Presurgical Tumor Tissue Detection Based On Resting-state FMRI

Gliomas are the most common primary intracranial tumors, accounting for approximately 70-80% of all brain tumors. Currently, the main treatment of glioma is neurosurgery. Accurate delineation of glioma tissue relative to surrounding functional eloquent areas helps maximize tumor resection and improves further outcome. Blood-oxygen-level-dependent functional magnetic resonance imaging (BOLD-fMRI) has been routinely adopted for presurgical functional mapping. Here we show the feasibility of using presurgical fMRI for tumor tissue detection for completely utilization of this type of imaging data. We introduced a novel tumor tissue detection method based on resting-state fMRI and independent component analysis (ICA) with a novel template-matching-based automatic tumor component identification algorithm named "discriminability index-based component identification" (DICI). Resting-state fMRI data of 32 glioma patients from three centers, plus the data of 28 patients from a fourth center with non-brain, musculoskeletal tumors were fed into our algorithm. Individual ICA with different total number of components (TNCs) were conducted to decompose rs-fMRI data into multiple components. The best-fitted tumor-related components with the optimized TNCs setting were automatically determined based on the DICI algorithm. The success rates in glioma tissue detection are 100%,100% and 93.75% for the data from the first three centers, respectively, and 85.19% for the data with musculoskeletal tumors. Tumors with various size, pathologies and location did not hinder the detection based on the current available samples. The DICI algorithm was also proven to be robust with the threshold applied to the components. Given the large inter-individual variability of human brain, this is also the first time that subject-specific TNCs is suggested (instead of using fixed TNCs across all subjects) to identify the optimized component of interest (tumor-related component in the current study). We also compared the performances of our implementation for tumor-related component identification with that of seed-based correlation method, demonstrating the advantage of ICA with DICI over seed-based approach. Finally, we proposed that the high success rate was achieved probably due to the ability of BOLD-rs-fMRI in characterizing abnormal vascularization, vasomotion and perfusion caused by tumors. The interesting findings suggest that BOLD-rs-fMRI is a promising non-invasive technique for comprehensive presurgical assessment.