Improvement of Functional Brain Parcellation Using Machine Learning Techniques

Abstract

Functional brain parcellation is fundamental to understanding large-scale neural organization and improving downstream neuroimaging analyses. Traditional atlas-based and classical clustering approaches, while widely used, often fail to capture complex, non-linear functional connectivity patterns and inter-individual variability present in high-dimensional resting-state fMRI data. This study proposes a machine learning–driven framework to enhance functional brain parcellation using large-scale datasets, including those from the Human Connectome Project. The methodology integrates advanced preprocessing, feature extraction from functional connectivity matrices, and multiple machine learning models, including unsupervised clustering, deep learning architectures, and graph-based neural networks. Model performance was evaluated using functional homogeneity metrics (Silhouette Score, Davies–Bouldin Index), spatial consistency measures (Dice Similarity Coefficient), and predictive utility in cognitive outcome modeling. Results demonstrate that the proposed framework outperformed traditional atlas-based, spectral clustering, and graph modularity approaches. The machine learning model achieved a higher mean silhouette score (0.61), a lower Davies–Bouldin Index (0.94), and improved Dice similarity (0.76), indicating enhanced intra-parcel coherence and inter-subject reproducibility. Furthermore, predictive modeling accuracy improved by 9–14%, with an 11% reduction in mean absolute error compared to atlas-based features. Statistical testing confirmed that these improvements were significant (p < 0.05). The findings suggest that machine learning techniques substantially improve the accuracy, stability, and practical utility of functional brain parcellation. By capturing non-linear connectivity structures and adapting to individual variability, the proposed framework advances computational brain mapping and supports more reliable neuroimaging analyses for research and potential clinical applications.

Introduction

Understanding the structural and functional organization of the human brain is a major challenge in neuroscience. Functional brain parcellation, which divides the brain into distinct regions with similar functional activity, is essential for mapping neural networks, studying brain activity, and identifying biomarkers for neurological and psychiatric disorders. Traditional approaches, such as atlas-based methods and connectivity-based clustering, have contributed significantly to brain mapping but often rely on predefined anatomical boundaries or linear assumptions that may not capture the complex patterns present in neuroimaging data.

With the development of advanced neuroimaging techniques like functional magnetic resonance imaging (fMRI) and large datasets from projects such as the Human Connectome Project, researchers can study brain connectivity with higher spatial and temporal resolution. However, these datasets are highly complex and contain noise, individual variability, and non-linear interactions between brain regions, which limit the effectiveness of traditional analysis methods. As a result, machine learning techniques have emerged as powerful tools for improving functional brain parcellation.

Machine learning methods—including clustering algorithms, graph-based models, convolutional neural networks (CNNs), and graph neural networks (GNNs)—can capture complex and non-linear relationships in neuroimaging data. These approaches enable more accurate and individualized brain parcellation, reflecting subject-specific connectivity patterns and improving applications such as disease classification, cognitive mapping, and predictive modeling. Despite these advantages, challenges remain related to interpretability, generalizability, computational cost, and validation across datasets.

The literature shows an evolution from anatomical atlases like the Brodmann map to connectivity-based and machine learning-driven methods. Atlas-based approaches provide standardization but fail to capture individual differences. Connectivity-driven clustering and graph-theoretical methods improved functional segmentation but are sensitive to noise and require predefined parameters. Recent research focuses on deep learning and individualized parcellation, which create personalized brain maps based on each subject’s connectivity profile. Emerging directions include multimodal data integration, explainable AI (XAI), and hybrid models combining graph theory with deep learning.

The study proposes a computational neuroimaging framework using machine learning to enhance functional brain parcellation. The methodology includes three main stages: data acquisition and preprocessing, feature extraction and model development, and clustering with validation. Neuroimaging data from the Human Connectome Project were used, with preprocessing steps such as motion correction, spatial normalization, temporal filtering, and noise removal to ensure data quality.

Functional connectivity features were extracted by creating connectivity matrices and graph representations of brain networks. Several machine learning models were applied, including K-means clustering, Gaussian Mixture Models, autoencoders, CNNs, and GNNs, to identify intrinsic functional subdivisions in the brain. The resulting clusters were converted into spatially coherent brain regions, and their quality was evaluated using silhouette scores, reproducibility tests, and comparisons with established functional atlases.

Results showed that the proposed machine learning framework produced better functional clustering, achieving a higher silhouette score compared to atlas-based, K-means, and spectral clustering methods. Overall, the study demonstrates that machine learning-based approaches can improve the accuracy, adaptability, and reliability of functional brain parcellation, contributing to more precise and personalized brain mapping in neuroscience.

