A cross-regional visual encoding method based on multi-gate mixture-of-experts (MMoE) is proposed to address the problem that functional magnetic resonance imaging (fMRI) signals of various visual regions in the human brain are heterogeneous, and existing single-region modeling methods ignore the influence of other brain regions and fail to effectively predict fMRI signals. First, the encoding of adjacent visual cortices is converted into a multi-task learning problem to simulate cross-regional interactions. Second, a parallel expert network is constructed to capture cross-regional shared features and region-specific information. Finally, an adaptive gating function is designed to dynamically fuse the outputs of experts, further to simulate the cross-regional interaction mechanism of the biological visual system. Experimental results demonstrate that the prediction accuracy of the model in each subregion of V1~V3 is significantly higher than that of the single-region baseline model, with the proportion of dominant voxels in V1v and V1d reaching 81.96% and 92.38%, respectively, validating the effectiveness of the visual encoding method integrating multi-region information.
HUANGL Q. Comprehensive exploration of visual working memory mechanisms using large-scale behavioral experiment[J]. Nature Communications, 2025,16:No.1383.
[2]
PARKJ, SOUCYE, SEGAWAJ, et al. Immersive scene representation in human visual cortex with ultra-wide-angle neuroimaging[J]. Nature Communications, 2024,15:No.5477.
[3]
ZHOUY K, GAOS, DENGL L, et al. A group based network analysis for Alzheimer’s disease fMRI data[J]. Scientific Reports, 2025,15:No.10888.
[4]
WANGE Y, FAHEYP G, DINGZ K, et al. Foundation model of neural activity predicts response to new stimulus types[J]. Nature, 2025,640(8058):470-477.
[5]
NONAKAS, MAJIMAK, AOKIS C, et al. Brain hierarchy score: which deep neural networks are hierarchically brain-like?[J]. iScience, 2021,24(9):No.103013.
[6]
KHERADPISHEHS R, GHODRATIM, GANJTABESHM, et al. Humans and deep networks largely agree on which kinds of variation make object recognition harder[J]. Frontiers in Computational Neuroscience, 2016,10:No.92.
[7]
POSPISILD A, PASUPATHYA, BAIRW. ‘Artiphysiology’reveals V4-like shape tuning in a deep network trained for image classification[J]. eLife, 2018,7:No.38242.
[8]
LASKARM N U, GIRALDOL G S, SCHWARTZO. Correspondence of deep neural networks and the brain for visual textures[EB/OL].(2018-06-07)[2025-11-04].
[9]
DODGES, KARAML. Can the early human visual system compete with deep neural networks[C]∥Proceedings of 2017 IEEE International Conference on Computer Vision Workshops. Piscataway,USA: IEEE, 2017:2798-2804.
[10]
GRUBERL Z, HARUVIA, BASRIR, et al. Perceptual dominance in brief presentations of mixed images: human perception vs. deep neural networks[J]. Frontiers in Computational Neuroscience, 2018,12:No.57.
[11]
FELLEMAND J, VAN ESSEND C. Distributed hierarchical processing in the primate cerebral cortex[J]. Cerebral Cortex, 1991,1(1):1-47.
[12]
MALACHR, LEVYI, HASSONU. The topography of high-order human object areas[J]. Trends in Cognitive Sciences, 2002,6(4):176-184.
[13]
DICARLOJ J, ZOCCOLAND, RUSTN C. How does the brain solve visual object recognition?[J]. Neuron, 2012,73(3):415-434.
[14]
ZHONGH X, WANGR B. A new discovery on visual information dynamic changes from V1 to V2: corner encoding[J]. Nonlinear Dynamics, 2021,105(4):3551-3570.
[15]
TOOTELLR B H, DALEA M, SERENOM I, et al. New images from human visual cortex[J]. Trends in Neurosciences, 1996,19(11):481-489.
[16]
ZHENGY J, JIAS S, YUZ F, et al. Unraveling neural coding of dynamic natural visual scenes via convolutional recurrent neural networks[J]. Patterns, 2021,2(10):No.100350.
[17]
SUNJ, LIM, CHENZ, et al. Contrast, attend and diffuse to decode high-resolution images from brain activities[J]. Advances in Neural Information Processing Systems, 2023, 36: 12332-12348.
[18]
ZHONGH X, WANGH Y, DAIW P, et al. Emergent orientation maps: mechanisms, coding efficiency and robustness[EB/OL].(2025-05-15)[2025-05-30].
[19]
ALLENE J, ST-YVESG, WUY H, et al. A massive 7T fMRI dataset to bridge cognitive neuroscience and artificial intelligence[J]. Nature Neuroscience, 2021,25(1):116-126.