Throughout the past decade, rs-fMRI has been validated in multiple studies as a technique for presurgical brain mapping, where results were highly concordant with functional maps extracted through tb-fMRI, intra-operative electro-cortical stimulation (ECS) and even direct visual identification of the motor cortex based on anatomical landmarks [20,21,22].
Multiple studies compared the reliability of rs-fmri analysis to tb-fMRI, ECS or both. Zhang el al. [23] conducted a study on 17 healthy controls and 4 patients, using SBA rs-fMRI. They were able to localize sensorimotor cortex compared to ECS and performed as well or better than tb-fMRI. In a study on 17 supratentorial glioma patients, SBA rs-fMRI was performed using a seed placed on the hand knob area of the healthy hemisphere. Compared to ECS, the overall sensitivity and specificity values of rs-fMRI were 90.91 and 89.41%, respectively. In the 15 patients who were able to perform tb-fMRI, the sensitivity and specificity values for the hand motor area were 78.57 and 84.76%, respectively. For these 15 patients who underwent both rs-fMRI and tb-fMRI, there was no statistical difference in sensitivity or specificity between these two methods [24]. Another study conducted on 10 healthy subjects and 25 patients with left sided brain tumors compared tb-fMRI, SBA rs-fMRI and visual anatomical localization of the hand motor found important differences in the locations (i.e., ˃ 20 mm) of the determined hand motor voxels by these three MR imaging methods [20]. In the current study, the sensorimotor cortex was detected using SBA and ICA maps in all patients, the pattern of activation was correlated to the anatomical structure regarding the expected activation area and the two networks showed similar activation areas in the lesional hemisphere.
Several studies used ICA rs-fMRI to extract the SMN, Kokkonen et al. [25] compared tb-fMRI and rs-fMRI using ICA in a group of patients with brain tumors of heterogeneous histopathology as well as a group of healthy controls, they found successfully localized the SMN in both groups, i.e. the brain tumors did not hamper visualization of the SMN in the cases. In the resting state group ICA, both the cases and the controls presented sensorimotor areas similar to the motor task group ICAs and to each other. Schnieider et al. [26] compared ICA rs-fMRI to tb-fMRI in a group of 19 brain tumor patients. They successfully and consistently identified motor cortex by using rs-fMRI and visual assessment by 3 radiologists for all patients. Even when subdividing the motor cortex into 3 segments (hand, face and foot), the sensitivities of rs-fMRI and tb-fMRI were comparable. Another study compared both SBA and ICA RSNs to ECS and found significant correlation between resting-state low frequency fluctuations of BOLD signal and the functional unity of distinct cortical areas confirmed by ECS. Additionally, the ICA method succeeded in detection of the sensori-motor and language networks in 90 (92%) and 73 (75%) cases respectively [27]. In the current study, using ICA, the sensorimotor network was identified in 100% of cases and the 4 components of language network were identified in 41% of cases. When comparing the SMN FC maps extracted using SBA and ICA, SBA maps were larger than ICA maps at a p value of ≤ 0.01.
Tie et al. [28] tried detecting ICAs related to language networks in preoperative patients, with 5 experts rating the language networks in ICAs generated from rs-fMRI data of 17 patients. They found moderate agreement between experts, moreover they found that 5 experts agreed upon 1 component most representative of the language network in 10 of the patients, which jumped to 14 upon consensus of only 4 experts. In the present study, using ICA, at least two of the components the language network was detected in 17 out of 22 patients whilst the 4 components of the language network were identified in 9 out of 22 patients.
The two main methods for ICA RSN recognition are subjective identification of these networks by experts [23, 25] and using automated software for detection [29], however the latter may be limited by structural changes caused by tumoral infiltration and mass effect of the tumor on target regions of interest [29]. Additionally, some of the RSNs are more readily reproducible than others, for example, in our study we were able to detect the SMN and DMN more frequently than other RSNs e.g. lateral visual, and DAN.
Thresholding FC maps to segregate actual activity from noise is a debatable subject, especially due to inter-individual differences in activation intensity [30]. A common practice in clinical pre-surgical fMRI is manual thresholding, which is operator dependent [30,31,32]. Some authors suggested methods to standardize the thresholding of the extracted FC maps including setting a percentage of the maximum intensity detected below which activations are discarded [30]. Even with manual thresholding, radiologists should be aware of the differences in intensity values of FC maps generated by their analysis pipeline. Even with manual thresholding, radiologists should be aware of the difference in the extent of FC maps generated by their analysis pipeline. In our study, the SBA pipeline generated larger FC maps, consequently, more thresholding parameters would be required on these maps for proper delineation of RNSs required for clinical pre-surgical mapping.
A recent meta-analysis of published preoperative tb-fMRI studies showed that preoperative fMRI planning -in addition to other advanced imaging techniques- reduced post-operative morbidity [33]. It is safe to project the same conclusions to rs-fMRI, since its maps are closely correlated to those of tb-fMRI, as previously mentioned. Nevertheless, the correlation between preoperative rs-fMRI and post-operative outcome -especially in pediatric patients- is a difficult subject due to several factors. First, there is no consensus on a clinical functional assessment method for pediatric brain tumors; in fact, many clinical assessment techniques for brain neoplastic lesions rely more on the day-to-day well being rather than gross and fine motor, sensory and language skills e.g. Karnofsky Performance Score, European Organisation for Research and Treatment of Cancer Quality of Life Core Questionnaire..etc. [34]. Other factors include different localization of the neoplastic lesions in relation to the canonical sensori-motor and language eloquent brain areas and their effect on these areas (e,g, displacement Vs infiltration) and surgical approach to the lesions (e.g. debulking Vs excision). The correlation of the tumoral effect on other RSNs e.g. DMN and post-operative cognitive outcome is similarly challenging as in-addition to the previous limitations-the role these networks play in cognitive functions is not yet fully understood [35]. One of the promising methods for better understanding these global and local cerebral RSNs is graph theory analysis, which may also play an important role in the future in the pre-operative planning and preservation brain network connectomics through critical node planning [36].
This study was not without limitations, the heterogeneous pathology and location of the tumors –even though reflects actual daily practice- falls short of accounting for their effect on the detection and extent of the extracted RSNs. The definition of the language network did not include some of the ROIs that may play a role in language processing e.g. premotor cortex. Finally, detection of language lateralization using rs-fMRI is still a challenging research subject with promising published data [37, 38].