background (recipient) brain scan
4 Lesion function mapping using mass univariate techniques
4.1
Introduction
The beginnings of functional localisation within the brain can be traced back to work by Paul Broca and Carl Wernicke in the mid 19th Century. Through careful
clinical observations of patients and subsequent post mortem examination of the patients’ brains, Broca correlated the ventro-posterior region of the frontal lobe to speech production, whilst Wernicke related the left posterior, superior temporal gyrus to language comprehension (Broca, 1861; Wernicke, 1874). With these discoveries, they revealed the human brain to possess a functionally specialised architecture. Their conclusions were based on the visual assessment on a number of brains, noting the size and location of injury and documenting regions of commonality. Their approach was, however, hindered by the small available sample size of suitable patients and the moral requirement to await their timely passing. The arrival of computed tomography (CT) and Magnetic Resonance Imaging (MRI) opened a new dimension to lesion function mapping (Damasio and Damasio, 1989; Rorden and Karnath, 2004). With these tools, brain injury could be visualized, quantified and
monitored in vivo. This has not only increased the number of patients available to study, but also increased the spatial resolution of analysis, improving
temporal resolution and providing a means to collect suitable control subjects. As a result over the last century and a half such studies have been critical to identifying the distinctive neural substrates of language (Bates et al., 2003; Dronkers et al., 2004), memory (Scoville and Milner, 1957), emotion (Adolphs et al., 1995; Calder et al., 2000), attention (Egly et al., 1994; Karnath et al., 2004; Mort et al., 2003) and intelligence (Gläscher et al., 2009).
Functional Magnetic Resonance Imaging (fMRI) has become a popular tool for investigating the functional architecture of the brain. The technique exploits the different magnetic properties of oxygenated and deoxygenated blood to generate a blood oxygen level dependent (BOLD) signal. The crux of fMRI studies is the association between increased neuronal activity and oxygen requirements. In this technique's favour is the spatial resolution of 1-5mm (Menon and Kim, 1999), and a temporal resolution of seconds. Importantly fMRI permits the observation of the brain but cannot interfere with brain function, therefore necessitating the manipulation of the experimental design to generate contrasts in the BOLD signal. As a consequence, fMRI provides a powerful approach to hypothesising putative critical regions but has greater difficulty in testing them (Aue et al., 2009). The necessity of a brain region for a putative function – arguably the strongest test – can only be established by showing a deficit when the function of the region is disrupted.
Inactivating brain areas experimentally cannot easily be done in the human. The technique of transcranial magnetic stimulation (TMS) provides a non- invasive approach to temporarily disrupting a region of the brain. Although temporally its resolution is high, in the order of milliseconds, its spatial
resolution is very limited, particularly centripetally, as the effects of stimulation are restricted to superficial cortical regions (Epstein et al., 1990; Rudiak and Marg, 1994; Walsh and Cowey, 2000; Zangen et al., 2005).
The only comprehensive means of establishing necessity is therefore the study of patients with naturally occurring focal brain lesions (Rorden and Karnath, 2004). The majority of these studies have involved a cohort of patients with statistics performed on the group rather than the individual. As a consequence the brain volumes must all be brought into spatial register, by transforming each image so as to align homologous regions between images, enabling point-by-point anatomical comparisons to be made across the cohort.
When the brain is damaged by a focal pathological process the pattern of damage generally bears no relation to the underlying functional architecture. Critically, the scale of damage in clinical cases (typically 10-2m) is substantially
greater than the scale of functional organisation suggested by the spatial heterogeneity of individual neuronal responses (<10-3m). To make population-
level inferences about the functional role of a given part of the brain from lesion data, previous studies have therefore relied on comparing sets of patients with large, inevitably overlapping, lesions to identify a critical locus much smaller than each individual lesion (e.g. (Karnath et al., 2004)). The comparison is made by applying a statistical test point-by-point for each part of the brain, discretized at some convenient spatial resolution, where each point is treated independently (Bates et al., 2003; Karnath et al., 2004). This mass-univariate approach assumes that the other, apparently non-critical areas damaged in each patient do not distort the localisation of the one, critical area that is common to them all: in short, that any spatial correlations in the pattern of damage within each patient are well-behaved. However, it is these hidden systematic biases in the natural patterns of damage that may invalidate our anatomical inferences. In relation to overwhelmingly the commonest lesion type used in such studies – ischaemic vascular – the assumption is fundamentally unsafe because the architecture of the vascular tree is highly stereotyped across individuals.
To illustrate the potential consequences of this mass univariate approach in the presence of hidden structure in the data, consider the two-dimensional synthetic example in figure 1.6, where damage to any part of area A alone may disrupt a putative function of interest but B plays no role in this function of interest. If the lesions used to map the functional dependence on A follow a stereotyped pattern where damage to any part of A is systematically associated with collateral damage to the non-critical area B, both areas may appear to be significantly associated even if B is irrelevant to the function of