Conclusion

This study investigated the improvement of functional brain parcellation through the integration of advanced machine learning techniques applied to large-scale resting-state fMRI datasets, including those from the Human Connectome Project. The findings demonstrate that data-driven machine learning approaches significantly enhance the functional homogeneity, reproducibility, and predictive utility of brain parcellation compared to traditional atlas-based and classical clustering methods. The proposed model achieved superior performance across multiple evaluation metrics, including higher silhouette scores, lower Davies–Bouldin indices, and improved Dice similarity coefficients. These results indicate that machine learning models particularly deep learning and graph-based architectures are capable of capturing complex, non-linear functional connectivity patterns that conventional methods may overlook. Moreover, enhanced parcellation quality translated into measurable gains in downstream predictive tasks, underscoring the practical utility of refined functional segmentation in cognitive modeling and potential clinical applications. Despite these advances, challenges remain regarding model interpretability, computational demands, and cross-dataset generalizability. Future research should focus on integrating multimodal imaging data, applying explainable artificial intelligence techniques, and validating findings across diverse populations and scanning environments.

References

[1] Y. Yang, C. Ye, and T. Ma, “A deep connectome learning network using graph convolution for connectome-disease association study,” Neural Netw., vol. 164, pp. 91–104, Jul. 2023, doi: 10.1016/j.neunet.2023.04.025. [2] J. Wang et al., “A Novel Deep Learning Tractography Fiber Clustering Framework for Functionally Consistent White Matter Parcellation Using Multimodal Diffusion MRI and Functional MRI,” in 2025 IEEE 22nd International Symposium on Biomedical Imaging (ISBI), Apr. 2025, pp. 1–5. doi: 10.1109/ISBI60581.2025.10980663. [3] D. S. Pishvai, A. Rondinella, F. Guarnera, and S. Battiato, “Advantages of brain parcellation in Multiple Sclerosis Lesion Segmentation,” in 2024 IEEE International Conference on Bioinformatics and Biomedicine (BIBM), Dec. 2024, pp. 6646–6653. doi: 10.1109/BIBM62325.2024.10822162. [4] Y. Li, A. Liu, X. Fu, M. J. Mckeown, Z. J. Wang, and X. Chen, “Atlas-guided parcellation: Individualized functionally-homogenous parcellation in cerebral cortex,” Comput. Biol. Med., vol. 150, p. 106078, Nov. 2022, doi: 10.1016/j.compbiomed.2022.106078. [5] J. Tan et al., “Application of improved graph convolutional network for cortical surface parcellation,” Sci. Rep., vol. 15, no. 1, p. 16409, May 2025, doi: 10.1038/s41598-025-00116-0. [6] J. Zhu et al., “BDEC: Brain Deep Embedded Clustering Model for Resting State fMRI Group-Level Parcellation of the Human Cerebral Cortex,” IEEE Trans. Biomed. Eng., vol. 73, no. 2, pp. 644–657, Feb. 2026, doi: 10.1109/TBME.2025.3590258. [7] L. A. Maglanoc et al., “Brain Connectome Mapping of Complex Human Traits and Their Polygenic Architecture Using Machine Learning,” Biol. Psychiatry, vol. 87, no. 8, pp. 717–726, Apr. 2020, doi: 10.1016/j.biopsych.2019.10.011. [8] C. Li et al., “DDEvENet: Evidence-based ensemble learning for uncertainty-aware brain parcellation using diffusion MRI,” Comput. Med. Imaging Graph., vol. 120, p. 102489, Mar. 2025, doi: 10.1016/j.compmedimag.2024.102489. [9] H. M. Lorzel and M. D. Allen, “Development of the next-generation functional neuro-cognitive imaging protocol - Part 1: A 3D sliding-window convolutional neural net for automated brain parcellation,” NeuroImage, vol. 286, p. 120505, Feb. 2024, doi: 10.1016/j.neuroimage.2023.120505. [10] W. Yin, S. Mostafa, and F. Wu, “Diagnosis of Autism Spectrum Disorder Based on Functional Brain Networks with Deep Learning,” J. Comput. Biol., vol. 28, no. 2, pp. 146–165, Feb. 2021, doi: 10.1089/cmb.2020.0252. [11] M. Mohammadi and S. Jamshidi, “Enhancing Brain Tumor Classification Using TrAdaBoost and Multi-Classifier Deep Learning Approaches,” Oct. 31, 2024, arXiv: arXiv:2411.00875. doi: 