Voodoo surgery? The distinct challenges of functional neuroimaging in clinical neurology

Sir, I read with interest the article by Lyon (2017), which provided an excellent cautionary note regarding the analysis and interpretation of functional MRI data. I concur with her overall conclusion that, in spite of recent scepticism, functional MRI remains a very useful technique, provided one is vigilant in maintaining best practice in terms of ‘good (experimental) design, analysis and reporting’. However, many of the issues that Lyon (2017) raises relate to functional MRI data from groups of (often healthy) participants to answer (often theoretical) questions on cognitive function. It is important to recognize that functional MRI is commonly applied in the clinical setting as well, which gives rise to a unique and fundamentally different set of challenges. Such applications are associated with more immediate and life-altering sequelae, and it behoves both clinicians and scientists to be aware of the distinct methodological limitations of such clinically-based applications. Consider, for example, patients with pathological brain lesions (e.g. a tumour or epileptogenic focus), who commonly undergo functional MRI to localize areas of ‘eloquent cortex’ that should be spared during lesion resection (Matthews et al., 2006; Chong and Cook, 2013). Even at the most fundamental level, the neurophysiological cornerstone of functional MRI—the blood oxygen level-dependent (BOLD) response—may not necessarily provide an accurate measure of neural activity in such patients. Brain tumours, which are often highly vascularized, can affect the localization and intensity of the BOLD signal through changes in vascular autoregulation, as well as through direct structural compression, resulting in a disruption of the tight neurovascular coupling that is seen in healthy physiology (Logothetis, 2002; D’Esposito et al., 2003; Lauritzen, 2005). In addition, concurrent disease processes (e.g. cerebrovascular disease), as well as certain commonly prescribed medications (e.g. dopaminergic drugs, certain anti-depressants) may alter either vascular flow or neurovascular coupling (D’Esposito et al., 2003). The consequence may be a BOLD signal that is attenuated, absent, or even inverted, and/or falsely localized (Peeters and Sunaert, 2007). With regards to task design, a challenge in preoperative functional MRI is determining, not necessarily what is ‘good’, but rather what is appropriate. The cognitive neuroscience literature is replete with well designed paradigms that probe many different aspects of cognition. However, in clinical practice, cognition has traditionally been modularized into monolithic constructs (e.g. ‘attention’, ‘memory’, ‘language’, ‘executive function’). Although this is clinically convenient, it becomes problematic when interpreting the nature of activations in an individual patient. For example, in the case of language, the distribution of activity will vary based on the nature of the task (e.g. word generation versus sentence completion; noun versus verb generation; comprehension versus word generation), as well as the baseline against which the conditions of interest are compared (active versus passive) (e.g. Zacà et al., 2013). Precisely which paradigms predict an improved postoperative outcome are yet to be doi:10.1093/brain/awx283 BRAIN 2017: 140; 1–3 | e76