Identification of morphological fingerprint in perinatal brains using quasi-conformal mapping and contrastive learning

Identification of morphological fingerprint in perinatal brains using quasi-conformal mapping and contrastive learning | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Identification of morphological fingerprint in perinatal brains using quasi-conformal mapping and contrastive learning Boyang Wang, Weihao Zheng, Ying Wang, Dalin Zhu, Yuchen Sheng This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-4602847/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 27 Mar, 2025 Read the published version in Brain Imaging and Behavior → Version 1 posted 11 You are reading this latest preprint version Abstract The morphological fingerprint in the brain is capable of identifying the uniqueness of an individual. However, whether such individual patterns are present in perinatal brains, and which morphological attributes or cortical regions better characterize the individual differences of neonates remain unclear. In this study, we proposed a deep learning framework that projected three-dimensional spherical meshes of three morphological features (i.e., cortical thickness, mean curvature, and sulcal depth) onto two-dimensional planes through quasi-conformal mapping, and employed the ResNet18 and contrastive learning for individual identification. We used the cross-sectional structural MRI data of 461 infants, incorporating with data augmentation, to train the model and fine-tuned the parameters based on 40 infants who had longitudinal scans. The model was validated on a fold of 20 longitudinal scanned infant data, and remarkable Top1 and Top5 accuracies of 85.90% and 92.20% were achieved, respectively. The sensorimotor and visual cortices were recognized as the most contributive regions in individual identification. Moreover, the folding morphology demonstrated greater discriminative capability than the cortical thickness. These findings provided evidence for the emergence of morphological fingerprints in the brain at the beginning of the third trimester, which may hold promising implications for understanding the formation of individual uniqueness in the brain during early development. morphological fingerprint MRI conformal mapping deep learning contrastive learning perinatal period Figures Figure 1 Figure 2 Figure 3 Figure 4 1. Introduction Significant progress has been achieved in understanding brain development and organization due to advances in neuroimaging and neuroscience (Giedd et al., 2010 [ 1 ]; Mills et al. 2014 [ 2 ]). There is growing awareness of the importance of differences in brain structure and function among individuals. These inter-individual variations may underlie differences in cognitive abilities, emotional processing, and susceptibility to neurological and psychiatric disorders (Dannlowski et al., 2012 [ 3 ]; Karama et al., 2014 [ 4 ]; Schmaal et al., 2017 [ 5 ]). Some researchers have explored the potential of brain function and structure for the usage of biometric identification (Bassi et al., 2018 [ 6 ]; Chen et al., 2018 [ 7 ]). While prior investigations have revealed that the adult brain exhibited stable structural and functional fingerprints for representing individual differences (Finn et al., 2015 [ 8 ]; Menon et al., 2019 [ 9 ]), it remains unclear whether the fingerprint in the brain emerges as early as during the third trimester—a critical period marked by the explosive growth of cortical anatomy and rapid establishment of structural and functional connectome. Magnetic Resonance Imaging (MRI) is a non-invasive imaging technique known for its capacity to yield high-resolution images for the examination of both brain structure and function. This technology has been widely applied to study infant brains, such as the early developmental patterns of morphology, microstructure, fiber tracts, and brain connectomes (Liu et al., 2021 [ 10 ]; Liu et al., 2021 [ 11 ]; Zheng et al., 2023 [ 12 ] ; Zheng et al., 2023 [ 13 ]). Studies have characterized individual differences in human brains through anatomical and functional connectivity. For example, the white matter tractography in adult brains could serve as an effective fingerprint for the identification of individuals (Yeh et al., 2016 [ 14 ]), and the identification accuracy of functional connectivity increases with age from childhood to adulthood (Vanderwal et al., 2021 [ 15 ]). Furthermore, recent studies have attempted to investigate the fingerprint in perinatal brains and achieved recognition rates of 62.22% and 78% based on structural (Ciarrusta et al., 2022 [ 16 ]) and functional (Hu et al., 2022 [ 17 ]) connectivity, respectively. These results demonstrated the fact that individual uniqueness of brain connectome emerges during early brain development. Similar as brain connectivity, cortical morphology could also serve as a valuable fingerprint for individual recognition (Wachinger et al., 2015 [ 18 ]; Aloui et al., 2018 [ 19 ]), which achieved remarkable performance that superior to the functional connectivity in differentiating adult individuals (Tian et al, 2021 [ 20 ]). In recent years, some studies have begun to explore cortical folding patterns for infant subject identification. For instance, Duan et al. successfully identified 1- and 2-year-old infants by using cortical folding information (i.e., curvature, convexity, and sulcus depth) of the