Résumé
Neurological and neurodevelopmental conditions are a major public health concern for which new therapies are urgently needed. The development of effective therapies relies on the precise mapping of the neural substrates causally involved in behaviour generation. Direct electrical stimulation (DES) performed during cognitive and neurological monitoring in awake surgery is currently considered the gold standard for the causal mapping of brain functions. However, DES is limited by the focal nature of the stimulation sites, hampering a real holistic exploration of human brain functions at the network level. We used 4'137 DES points derived from 612 glioma patients in combination with human connectome data - resting-state functional MRI, N=1'000 and diffusion weighted imaging, N=284, respectively - to provide a multimodal description of the causal macroscale functional networks subtending twelve distinct behavioural domains. To probe the validity of our procedure, we compared the network topographies of healthy and clinical populations; we tested the predictive capacity of direct electrical stimulation derived networks; we quantified the coupling between structural and functional connectivity, and we built a multivariate model able to quantify single subject deviations from a normative population. Lastly, we probed the translational potential of DES derived functional networks by testing their specificity and sensitivity in identifying critical neuromodulation targets and neural substrates associated with post-operative language deficits. The combination of DES and human connectome data resulted in an average 29.4-fold increase in whole brain coverage compared to direct electrical stimulation alone. DES derived functional networks are predictive of future stimulation points (97.8% accuracy), and strongly supported by the anatomical connectivity of subcortical stimulations. We did not observe any significant topographical differences between the patients and the healthy population at both group and single subject level. Showcasing concrete clinical applications, we found that DES derived functional networks overlap with effective neuromodulation targets across several functional domains, show a high degree of specificity when tested with the intracranial stimulation points of a different stimulation technique, and can be effectively used to characterise post-operative behavioural deficits. The integration of DES with the human connectome fundamentally advances the quality of the functional mapping provided by DES or functional imaging alone. DES derived functional networks can reliably predict future stimulation points, have a strong correspondence with the underlying white matter, and can be used for patient specific functional mapping. Possible applications range from psychiatry and neurology up to neuropsychology, neurosurgery, and neurorehabilitation.