The roadblocks to AI adoption in surgery: Data, real‐time applications and ethics

Abstract

This paper explores the critical barriers to AI integration in surgery, focusing on challenges in data management, real-time application and ethics. It emphasizes the need for robust data-sharing frameworks, advancements in real-time processing capabilities and ethical guidelines to ensure responsible AI deployment. Artificial intelligence (AI) has emerged as a transformative force in medicine, offering unprecedented opportunities to revolutionize various aspects of healthcare. In particular, AI has the potential to significantly enhance surgeons' capabilities by leveraging advanced data analytics and sophisticated decision-making tools [1]. The true promise of AI in surgical practice lies in its remarkable ability to process vast and complex datasets in real-time (RT), thereby providing surgeons with actionable insights that have the potential to improve patient outcomes markedly. Nevertheless, the journey toward the full integration of AI into surgical practice is laden with numerous challenges that must be meticulously and systematically addressed. This commentary delves into these challenges, emphasizing the critical areas of data acquisition, the feasibility of RT applications and the ethical considerations that arise. Each aspect is pivotal to AI's successful and responsible implementation in surgical environments, where the stakes are high, and the potential for impact is profound. The foundation of any effective AI system is the availability of high-quality, annotated data. In the surgical domain, this requirement presents significant challenges. Surgical procedures generate vast amounts of data, particularly video recordings and imaging, essential for training AI models, where 1 min of surgical video is estimated to contain 25 times the amount of data found in computed tomography imaging [2]. However, the variability in surgical techniques, patient anatomy and the inconsistent use of data collection methods across different institutions create substantial hurdles in building comprehensive datasets that are both diverse and standardized. A further challenge in this context is that data annotation in surgery is time-consuming and highly specialized, requiring expert surgeons to label critical anatomical structures, surgical steps and potential outcomes. This labor-intensive process significantly limits the scale of datasets that can be used for training AI models. Moreover, there is a general reluctance among surgeons to record and share surgical data, primarily due to concerns over patient privacy and the potential legal implications of such recordings. This hesitancy limits the amount of data available for AI model training, resulting in models that may lack the robustness and generalizability needed for widespread clinical use. To address these data challenges, it is critical to foster collaboration between institutions to create standardized data-sharing frameworks. Standardized data collection and annotation methods across healthcare systems would significantly improve AI training. Furthermore, incentives should be created to encourage sharing of surgical data while ensuring stringent patient privacy protections through ethical and legal safeguards. Blockchain technology could also be explored to share data across institutions while maintaining privacy securely. Furthermore, the variability in the quality of the collected data further complicates the training process, as AI models require consistent and high-quality inputs to produce reliable outputs. To address these challenges, there is a need to develop standardized protocols for data collection and annotation, ensuring that data from different sources can be seamlessly integrated into AI systems. For instance, Meireles et al. [3] have proposed an established hierarchy for annotating temporal events in surgery, comprising temporal models, actions and tasks, tissue characteristics and general anatomy and software and data structure. Incentives should be provided