corresponding neonate (Duan et al.,2019 [ 21 ]; Duan et al.,2020 [ 22 ]). Nevertheless, it remains unclear whether the individual variations in human brain morphology already appear as early as the beginning of the third trimester. Deep learning methods can achieve more advanced feature representations in brain images. Multiple deep learning models have been applied to resolve challenges in brain image analysis in recent years. For example, the convolutional neural network (CNN) has been utilized for the detection of brain lesion (Chen et al., 2020 [ 23 ]) and white matter abnormalities (McKinley et al., 2019 [ 24 ]), as well as for the diagnosis of brain disorders (Esmaeilzadeh et al., 2018 [ 25 ]; Qureshi et al., 2019 [ 26 ]). Convolutional networks based on the whole-brain cortical surface (Mostapha et al. 2018 [ 27 ]) have offered new perspectives for studying human brain MRI and cortical morphology. However, 3D-CNN methods are often unsuitable for small-sample datasets due to the large number of parameters and high computational demands. On the other hand, 2D-CNN methods have lower computational requirements and can be more easily embedded into a mature model architecture. T. W. Meng et al. have introduced Teichmüller Extremal Mapping of POint clouds (TEMPO), a quasi-conformal mapping method for conformally mapping a simply-connected open triangle mesh to a 2D rectangle space [ 28 ]. The TEMPO method can effectively preserve conformality, reducing the loss of local features and geometric structures in the mapping of the original point cloud. Leveraging deep learning methods enables the automatic acquisition of more high-dimensional features of cortical morphology through multilevel nonlinear transformations, thereby simplifying the model’s principles and procedures via end-to-end learning. Therefore, we utilized quasi-conformal mapping to project the 3D brain mesh onto a 2D plane and employed 2D-CNN to extract individual cortical morphological feature representations. The present study aims to validate the existence of morphological fingerprints in perinatal brains. Each hemispheric surface of an individual subject was inflated to a sphere and was then projected to a 2D plane through TEMPO. We propose a contrastive learning framework based on the pretrained ResNet18 encoder to recognize an individual at term-equivalent age by using his or her brain MRI acquired at birth. The attention mechanism is incorporated to fuse features from different partitions generated from the 3D-to-2D mapping. The effectiveness of each morphological feature is assessed to demonstrate their contributions to the individual fingerprint in perinatal brains. 2. Materials and Methods 2.1 Dataset Imaging data used in this study was obtained from the Developing Human Connectome Project (dHCP-release 3, https://biomedia.github.io/dHCP-release-notes/acquire.html ). The dHCP study was conducted in accordance with the Declaration of Helsinki and was approved by the United Kingdom Health Research Ethics Authority (reference no. 14/LO/1169). Written informed consent was obtained from all participating families before the scan. All participants received MRI scans performed using a 3T Philips Achieva system [ 29 ]. T2-weighted (T2w) multi-slice fast spin-echo images were obtained in both sagittal and axial slice stacks. These images had an in-plane resolution of 0.8x0.8mm² and 1.6mm slices, with an overlap of 0.8mm. The imaging parameters were as follows: repetition time (TR) = 12000ms, echo time (TE) = 156ms, SENSE factor of 2.11 for axial and 2.60 for sagittal slices. Additionally, a 3D MPRAGE scan was performed with 0.8mm isotropic resolution. The dataset included 783 infants in total, comprising 682 infants who underwent MRI scanning one time and 101 infants who were scanned twice (scanned at approximately 1–2 weeks after birth and term-equivalent age, respectively) or more. We selected 400 sets of once sampling data and 62 pairs of twice sampling data for the experiment. Participants who have low-quality imaging data or focal brain abnormality (radiology score > 3, an evaluation by the dHCP) were discarded. Detailed demographic information of the selected infants are given in Table 1 . Table 1 Demographic information of the infants included in this study. Groups Median birth age (weeks) Birth age range (weeks) Median scan age (weeks) Scan age range (weeks) Male/ Female Cross-sectional cohort 39.86 [24.71–43.57] 41.14 [26.86–45.14] 215/ 185 Longitudinal cohort (1st scan) 31.86 [23.86–40.14] 34.71 [26.71–42.71] 32/29 Longitudinal cohort (2nd scan) 31.86 [23.86–40.14] 41.29 [38.43–44.86] 32/29 2.2 Dataset Processing Image preprocessing followed the pipeline proposed by the dHCP [ 30 ]. In summary, the process first involved bias correction and brain extraction of the motion-corrected T2-weighted image. This was followed by segmenting the brain into different tissue types using Draw-EM algorithm. White-matter mesh was then extracted and expanded to fit the pial surface. The cortical thickness was estimated based on the Euclidean distance between the white and pial surface. We generated the inflated surfaces from the white surface, which were then projected to a sphere for surface registration. The mean surface curvature and sulcal depth (mean surface convexity/concavity) were respectively estimated from the white surface and from inflation. 