BACKGROUND:Intraoperative real-time optical coherence tomography (iOCT) enables a dynamic visualization of retinal structures at the micrometer level during vitreoretinal surgery and provides additional information that can directly influence the course of surgery. The aim of this study is the presentation of the clinical utility, limitations and current evidence base for the use of iOCT in vitreoretinal surgery. MATERIAL AND METHODS:A narrative review was conducted based on published studies, case series, subanalyses of prospective cohorts and systematic reviews addressing the use of iOCT in vitreoretinal procedures. The literature search was primarily performed using PubMed/MEDLINE. RESULTS:During epiretinal membrane (ERM) and internal limiting membrane (ILM) peeling, iOCT can assist in identifying an appropriate initiation site for membrane peeling. It also enables real-time assessment of the tissue response and facilitates verification of complete membrane removal by detecting residual fragments. In macular hole surgery, iOCT enables intraoperative visualization of hole morphology and reliable assessment of ILM flap positioning, including after fluid-air exchange. In retinal detachment and proliferative vitreoretinopathy (PVR) surgery, iOCT not only facilitates the detection of residual subretinal fluid but also supports the evaluation of dissection planes and tractional membranes. DISCUSSION:Although many studies report a substantial added value of intraoperative OCT, a significant functional benefit, such as improvements in visual acuity outcomes, recurrence rates or complication rates has not yet been conclusively demonstrated. Nevertheless, the ability to dynamically assess retinal structures at micrometer resolution makes it possible to guide intraoperative decision-making, minimize iatrogenic tissue damage and enhance visualization, thereby providing a clear practical advantage, especially in complex cases or when microscopic visualization is limited. To date, limited evidence, high costs and heterogeneous technical implementations with variable clinical applicability have hindered widespread adoption of iOCT; however, in light of considerable technological advancements in recent years, increasing acceptance of this intraoperative imaging modality can be observed.
Background The increasing complexity, interaction, and user acceptance of generative artificial intelligence models (AI) can lead to unexpected, dangerous actions or behaviours that run counter to the mo dels' intended purpose. The aim of this narrative review is to identify examples of such rogue AI, outline the implications they might have for the field of anaes thesiology, and to find approaches to solutions. Methods For a narrative review, a PubMed and a Google Scholar search were conducted with the strings "Artificial intelligence / Machine learning AND/OR Rogue AI" as well as a Google search for exemplary cases of rogue AI. Scientific articles, journalistic reports as well as grey litera ture were included. Results A total of 12 exemplary scientific artic les, one case and three security reports, one professional association communi cation and 9 journalistic reports were identified. These included manipulative or extortionate behaviour of AI models, misdiagnoses caused by hallucinations or insufficiently trained or validated mo dels, examples of racist bias due to in adequate datasets, refusals to execute input commands, and unsolvable en cryptions. In the field of cybersecurity, studies reported on hidden backdoors, hacking, and manipulation of source code or training data. The control of eco nomic processes by AI could also lead to potential financial losses. No rogue AI was found to be directly implemented in the medical field. In anaesthesiology, for example, this could lead to problems affecting doctor patient interactions, the malfunction or takeover of medical de vices by AI, over or undertreatment due to bias issues, misdiagnoses, and a disrupted doctor AI or patient AI inter action. Conclusion The fundamental aspects of the rogue AI problem already exist today. In the fu ture, the problem could worsen with self learning and self optimising AI systems that are interconnected at all levels within hospitals. Approaches needed to solve these problems consist in comply ing with biomedical ethical guidelines, principles of fairness, transparency, le gislative frameworks like the EU AI Act, and extended cybersecurity against ex ternal attacks and uncontrolled internal usage. Next to an effective user training, continuous human oversight and cor rection mechanisms, as well as realtime monitoring during operation, should also be consistently ensured.
