The age of Artificial Intelligence in neurosurgical practice and educational paradigms – Considerations from the EANS Ethico-Legal and Young Neurosurgeons’ committees

To the Editor,

In recent years, there has been a substantial increase in publications on artificial intelligence (AI) in neurosurgery. This reflects not only the technical progress but also the scientific interest and potential benefits in a powerful emerging technology. While there are well defined and accepted classifications of AI, either based on their functionalities and likeness to human intelligence (Reactive Machines, Limited Memory, Theory of Mind, and Self Aware AI), or based on their technological capabilities and potential to surpass human intelligence (Narrow Intelligence, General Intelligence, Super Intelligence), commonly speaking the term of AI most often refers to the subfields machine learning (ML) and deep neural networks (Benke and Benke, 2018; Lopez-Jimenez et al., 2020). The general areas of application in medicine include diagnostics, prediction of outcome, and complications, as well as the creation of treatment models (Staartjes et al., 2020a). When applied correctly, AI enables the step towards more individualized medicine by optimized information on diagnosis, treatment options, potential risks and the associated prognosis (Bonsanto and Tronnier, 2020; Schwalbe and Wahl, 2020; Azad et al., 2021).

As younger generation of the large a committee of the EANS family, the Young Neurosurgeons Committee led an effort to appraise the scientific direction of AI in Neurosurgery, including its practical as well as educational applications, and with the support of the Ethico-Legal Committee, we would like to we tried to provide a reflection on the bioethical implication of this technological revolution.

1. Practice and education

In the digitalized world of neurosurgery, documentation and administration is an important but also growing part of daily medical work, which is still mainly completed by physicians (Stienen et al., 2016). This takes up valuable time and resources, which could be otherwise invested in patient care, research, education and training. In this context, AI-based tools offer the potential to simplify or even completely take over a variety of administrative tasks and manage data more effectively (Shanafelt et al., 2016; Greisman and DiGiorgio, 2023). There are already approaches using natural language processing (NLP) by artificial neural networks to automatically generate discharge reports for patients based on surgical reports, patient information, and patient records (Liu et al., 2023). Other large-scale language models (LLM) currently find wide application in the healthcare sector (Ali et al., 2023; Peng et al., 2023; Stroop et al., 2023).

In addition to data management, AI-based applications could also play a key role in data analysis when it comes to identifying correlations and complex patterns in large data sets (Staartjes and Stienen, 2019). Current developments might have a significant impact on patient selection, diagnosis, treatment plans and outcome prediction (Schwalbe and Wahl, 2020; El-Hajj et al., 2023). For example, AI algorithms already enables the providing analysis of patient profiles and medication lists have been fostering pharmacy automation and enabling robotic drug delivery systems. These software systems are linked to a warning system, which indicates potential drug interactions. AI might be used to improve clinical practice and overcome barriers concerning diagnosis and clinical decision-making. There are already applications that use machine learning models to better predict the development of measurements (e.g. intracranial pressure) or the prognosis of surgical outcomes. This makes it possible to improve the interpretation of data patterns in everyday clinical practice and to provide individualized recommendations for action (Schweingruber et al., 2022; Staartjes et al., 2020b, 2022; de Jong et al., 2021). Even if the clinical prediction models for personalized medicine are promising, clinical practicability is not yet fully established in many applications.

The combination of AI for image segmentation with robotics or other surgical applications such as neuronavigation pursues the realistic goal of increasing accuracy while reducing the error rate (Panesar et al., 2020). Surgical Phase Recognition (SPR) is, for instance, an AI-based technology which contains the potential to use data from surgical videos and to identify surgical steps in order to provide intra-operative decision support in the second instance (Jumah et al., 2022). In addition to advancing surgical practice, such technologies might present valuable tools for surgical training. During simulated operations, AI software might offer suggestions or warnings prior to critical steps to assist and correct the trainee. However, if the underlying algorithms of existing models are not properly understood, the implementation might lead to systematic errors with potentially disastrous consequences. In addition to loss of surgical competence, over-reliance on technology and a lack of basic understanding are potential risk factors that could lead to a certain degree of ignorance and neglect for the risks and complications of a fully or semi-automated workflow (Panesar et al., 2020).

AI applications will also play an increasingly important role in neurosurgical education and surgical training. Occupational health, safety legislation and increasing bureaucratic requirements reduced significantly the exposure time of residents spent in the operating theatre, which is an essential part of their training (Stengel et al., 2022a; Stienen et al., 2020). In addition to conventional training methods such as cadaver training, modern technologies like mixed reality are already available today and have proven to be a relevant part of neurosurgical training (Stengel et al., 2022b). In addition to practical training support, AI enables new training concepts in the context of adaptive e-learning with the aim of improving weaknesses in a more targeted and effective way (Hickmann et al., 2022). We believe this type of training will become increasingly important for residents in the future.

2. Bioethical considerations

In addition to all benefits expected from AI-based technology, such as personalized medicine, reduction of risks and complications due to increased precision as well as cost reduction, the general and bioethical risks should not be underestimated (Abdullah et al., 2021). Implementing AI systems requires significant initial investment and ongoing maintenance costs, which need to be weighed against potential long-term savings and improved efficiencies. In addition, data protection and privacy concerns, particularly under the General Data Protection Regulation in Europe, require careful consideration of data handling, storage and processing protocols.AI can easily surpass human data management and memory capacity. The challenge seems to be to maintain personal integrity and responsibility for moral behavior through AI applications (Mathiesen and Broekman, 2022). Safety and ethical concerns need to be addressed before implementing AI in the clinical workflow. Failure to consider the risks of bias and health inequalities that may arise from AI applications can have a drastic impact on patient care if decisions in everyday clinical practice are based on these models (Uche-Anya et al., 2022; Goisauf and Cano Abadía, 2022). This is particularly relevant when AI systems are trained on datasets that may not be representative of diverse patient populations or healthcare settings. Ultimately, applications are only as good as the models they are conceptualized on and the data which is used to train the models. To ensure responsible implementation, we advocate the establishment of European guidelines that include standardised protocols for data collection and sharing, regular evaluation of the performance of AI systems in different patient populations, and clear governance structures for the implementation of AI in clinical practice.

3. Conclusion

Communities such as the EthicoLegal Committee and the Young Neurosurgeons Network (YNN) of the EANS provide a suitable platform for establishing high quality studies with clinical relevance at European level, as research in this field requires multidisciplinary international collaboration between computer scientists, biostatisticians and clinical researchers. Further projects could demonstrate the full capacity of this modern technological development.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Handling Editor: Dr W Peul