2.3 Teichmüller Mapping Despite the promising spherical convolutional neural network (SCNN) models on (Cohen et al., 2018 [ 31 ]; C. Esteves et al., 2018 [ 32 ]; C. Jiang et al., 2019 [ 33 ]), the SCNNs often suffer from substantial computational complexity, e.g., extensive parameters, which make them difficult to converge on small sample datasets. Therefore, we projected the 3-dimensional spherical mesh of the brain surface to 2-dimensional and employed the convolutional neural network (CNN) model for analysis. To achieve this goal, a quasi-conformal mapping method (Teichmüller Mapping) was employed, which is a geometric transformation method based on complex variational functions. Conformal mapping facilitates the mapping of one space to another while preserving local angles between intersecting curves or surfaces, thereby ensuring the preservation of shapes and angles within a small region of the original space. In general, its formula representation is as follows: $$f\text{*}d{s}_{N}^{2}=\lambda d{s}_{M}^{2}$$ where M and N are two Riemann surfaces, f is the mapping function: M→N, and λ is a positive scalar function. A generalization of conformal mapping is quasi-conformal mapping, which allows for a certain degree of angle and shape distortion to adapt to more complex data structures and transformations. Teichmüller Mapping is such a type of quasi-conformal mapping that can induce uniform conformality distortions in the target point cloud, thereby preserving stable relative positions and local shapes. The general formula for a Teichmüller mapping (T-map) can be expressed as follows: let M and N be two Riemann surfaces, and \(f:M\to N\) be a quasi-conformal mapping. If it is associated with a quadratic differential \(q=\varphi d{z}^{2}\) , where \(\varphi :M\to \mathbb{C}\) is a holomorphic function, and its associated Beltrami coefficient is of the form \(\mu \left(f\right)=k\frac{\varphi }{\left|\varphi \right|}\) , where k<1 is a constant, and the quadratic differential q is non-zero and satisfies \(\left|\right|q{\left|\right|}_{1}={\int }_{{S}^{1}}\left|\varphi \right|<{\infty }\) , then f is called a Teichmüller mapping (T-map) associated with the quadratic differential \(q\) . In the present study, we used the TEMPO method for projection. Specifically, we adopt an approach of segmenting the hemispherical plane along the x = 0 plane in the coordinate system and subsequently applying the TEMPO method separately to the two hemispheres for mapping. Then, we interpolated the mapped mesh to obtain a two-dimensional matrix. Although this approach sacrificed the continuity of the data around the segmentation curve, it significantly reduced the area distortion from mapping deformation and preserved the overall continuity of other areas, offering a more desirable outcome for our study. The projected meshes on 2D plane are shown in Fig. 1 . 2.4 Data Augmentation Data augmentation is a commonly employed technique to address the issue of insufficient data and lack of data diversity. Its primary objective is to expand the dataset by applying diverse transformations and processing to the original data. In the context of images, various augmentation methods have been established, including random cropping, rotation, flipping, random noise, Gaussian blur, and color transformations such as brightness, contrast, and saturation adjustments (Cubuk et al., 2018 [ 34 ]; Shorten et al., 2019 [ 35 ]; T. Chen et al., 2020 [ 36 ]). We used data augmentation to expand single sampling data into sample pairs, which can be used for model pre-training. We employed rotation, random noise, and Gaussian blur methods as shown in Fig. 1 . 2.5 Model Architecture As illustrated in Fig. 2 a, we present a contrastive learning-based feature extraction framework. We aligned the 3D grids of the left and right brain hemispheres onto spheres, and then partitioned each sphere into two hemispheres. Subsequently, we employed the TEMPO method to map the spherical grids onto planar grids and compute four 3×224×224 matrices through interpolation. These cortical morphological matrices were then fed into the feature extraction module to extract cortical morphological feature fingerprints of individual subjects (Fig. 2 b). These fingerprints (vectors of length 512) were ultimately utilized for calculating pairwise similarities for individual identification. Notably, in the feature extraction process, we employed a channel-wise attention mechanism-based excitation module for fusing features from different brain partitions and weight allocation to feature map channels, as shown in Fig. 2 c. 2.6 Contrastive Learning Contrastive learning is a self-supervised method for training models to learn meaningful data representations. It works by comparing the similarity between pairs of data samples. In essence, it creates pairs of positive and negative samples and uses a loss function to make the feature representations of positive pairs more similar while making those of negative pairs more dissimilar. A notable framework for contrastive learning is SimCLR (short for Simple Contrastive Learning of Representations) [ 34 ], which used the NT-Xent loss function to maximize the similarity of positive pairs and minimize the similarity of negative pairs in the feature space. The NT-Xent loss function is defined as: $$L=-\frac{1}{N}\sum _{i=1}^{N}\left(\text{l}\text{o}\text{g}\left(\frac{\text{e}\text{x}\text{p}\left(-\parallel {z}_{i}-{z}_{i}^{+}\parallel /\tau \right)}{\sum _{j=1,j\ne i}^{N}\text{e}\text{x}\text{p}\left(-\parallel {z}_{i}-{z}_{j}^{+}\parallel /\tau \right)}\right)\right)$$ where N is the batch size, \(\parallel {z}_{i}-{z}_{i}^{+}\parallel\) is the Euclidean distance between the i -th sample and its positive sample, \(\parallel {z}_{i}-{z}_{j}^{+}\parallel\) is the Euclidean distance between the i -th sample and the j -th sample’s corresponding positive sample. τ denotes a temperature parameter. In this study, we used the SimCLR as the main contrastive learning framework to extract morphological feature representations of individual differences in the perinatal cerebral cortex. We constructed positive sample pairs by using multiple data augmentation methods on single sampling data and used the other samples in the same batch to construct negative pairs. Specifically, we adopted a contrastive loss similar to the classical Siamese network [ 37 ]. We constructed multiple negative sample pairs while adopting a non-softmax computation of the loss function. The cross-entropy loss that was separate for the positive and negative sample pairs was calculated and summed up. The loss function was defined as follows: $$L=\sum _{i=1}^{N}\left(\left(1-{y}_{i}\right)\text{*}{\text{dist}}_{i}^{2}+y\text{*}{\text{clamp}}_{\text{min}}{\left(m-{\text{dist}}_{i},0\right)}^{2}\right)$$ $${\text{dist}}_{i}=\parallel {x}_{1i}-{x}_{2i}\parallel ,{\text{clamp}}_{\text{min}}\left(a,b\right)=\left\{\begin{array}{ll}a,& \text{if }a\ge b\\ b,& \text{if }a<b\end{array}\right.$$ where N is the batch size, \({y}_{i}\) is the label indicating whether the i -th sample pair was a positive or negative sample, \({x}_{1i}\) and \({x}_{2i}\) represent the two samples of the i-th sample pair respectively, and m was an artificially set margin. In this framework, we used the ResNet18 backbone (He et al., 2016 [ 32 ]) as the brain map encoder, as shown in Fig. 2 -(b). For each morphological feature, four sets of projection maps (medial and lateral projections of the left and right brain, respectively) per sample were fed into the parameter-sharing ResNet18 to extract four feature maps. 2.7 Excitation Module The excitation module aimed to enhance the overall representation capability of the feature vectors to the morphological features of the samples through the channel attention mechanism, as shown in Fig. 2 c. It achieved the fusion of features from four different partitions. Similar to the SE-Nets (Hu et al., 2018 [ 38 ]), the channel weights were learned by two fully connected layers that mapped the feature vector to a channel attention vector. This channel attention vector was applied to the input feature map to weigh the channels. However, different from the previous literature, our excitation module optimized the assignment of the weight to three channels of morphological features and four channels (partitions) of projection maps simultaneously, which was arranged before the final layer of the neckbone network. Adding a hyperparameter weight_scale was intended to restore the weights to a distribution with a mean of 1 or approximately 1 to maintain the consistency of the data scale. The feature vector of the sample was obtained by pooling and a simplified projection through a fully connected layer. In addition to the excitation method, we employed other two fusion techniques: a decision-level fusion approach based on a voting mechanism and a feature-level fusion method using a two-layer Multilayer Perceptron (MLP). The former averaged similarity matrices calculated on four partitions and then made identification decisions based on the averaged similarity matrix. The latter utilized a two-layer MLP to map the input features. 2.8 Training Procedure and Evaluation Metrics For the training process, we initially excluded the excitation module and employed the original structure during the training of the ResNet18 backbone. The network parameters were trained by using primarily augmented sample pairs for pre-training and several two-shot data for fine-tuning, as illustrated in the stage1 of Fig. 3 . After that, we froze the ResNet18 and updated the parameters in the excitation module and the FC layer (stage2 of Fig. 3 ). Finally, the trained feature extraction module was applied to the twice scanned samples. We calculated and compared the similarity of the feature vectors between two time points of the same subject (self-self similarity) and the feature similarity between a neonate at birth and other infants scanned at term-equivalent age (self-other similarity). The Euclidean distance was used to measure the similarity between individual brains (Eickhoff et al.,2005 [ 39 ]; Al-Saffar et al., 2020 [ 40 ]). We utilized the Top 1 accuracy (where self-similarity surpasses all self-other similarities) and Top 5 accuracy (where self-similarity ranks among the top five similarities between itself and all other samples) as metrics to assess the effectiveness of the model. 