed from within the development process. In our case, this was achieved by integrating citizen feedback into iterative design loops within our interdisciplinary team, fostering closer alignment between AI functionalities and public values. Through focus group discussions with citizens and a constructivist grounded theory approach, we identified three key areas of concern: the evolving doctor–patient relationship, patient agency in AI-supported care, and the influence of specific medical contexts on public evaluations of AI. This article illustrates how these citizen perspectives can be meaningfully connected with the medical and technical considerations shaping the development of AI.

The statement “Decision Support in Medical Practice through Artificial Intelligence,” (German: “Entscheidungsunterstützung ärztlicher Tätigkeit durch Künstliche Intelligenz”) by the Central Ethics Commission at the German Medical Association critically evaluates the implementation of Clinical Decision Support Systems (CDSS) from medical, ethical, and legal standpoints. While the statement highlights numerous advantages and challenges inherent in the use of AI in healthcare, it remains vague, particularly concerning the assignment of accountability, the delineation of professional responsibility, and the technical feasibility of integrating these systems in real-world clinical workflows. In the present paper, we scrutinize the statement from both philosophical and technical angles, offering concrete proposals aimed at clarifying the ethical ambiguities while also addressing the technological constraints that currently impede widespread adoption. By doing so, we aim to critically evaluate the governmental document against practical realities and bridge the gap between theoretical possibilities and practical applicability, thereby fostering a more nuanced and responsible use of AI in contemporary medical practice.

As artificial intelligence (AI) systems become increasingly integrated into critical domains, ensuring their responsible design and continuous development is imperative. Effective AI quality management (QM) requires tools and methodologies that address the complexities of the AI lifecycle. In this paper, we propose an approach for AI lifecycle planning that bridges the gap between generic guidelines and use case-specific requirements (MQG4AI). Our work aims to contribute to the development of practical tools for implementing Responsible AI (RAI) by aligning lifecycle planning with technical, ethical and regulatory demands. Central to our approach is the introduction of a flexible and customizable Methodology based on Quality Gates, whose building blocks incorporate RAI knowledge through information linking along the AI lifecycle in a continuous manner, addressing AIs evolutionary character. For our present contribution, we put a particular emphasis on the Explanation stage during model development, and illustrate how to align a guideline to evaluate the quality of explanations with MQG4AI, contributing to overall Transparency.

Continuing the digital revolution, AI is capable to transform our world. Thanks to its novelty, we can define how we, as a society, envision this fascinating technology to integrate with existing processes. The EU AI Act follows a risk-based approach, and we argue that addressing the human influence, which poses risks along the AI lifecycle is crucial to ensure the desired quality of the model’s transition from research to reality. Therefore, we propose a holistic approach that aims to continuously guide the involved stakeholders’ mindset, namely developers and domain experts, among others towards Responsible AI (RAI) lifecycle management. Focusing on the development view with regard to regulation, our proposed four pillars comprise the well-known concepts of Generalizability, Adaptability and Translationality. In addition, we introduce Transversality (Welsch in Vernunft: Die Zeitgenössische Vernunftkritik Und Das Konzept der Transversalen Vernunft, Suhrkamp, Frankfurt am Main, 1995), aiming to capture the multifaceted concept of bias, and base the four pillars on Education, and Research. Overall, we aim to provide an application-oriented summary of RAI. Our goal is to distill RAI-related principles into a concise set of concepts that emphasize implementation quality. Concluding, we introduce the ethical foundation’s transition to an applicable ethos for RAI projects as part of on-going research.

Companies increasingly integrate Artificial Intelligence (AI) into their applications to stay competitive. However, the efficient and successful development and deployment of AI applications requires complex setups. Providing developers with all the required resources, applications, and services for developing and deploying AI applications without reinventing the wheel remains challenging. Therefore, we propose a domainand workflow-agnostic reference architecture (RA) for an onpremises AI Platform that supports teams throughout the entire AI lifecycle and is reusable across multiple projects. Additionally, we present an evaluation strategy to validate the RA.