10.48550/arXiv.2411.00875. [12] L. Wang et al., “Functional Brain Network Measures for Alzheimer’s Disease Classification,” IEEE Access, vol. 11, pp. 111832–111845, 2023, doi: 10.1109/ACCESS.2023.3323250. [13] Y. Zhu, X. Li, C. Niu, F. Wang, and J. Ma, “Geometric deep learning with adaptive full-band spatial diffusion for accurate, efficient, and robust cortical parcellation,” Med. Image Anal., vol. 101, p. 103492, Apr. 2025, doi: 10.1016/j.media.2025.103492. [14] G. Lohmann et al., “Improving the reliability of fMRI-based predictions of intelligence via semi-blind machine learning,” Nov. 30, 2023, bioRxiv. doi: 10.1101/2023.11.03.565485. [15] A. Kujawa, R. Dorent, S. Ourselin, and T. Vercauteren, “Label Merge-and-Split: A Graph-Colouring Approach for Memory-Efficient Brain Parcellation,” in Medical Image Computing and Computer Assisted Intervention – MICCAI 2024, M. G. Linguraru, Q. Dou, A. Feragen, S. Giannarou, B. Glocker, K. Lekadir, and J. A. Schnabel, Eds., Cham: Springer Nature Switzerland, 2024, pp. 350–360. doi: 10.1007/978-3-031-72114-4_34. [16] B. Wang et al., “Multigroup recognition of dementia patients with dynamic brain connectivity under multimodal cortex parcellation,” Biomed. Signal Process. Control, vol. 76, p. 103725, Jul. 2022, doi: 10.1016/j.bspc.2022.103725. [17] M. Ghofrani-Jahromi et al., “Prognostic enrichment for early-stage Huntington’s disease: An explainable machine learning approach for clinical trial,” NeuroImage Clin., vol. 43, p. 103650, Jan. 2024, doi: 10.1016/j.nicl.2024.103650. [18] S. Hahn et al., “Performance scaling for structural MRI surface parcellations: a machine learning analysis in the ABCD Study,” Cereb. Cortex, vol. 33, no. 1, pp. 176–194, Jan. 2023, doi: 10.1093/cercor/bhac060. [19] J. Gómez-Ramírez, M. A. Fernández-Blázquez, and J. J. González-Rosa, “Prediction of Chronological Age in Healthy Elderly Subjects with Machine Learning from MRI Brain Segmentation and Cortical Parcellation,” Brain Sci., vol. 12, no. 5, Apr. 2022, doi: 10.3390/brainsci12050579. [20] S. Huang et al., “Towards Accurate Fetal Brain Parcellation via Hierarchical Network and Loss,” in Perinatal, Preterm and Paediatric Image Analysis, D. Link-Sourani, E. Abaci Turk, C. Macgowan, J. Hutter, A. Melbourne, and R. Licandro, Eds., Cham: Springer Nature Switzerland, 2025, pp. 70–81. doi: 10.1007/978-3-031-73260-7_7. [21] P. Khandelwal et al., “Surface-Based Parcellation and Vertex-wise Analysis of Ultra High-resolution ex vivo 7 tesla MRI in Alzheimer’s disease and related dementias,” in Machine Learning in Clinical Neuroimaging, D. R. Bathula, A. Benet Nirmala, N. C. Dvornek, S. T. Govindarajan, M. Habes, V. Kumar, A. Nebli, T. Wolfers, and Y. Xiao, Eds., Cham: Springer Nature Switzerland, 2025, pp. 134–144. doi: 10.1007/978-3-031-78761-4_13. [22] S. Huang, K. Zhang, F. Zhu, Z. Ding, G. Chen, and D. Shen, “Semi-supervised Fetal Brain Parcellation via Hierarchical Learning Framework,” Med. Image Anal., vol. 107, p. 103835, Jan. 2026, doi: 10.1016/j.media.2025.103835. [23] M. R. Ahmed, Y. Zhang, Y. Liu, and H. Liao, “Single Volume Image Generator and Deep Learning-Based ASD Classification,” IEEE J. Biomed. Health Inform., vol. 24, no. 11, pp. 3044–3054, Nov. 2020, doi: 10.1109/JBHI.2020.2998603. [24] K. B. See, D. J. Arpin, D. E. Vaillancourt, R. Fang, and S. A. Coombes, “Unraveling somatotopic organization in the human brain using machine learning and adaptive supervoxel-based parcellations,” NeuroImage, vol. 245, p. 118710, Dec. 2021, doi: 10.1016/j.neuroimage.2021.118710. [25] F. Zhao, Z. Wu, L. Wang, W. Lin, S. Xia, and G. Li, “A Deep Network for Joint Registration and Parcellation of Cortical Surfaces,” in Medical Image Computing and Computer Assisted Intervention – MICCAI 2021, M. de Bruijne, P. C. Cattin, S. Cotin, N. Padoy, S. Speidel, Y. Zheng, and C. Essert, Eds., Cham: Springer International Publishing, 2021, pp. 171–181. doi: 10.1007/978-3-030-87202-1_17.

Copyright

Copyright © 2026 Babagana Ali Dapshima, Samaila Kasimu Ahmad, Azira Bata Mshelia, Falmata Mustapha, Salisu Abubakar Sadiq, Mohammed Hussaini Imam. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.