3. Results 3.1. Experimental Setup We used the ResNet18 backbone network and its pre-trained network parameters provided by Pytorch on the ImageNet dataset for initialization. To train the ResNet18 network, we augmented all single-sampled data to create positive and negative sample pairs. These pairs were then used for the initial training of ResNet18. Subsequently, we performed fine-tuning of ResNet18 and trained the feature fusion module. The training process employed a learning rate of 5e-4 with a momentum of 0.9 and a weight decay of 5e-5. Each training session comprised 8 epochs and the SGD optimizer was used. For all experiments, we conducted triple-fold cross-validation using twice sampled data (i.e., the real sample pairs). In each fold, we used 40 and 20 real sample pairs for training and testing, respectively. We repeated the experiments for 30 rounds to ensure robustness and obtained weighted accuracies for identification. 3.2. Experiment Results 3.2.1. Individual Recognition Our model achieved a notable Top 1 accuracy of 85.90% and a Top 5 accuracy of 92.20%. These results suggested that the self-similarity of the prenatal brains between birth and term-equivalent ages tended to be higher than self-other similarities. We also compared the accuracies derived from different fusion strategies for fusing features from left lateral, left medial, right lateral and right medial brain. (e.g., excitation, voting and MLP). The excitation method outperformed other fusion methods. The voting strategy that integrated judgments from different partitions also demonstrated high Top 1 accuracy of 85.25% and Top 5 accuracy of 91.75%. The method using a MLP for feature fusion achieved the Top 1 accuracy of 77.78% and Top 5 accuracy of 87.39%. Overall, the excitation method exhibited superior performance. The detailed model performance is shown in Table 2 . Table 2 Model performance comparison between different fusion methods and different feature channels selected. Fusion methods Accuracy (%) Morphological Features Cortical thickness Mean curvature Sulcal depth All Excitation TOP 1 25.00 82.00 73.34 85.90 TOP 5 60.35 87.65 85.00 92.20 MLP TOP 1 29.31 75.08 71.45 77.78 TOP 5 49.81 85.08 82.13 87.39 Decision Level TOP 1 37.15 81.67 73.65 85.25 TOP 5 61.17 87.67 84.00 91.75 Concatenation TOP 1 23.65 81.65 73.34 16.70 TOP 5 59.35 87.00 83.00 55.70 3.2.2. Contributions of Morphological Features and Brain Regions To explore the contributions of each morphological feature to the recognition task, we conducted single-channel comparison experiments for the three morphological features: curvature, thickness, and sulcus. Mean curvature achieved the highest Top 1 and Top 5 accuracies of 82.00% and 87.65%, respectively, which outperformed the other two features. Cortical thickness did not exhibit any discriminative power. We then explored the contributions of different brain regions in the recognition task. We utilized the weights derived from the excitation module as the contribution weights of brain regions shown in Fig. 4 a, and the weights of brain regions were mapped back to the cortical surface (as illustrated in Fig. 4 b. The parietal and occipital cortices of the right hemisphere and the antero-medial temporal cortices of the right hemisphere showed higher weights relative to other regions, suggesting these regions may have evident morphological differences among individuals. 4. Discussion In this study, we proved the existence of morphological fingerprints in neonatal cerebral cortex through a deep learning model. We achieved a remarkable Top 1 accuracy of 85.90% in the individual recognition. This suggested that certain cortical morphological fingerprints are already formed as early as the beginning of the third trimester and maintained stable during the perinatal period, which can serve as an effective fingerprint for recognizing individual neonate. Training the ResNet18 encoder with a contrastive learning approach significantly improved the recognition accuracy, demonstrating the effectiveness of the framework in learning cortical morphological features. Additionally, the incorporation of the excitation module based on attention mechanisms also greatly enhanced the model performance, outperforming both the comparative voting and MLP methods. We speculated that the improvement might be attributed to the excitation module enabling the model to prioritize brain regions with significant individual variability, reducing the impact of other regions on individual recognition and thus enhancing identification accuracy (Bodapati et al., 2021 [ 41 ]). Moreover, the excitation module enabled the direct assessment of the contribution of different brain regions, providing a convenient way to analyze the recognition contribution rates of different brain areas. Our findings indicated that both cortical curvature and sulcal depth exhibited individual variations, and the former achieved the best recognition rate, emphasizing the crucial role of the cortical folding morphology in characterizing individual differences at early developmental stage. On the other hand, cortical thickness did not exhibit discriminative power to individual recognition. Previous studies have suggested that the individual variability of cortical folding patterns has been established at term age (Duan et al., 2019 [ 42 ]), while the cortical thickness showed relative longer maturation period. Specifically, the cortical thickness typically undergoes rapid growth shortly after birth, peaking around 14 months after birth and subsequently decreases (Wang et al., 2019 [ 43 ]). Therefore, we hypothesized that the earlier maturation of folding patterns imparts greater individual variability to cortical morphology, leading to higher accuracy in individual recognition. Additionally, sensitivity to noise may be another potential factor leading to lower recognition efficiency