Early recognition of clinical deterioration is crucial for timely intervention, especially during Emergency Medical Services (EMS) encounters. Early Warning Scores (EWS) translate raw vital signs into a clinically transparent risk scale. However, research on prehospital EWS applications is limited and often focuses on in-hospital outcomes and single snapshots, neglecting short-term risk trajectories. This paper explores whether EWS trends, captured just before initial EMS intervention, convey additional information and can predict the return of spontaneous circulation (ROSC) during out-of-hospital cardiac arrest encounters. In a retrospective study of 4,394 cardiac arrest encounters from the 2021-2023 National EMS Information System (NEMSIS), we applied eight different EWS models at every documented vital sign measurement and derived time-normalized preintervention features, including slope, mean, area under the EWS curve (AUC), and exponentially weighted average (EWA). Informational value was quantified with nonparametric tests and L1-regularized logistic regression models targeting prehospital ROSC. Our findings demonstrate that short-term EWS dynamics encode measurable patterns of clinical deterioration, achieving moderate predictive performance (AUROC: 0.665) and advancing the current understanding of prehospital risk assessment. These results highlight the potential of incorporating vital sign trajectories into real-time, data-driven decision-support tools for EMS and motivate further exploration with more flexible, AI-based modeling approaches.

Deep neural networks for classification and segmentation of complex images require a sufficient amount of annotated data. Model performance depends significantly on the number of samples and the class balance within the dataset. In particular, anomalies and edge cases are often underrepresented, as they occur less frequently than control samples under normal conditions. However, collecting and annotating such data is usually associated with significant financial and time costs, as seen in applications like quality control in production. Moreover, in the medical sector, additional challenges such as data privacy can hinder the compilation of sufficient datasets. Recent studies show that generated synthetic samples, in combination with real samples, have the potential to expand underrepresented classes, increase variance and thereby improve model performance. However, generating realistic and complex multidimensional data is challenging and typically requires extensive domain knowledge.This work introduces a simple, automated concept to expanding complex image datasets with generated and fused hybrid training samples, providing an intermediate solution between purely real and entirely synthetic samples. Using a 3D convolutional variational autoencoder, new segmented variations of anomalies were generated and fused with real control samples. A comparison study was conducted, with the expanded hybrid dataset, along with the isolated real data, to examine the impact of generated hybrid samples. Our results demonstrate that the inclusion of synthetic anomalies enhances the training process and improves detection performance on 3D medical MRI data. The most promising models, trained on single sequences, achieved consistent or improved results across all evaluation metrics compared to models trained exclusively on real data. Since the proposed concept shows promising results, hybrid sample generation could become a valuable tool for data preparation and augmentation to optimize complex datasets and improve model training.

Large language models (LLMs) are now routinely used to autonomously execute complex tasks, from natural language processing to dynamic workflows like web searches. The usage of tool-calling and Retrieval Augmented Generation (RAG) allows LLMs to process and retrieve sensitive corporate data, amplifying both their functionality and vulnerability to abuse. As LLMs increasingly interact with external data sources, indirect prompt injection emerges as a critical and evolving attack vector, enabling adversaries to exploit models through manipulated inputs. Through a systematic evaluation of indirect prompt injection attacks across diverse models, we analyze how susceptible current LLMs are to such attacks, which parameters, including model size and manufacturer, specific implementations, shape their vulnerability, and which attack methods remain most effective. Our results reveal that even well-known attack patterns continue to succeed, exposing persistent weaknesses in model defenses. To address these vulnerabilities, we emphasize the need for strengthened training procedures to enhance inherent resilience, a centralized database of known attack vectors to enable proactive defense, and a unified testing framework to ensure continuous security validation. These steps are essential to push developers toward integrating security into the core design of LLMs, as our findings show that current models still fail to mitigate long-standing threats.

Companies increasingly integrate Artificial Intelligence (AI) into their applications to stay competitive. However, the efficient, successful, and certifiable development of AI applications requires complex setups, including computing resources, data stores, and training pipelines. Providing developer teams with all the necessary resources, tools, and services without reinventing the setup for every project remains a significant challenge. To address this, we propose a domain- and workflow-agnostic reference architecture for an on-premises AI Platform-as-a-Service (PaaS) that supports teams throughout the entire AI lifecycle and is reusable across multiple projects. To establish a shared understanding of the functionalities that such a platform should provide, we outline a set of general platform and MLOps capabilities. We validate the proposed reference architecture against these defined capabilities. This includes evaluating its suitability for on-premises deployment, its ability to support domain- and workflow-agnostic ML development, and its capacity to manage multiple concurrent projects. Additionally, to demonstrate its practical applicability, we present a proof-of-concept implementation composed of open-source components. This implementation can serve as a starting point for teams to build customized AI platforms, potentially reducing the initial setup effort.

