Table 1.

Barriers and solutions for integration of AI into brain tumour surgery.

Barrier	Proposed solution
Requirement of large datasets to train existing ML programs	Creation of international databases as repositories for training data for brain tumours. Collaboration between neurosurgical oncology units. Synthetic multi-parametric MRI image generation.
Selection bias of training data	Ensure a wide range of demographics used to train ML programs. Use of international databases as repositories for training data.
Patient confidentiality concerns when sharing patient data between units to train ML platforms	Robust scrutiny of data governance for existing databases. Development of technologies in accordance with existing ethical and legal frameworks. Synthetic multi-parametric MRI image generation.
Slow progress in advancing ML programming	International collaboration between ML programming teams. Publishing code for all newly developed ML platforms, making code widely available for further development and scrutiny.
“Black box” conundrum	Ensure that human users can understand and trace all predictions and decisions made by future ML platforms.
Poor contextualisation of uncertainty by ML programs	Ensure that ML platforms developed for use in brain tumour management are used in tandem with clinicians, who are better able to contextualise and explain uncertainty.