of thickness features. Moreover, we observed that primary cortex (e.g., somatosensory and visual) carried higher attention weights, indicating greater inter-individual differences in these regions. Given that the primary regions experienced more pronounced development than the high-order cortex in the second trimester (Gilmore et al., 2018 [ 44 ]; Duan et al., 2019 [ 42 ]), we speculated that the high attention weights assigned to the primary cortex may be attributed to their stable morphology maturity throughout the third trimester. Our model may hold potential in clinical applications. While previous studies revealed differential patterns of development across brain measures or regions, very few examine heterogeneity in the developmental patterns between persons. Examining within-subject change over time is necessary to depict typical development of an individual as well as identifying possible atypical alterations [ 45 ]. For example, a significant dissimilarity observed in an individual brain compared to its earlier stage may indicate abnormal development. Furthermore, our model largely reduced training parameters by projecting cortical spherical surface to two-dimensional space, which would be suitable for the computation on small dataset, such as infant data. In addition, by accounting for the individual variability in alteration patterns associated with brain disorders, our approach has the potential to correct inter-subject discrepancies in brain morphology and enable clinical models to focus on disease-specific changes. Our work still has some limitations. First, the longitudinal dataset was relatively small. The generalizability of this model and the reproducibility of our findings should be examined on a larger independent dataset. Second, the conformal transformation may also lead to area distortion during projection, especially near the split curve. Such distortion may potentially influence the recognition accuracy. Some novel methods for addressing these challenges should be developed in the future. In addition, our study is still at experimental stage, it is necessary to validate the results from both methodological and clinical perspectives. 5. Conclusions In conclusion, we introduce a deep learning framework that maps the three-dimensional spherical surface onto a two-dimensional image for computational simplification, facilitating the investigation of individual differences in neonatal brain structures. Our study presents the initial evidence of morphological fingerprints emerging during the perinatal period and highlights the potential for identifying atypical brain development of an individual. Declarations Conflicts of Interest The authors declare no conflict of interest. Funding This work was funded by the National Natural Science Foundation of China (62202212) and the STI-2030 major project (2021ZD0200800). Author Contribution Conceptualization, B.W., Y.S., D.Z. and W.Z.; methodology, W.Z., and Y.W.; validation, B.W. and W.Z.; formal analysis, B.W. and Y.W.; data curation, Y.W.; writing---original draft preparation, B.W. and W.Z.; writing---review and editing, W.Z, D.Z., and Y.S.; supervision, W.Z, and Y.S.; funding acquisition, W.Z.. All authors have reviewed and agreed to the published version of the manuscript. Data Availability We thank all the contributors to the dHCP dataset that support our study. This dataset is openly available at http://www.developingconnectome.org. References Giedd, J. N., & Rapoport, J. L. (2010). Structural MRI of pediatric brain development: what have we learned and where are we going? Neuron 67(5) , 728–734. Mills, K. L., Goddings, A. L., Clasen, L. S., Giedd, J. N., & Blakemore, S. J. (2014). The developmental mismatch in structural brain maturation during adolescence. Developmental neuroscience , 36 (3–4), 147–160. Dannlowski, U., Stuhrmann, A., Beutelmann, V., Zwanzger, P., Lenzen, T., Grotegerd, D., Domschke, K., Hohoff, C., Ohrmann, P., Bauer, J., & Lindner, C. (2014). Limbic scars: long-term consequences of childhood maltreatment revealed by functional and structural magnetic resonance imaging. Biological psychiatry , 71 (4), 286–293. Karama, S., Bastin, M. E., Murray, C., Royle, N. A., Penke, L., Muñoz Maniega, S., Gow, A. J., Corley, J., Valdés Hernández, M., Lewis, J. D., & Rousseau (2014). M.É. Childhood cognitive ability accounts for associations between cognitive ability and brain cortical thickness in old age. Molecular psychiatry , 19 (5), 555–559. Schmaal, L., Hibar, D. P., Sämann, P. G., Hall, G. B., Baune, B. T., Jahanshad, N., Cheung, J. W., van Erp, T. G., Bos, D., Ikram, M. A., & Vernooij, M. W. (2017). Cortical abnormalities in adults and adolescents with major depression based on brain scans from 20 cohorts worldwide in the ENIGMA Major Depressive Disorder Working Group. Molecular psychiatry , 22 (6), 900–909. Bassi, M., & Triverbi, P. Human biometric identification through brain print. In 2018 Second International Conference on Electronics, Communication and Aerospace Technology (ICECA) 2018, 1514–1518. Chen, Shiyang, Xiaoping, & Hu (2018). Individual identification using the functional brain fingerprint detected by the recurrent neural network. Brain connectivity , 8 (4), 197–204. Finn, E. S., Shen, X., Scheinost, D., Rosenberg, M. D., Huang, J., & Chun, M. M. (2015). Xenophon Papademetris, and R. Todd Constable. Functional connectome fingerprinting: identifying individuals