Usually, mature Artificial Intelligence (AI) projects are developed by a team of various members, such as data engineers, data scientists, software engineers and machine learning (ML) engineers. They often pursue highly heterogeneous approaches, leading to new challenges in collaboration, particularly regarding software quality, data versioning and the traceability of model metrics and other resulting artifacts. These challenges are further intensified when AI projects rely on dynamic datasets, introducing an entirely new dimension that teams must deal with. Adopting principles from the machine learning operations (MLOps) paradigm becomes essential in this context. To go beyond existing process models and develop actionable guidelines, our work introduces a Git workflow for AI projects. We present basic instructions for data and code while outlining a minimal infrastructure setup. Building upon abstract concepts, we delve into concrete, actionable steps by examining the proposed branching workflow. Through a case study, we apply the development methodology to two use cases and demonstrate that the principles and approaches positively impact project outcomes.

Background and aims Peroral endoscopic myotomy (POEM) is a standard treatment option for achalasia patients. Treatment response varies due to factors such as achalasia type, degree of dilatation, pressure and distensibility indices. We present an innovative approach for treatment response prediction based on an automatic three-dimensional (3-D) reconstruction of the tubular oesophagus (TE) and the lower oesophageal sphincter (LES) in patients undergoing POEM for achalasia.Methods A software was developed, integrating data from high-resolution manometry, timed barium oesophagogram and endoscopic images to automatically generate 3-D reconstructions of the TE and LES. Novel normative indices for TE (volume×pressure) and LES (volume/pressure) were automatically integrated, facilitating pre-POEM and post-POEM comparisons. Treatment response was evaluated by changes in volumetric and pressure indices for the TE and the LES before as well as 3 and 12 months after POEM. In addition, these values were compared with normal value indices of non-achalasia patients.Results 50 treatment-naive achalasia patients were enrolled prospectively. The mean TE index decreased significantly (p<0.0001) and the mean LES index increased significantly 3 months post-POEM (p<0.0001). In the 12-month follow-up, no further significant change of value indices between 3 and 12 months post-POEM was seen. 3 months post-POEM mean LES index approached the mean LES of the healthy control group (p=0.077).Conclusion 3-D reconstruction provides an interactive, dynamic visualisation of the oesophagus, serving as a comprehensive tool for evaluating treatment response. It may contribute to refining our approach to achalasia treatment and optimising treatment outcomes.Trial registration number 22-0149.

Aims Achalasia is a chronic esophageal motility disorder. Peroral endoscopic myotomy (POEM) is a standard treatment option for patients with achalasia. However, treatment response varies due to factors such as achalasia type, degree of dilatation, pressure, and distensibility indices. This study presents an innovative approach for risk stratification, treatment response prediction and pathophysiological understanding based on an automatic three-dimensional (3-D) reconstruction of the tubular esophagus (TE) and the lower esophageal sphincter (LES) in patients undergoing POEM for achalasia.

Software testing is vital for strengthening the security of embedded systems by identifying and rectifying code errors, flaws and vulnerabilities. This is particularly significant when addressing vulnerabilities associated with side-channel attacks, given that they introduce a distinctive class of vulnerabilities, primarily subject to manual testing procedures. Manual testing remains prevalent despite advances in automation, posing challenges, particularly for complex environments. This research aims to automate embedded software testing on hardware in a modular and scalable manner, addressing the limitations of manual testing. We present a system designed to automate testing, including Side-Channel Analysis (SCA), in Continuous Integration (CI) environments, emphasizing accessibility and collaboration through open-source tools. Our evaluation setup based on GitLab, Jenkins and the ChipWhisperer framework shows that automating and integrating SCA in CI environments is possible in an efficient way.