A Novel Graph Neural Network to Localize Eloquent Cortex in Brain Tumor Patients from Resting-State fMRI Connectivity
- 699 Downloads
We develop a novel method to localize the language and motor areas of the eloquent cortex in brain tumor patients based on resting-state fMRI (rs-fMRI) connectivity. Our method leverages the representation power of convolutional neural networks through specialized filters that act topologically on the rs-fMRI connectivity data. This Graph Neural Network (GNN) classifies each parcel in the brain into eloquent cortex, tumor, or background gray matter, thus accommodating varying tumor characteristics across patients. Our loss function also reflects the large class-imbalance present in our data. We evaluate our GNN on rs-fMRI data from 60 brain tumor patients with different tumor sizes and locations. We use motor and language task fMRI for validation. Our model achieves better localization than linear SVM, random forest, and a multilayer perceptron architecture. Our GNN is able to correctly identify bilateral language areas in the brain even when trained on patients whose language network is lateralized to the left hemisphere.
KeywordsRs-fMRI Graph Neural Network Language localization
This work was supported by the National Science Foundation CAREER award 1845430 (PI: Venkataraman) and the Research & Education Foundation Carestream Health RSNA Research Scholar Grant RSCH1420.
- 3.Langs, G., et al.: Functional geometry alignment and localization of brain areas. In: Advances in Neural Information Processing Systems, pp. 1225–1233 (2010)Google Scholar
- 4.Nandakumar, N., et al.: Defining patient specific functional parcellations in lesional cohorts via Markov random fields. In: Wu, G., Rekik, I., Schirmer, M.D., Chung, A.W., Munsell, B. (eds.) CNI 2018. LNCS, vol. 11083, pp. 88–98. Springer, Cham (2018). https://doi.org/10.1007/978-3-030-00755-3_10CrossRefGoogle Scholar
- 5.Gohel, S., et al.: Resting-state functional connectivity of the middle frontal gyrus can predict language lateralization in patients with brain tumors. Am. J. Neuroradiol. (2019)Google Scholar
- 8.Kamnitsas, K., et al.: Efficient multi-scale 3D CNN with fully connected CRF for accurate brain lesion segmentation. MedIA 36, 61–78 (2017)Google Scholar
- 9.Khosla, M., Jamison, K., Kuceyeski, A., Sabuncu, M.R.: 3D convolutional neural networks for classification of functional connectomes. In: Stoyanov, D., et al. (eds.) DLMIA/ML-CDS -2018. LNCS, vol. 11045, pp. 137–145. Springer, Cham (2018). https://doi.org/10.1007/978-3-030-00889-5_16CrossRefGoogle Scholar