using patterns of brain connectivity. Nature neuroscience , 18 (11), 1664–1671. Menon, S. S., & Krishnamurthy, K. (2019). A comparison of static and dynamic functional connectivities for identifying subjects and biological sex using intrinsic individual brain connectivity. Scientific reports , 9 (1), 5729. Liu, T., Gao, F., Zheng, W., You, Y., Zhao, Z., Lv, Y., Chen, W., Zhang, H., Ji, C., & Wu, D. (2021). Diffusion MRI of the infant brain reveals unique asymmetry patterns during the first-half-year of development. Neuroimage , 242 , 118465. Liu, T., Zhang, H., You, Y., Zheng, W., Zhao, Z., Liu, T., Su, X., Tian, F., Zhang, Y., & Wu, D. (2021). Brain developmental differences between preterm-born twins and singletons: a multi-modal MRI Study. Jpr , 8 , 276–285. Zheng, W., Wang, X., Liu, T., Hu, B., & Wu, D. (2023). Preterm-birth alters the development of nodal clustering and neural connection pattern in brain structural network at term‐equivalent age. Human Brain Mapping , 44 (16), 5372–5386. Zheng, W., Zhao, L., Zhao, Z., Liu, T., Hu, B., & Wu, D. (2023). Spatiotemporal Developmental Gradient of Thalamic Morphology, Microstructure, and Connectivity from the Third Trimester to Early Infancy. Journal of Neuroscience , 43 (4), 559–570. Yeh, F. C., Vettel, J. M., Singh, A., Poczos, B., Grafton, S. T., Kirk, I., Erickson, Wen-Yih, I., Tseng, & Timothy, D. (2016). Verstynen. Quantifying differences and similarities in whole-brain white matter architecture using local connectome fingerprints. PLoS computational biology 12(11) , e1005203. Vanderwal, T., Eilbott, J., Kelly, C., Frew, S. R., Todd, S., Woodward, M. P., & Milham (2021). and F. Xavier Castellanos. Stability and similarity of the pediatric connectome as developmental measures. NeuroImage 226 , 117537. Ciarrusta, J., Christiaens, D., Fitzgibbon, S. P., Dimitrova, R., Hutter, J., Hughes, E., Duff, E., et al. (2022). The developing brain structural and functional connectome fingerprint. Developmental Cognitive Neuroscience , 55 , 101117. Hu, D., Wang, F., Zhang, H., Wu, Z., Zhou, Z., Li, G., Wang, L., Lin, W., Li, G., & UNC/UMN Baby Connectome Project Consortium. (2022). Existence of functional connectome fingerprint during infancy and its stability over months. Journal of Neuroscience , 42 (3), 377–389. Wachinger, C., Golland, P., Kremen, W., & Fischl, B. (2015). Martin Reuter, and Alzheimer’s Disease Neuroimaging Initiative. BrainPrint: A discriminative characterization of brain morphology. Neuroimage , 109 , 232–248. Aloui, K., Nait-Ali, A., & Naceur, M. S. (2018). Using brain prints as new biometric feature for human recognition. Pattern Recognition Letters 2018, 113 , 38–45. Tian, Y., Yeo, B. T., Cropley, V., & Zalesky, A. (2021). High-resolution connectomic fingerprints: Mapping neural identity and behavior. Neuroimage , 229 , 117695. Duan, D., Xia, S., Wu, Z., Wang, F., Wang, L., Lin, W., & Gilmore, J. H. (2019). Dinggang Shen, and Gang Li. Cortical Foldingprints for Infant Identification. In 2019 IEEE 16th International Symposium on Biomedical ImagingISBI 396–399. Duan, D., Xia, S., Rekik, I., Wu, Z., Wang, L., Lin, W., & Gilmore, J. H. (2020). Dinggang Shen, and Gang Li. Individual identification and individual variability analysis based on cortical folding features in developing infant singletons and twins. Human brain mapping , 41 (8), 1985–2003. Chen, H., Qin, Z., Ding, Y., Tian, L., & Qin, Z. (2020). Brain tumor segmentation with deep convolutional symmetric neural network. Neurocomputing , 392 , 305–313. McKinley, R., Wepfer, R., Aschwanden, F., Grunder, L., Muri, R., Rummel, C., Verma, R., et al. (2019). Simultaneous lesion and neuroanatomy segmentation in multiple sclerosis using deep neural networks. Scientific reports , 11 (1), 1087. Esmaeilzadeh, S., Yang, Y., & Adeli, E. (2018). End-to-end parkinson disease diagnosis using brain mr-images by 3d-cnn. arXiv preprint arXiv :180605233. Qureshi, M. N., Iqbal, J., Oh, & Lee, B. (2019). 3D-CNN based discrimination of schizophrenia using resting-state fMRI. Artificial intelligence in medicine , 98 , 10–17. Mostapha, M., Kim, S. H., Wu, G., Zsembik, L., Pizer, S., & Styner, M. (2018). ISBI. Non-euclidean, convolutional learning on cortical brain surfaces. In 2018 IEEE 15th International Symposium on Biomedical Imaging 527–530. Meng, T., Wei, G., & Choi, P. T. (2016). Lok Ming Lui. Tempo: feature-endowed Teichmuller extremal mappings of point clouds. SIAM Journal on Imaging Sciences , 9 (4), 1922–1962. Hughes, E. J., Winchman, T., Padormo, F., Teixeira, R., Wurie, J., Sharma, M., Fox, M., Hutter, J., Cordero-Grande, L., Price, A. N., & Allsop, J. (2017). A dedicated neonatal brain imaging system. Magnetic resonance in medicine , 78 (2), 794–804. Makropoulos, A., Robinson, E. C., Schuh, A., Wright, R., Fitzgibbon, S., Bozek, J., Counsell, S. J., Steinweg, J., Vecchiato, K., Passerat-Palmbach, J., & Lenz, G. (2018). The developing human connectome project: A minimal processing pipeline for neonatal cortical surface reconstruction. Neuroimage , 173 , 88–112. Cohen, T. S., & Geiger, M. (2018). Jonas Köhler, and Max Welling. Spherical cnns. arXiv preprint arXiv:1801.10130. Esteves, C., Allen-Blanchette, C., Makadia, A., & Daniilidis, K. (2018). Learning so (3) equivariant representations with spherical cnns. In Proceedings of the European Conference on Computer Vision (ECCV). 52–68. Chiyu Max Jiang, Huang, J., & Kashinath, K. (2019). Prabhat, Philip Marcus, and Matthias Nießner. Spherical CNNs on Unstructured Grids. In ICLR (Poster). Cubuk, E. D., Zoph, B., Mane, D., Vasudevan, V., & Quoc, V. (2018). Le. Autoaugment: Learning augmentation policies from data. arXiv preprint arXiv :180509501. Shorten, C., & Taghi, M. (2019). Khoshgoftaar. A survey on image data augmentation for deep learning. Journal of big data , 6 (1), 1–48. Chen, T., Kornblith, S., Norouzi, M., & Hinton, G. A simple framework for contrastive learning of visual representations. In International conference on machine learning 2020, 1597–1607. Chopra, S., Hadsell, R., Yann, & LeCun (2005). Learning a similarity metric discriminatively, with application to face verification. In 2005 IEEE computer society conference on computer vision and pattern recognition (CVPR’05). 539–546. He, K., Zhang, X., Ren, S., & Sun, J. (2016). Deep residual learning for image recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition. 770–778. Eickhoff, S., Walters, N. B., Schleicher, A., Kril, J., Egan, G. F., Zilles, K., Watson, J. D. G., & Amunts, K. (2005). High-resolution MRI reflects myeloarchitecture and cytoarchitecture of human cerebral cortex. Human brain mapping , 24 (3), 206–215. Al-Saffar, Z. A., Tülay, & Yildirim (2020). A novel approach to improving brain image classification using mutual information-accelerated singular value decomposition. Ieee Access : Practical Innovations, Open Solutions , 8 , 52575–52587. Bodapati, J. D., Shareef, S. N., Naralasetti, V., & Mundukur, N. B. (2021). Msenet: Multi-modal squeeze-and-excitation network for brain tumor severity prediction. International Journal of Pattern Recognition and Artificial Intelligence 2021, 35(07) , 2157005. Duan, D., Xia, S., Rekik, I., Meng, Y., Wu, Z., Wang, L., Lin, W., Gilmore, J. H., Shen, D., & Li, G. (2019). Exploring folding patterns of infant cerebral cortex based on multi-view curvature features: Methods and applications. Neuroimage , 185 , 575–592. Wang, F., Lian, C., Wu, Z., Zhang, H., Li, T., Meng, Y., Wang, L., Lin, W., Shen, D., & Li, G. (2019). Developmental topography of cortical thickness during infancy. Proceedings of the National Academy of Sciences 116(32) , 15855–15860. Gilmore, J. H., Knickmeyer, R. C., & Gao, W. (2018). Imaging structural and functional brain development in early childhood. Nature Reviews Neuroscience , 19 (3), 123–137. Becht, A. I., Kathryn, L., & Mills (2020). Modeling individual differences in brain development Biological Psychiatry 88.1 : 63–69. Additional Declarations No competing interests reported. Cite Share Download PDF Status: Published Journal Publication published 27 Mar, 2025 Read the published version in Brain Imaging and Behavior → Version 1 posted Editorial decision: Revision requested 31 Oct, 2024 Reviews received at journal 22 Sep, 2024 Reviews received at journal 19 Sep, 2024 Reviewers agreed at journal 19 Sep, 2024 Reviewers agreed at journal 17 Sep, 2024 Reviewers agreed at journal 11 Aug, 2024 Reviewers agreed at journal 08 Aug, 2024 Reviewers invited by journal 06 Aug, 2024 Editor assigned by journal 06 Aug, 2024 Submission checks completed at journal 20 Jun, 2024 First submitted to journal 18 Jun, 2024 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-4602847","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":316690208,"identity":"a501855a-29ee-403f-aaa9-eb2ce3ef1dde","order_by":0,"name":"Boyang Wang","email":"","orcid":"","institution":"Lanzhou Jiaotong University","correspondingAuthor":false,"prefix":"","firstName":"Boyang","middleName":"","lastName":"Wang","suffix":""},{"id":316690210,"identity":"21f48705-008a-461a-b7ed-9a6c8991a1bf","order_by":1,"name":"Weihao Zheng","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA+ElEQVRIiWNgGAWjYDACCQjFY8DAwPjgAZidQLwWZoMEUrQwALWwSRClRX5287GHX9tsZMzZDx+rSPhzmIGfPceA4ecO3FoY5xxLN5ZtS+Ox7ElLu5HAc5hBsueNAWPvGdxamCVyzKQltx3mMTiQY3YjQeIwg8GNHANmxjbcWtgk8r8BtfznMTj/xqwgweAwgz0hLTwSOWySH7cd4AEabsaQkAC0RYKAFgmJNDNpxn/JQC3PkiUSDqTzSJx5VnCwF48W+RnJzyR/nLGzNziffPDDhz/WcvztyRsf/MSjBRwEPMguBREH8GsABvQPQipGwSgYBaNgZAMAdGdNrTszEvcAAAAASUVORK5CYII=","orcid":"","institution":"Lanzhou University","correspondingAuthor":true,"prefix":"","firstName":"Weihao","middleName":"","lastName":"Zheng","suffix":""},{"id":316690211,"identity":"fe8f55d1-dd9b-4e44-9d6c-0e63e2d506ae","order_by":2,"name":"Ying Wang","email":"","orcid":"","institution":"Lanzhou University","correspondingAuthor":false,"prefix":"","firstName":"Ying","middleName":"","lastName":"Wang","suffix":""},{"id":316690213,"identity":"33308c6a-933e-4a73-a5b1-45d0cdea9cd7","order_by":3,"name":"Dalin Zhu","email":"","orcid":"","institution":"Gansu Provincial Maternal and Child Health Hospital","correspondingAuthor":false,"prefix":"","firstName":"Dalin","middleName":"","lastName":"Zhu","suffix":""},{"id":316690214,"identity":"6af6250a-18bb-4a2e-ac40-eeff74481644","order_by":4,"name":"Yuchen Sheng","email":"","orcid":"","institution":"Lanzhou Jiaotong University","correspondingAuthor":false,"prefix":"","firstName":"Yuchen","middleName":"","lastName":"Sheng","suffix":""}],"badges":[],"createdAt":"2024-06-19 03:23:28","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-4602847/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-4602847/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1007/s11682-025-00998-8","type":"published","date":"2025-03-27T15:57:22+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":60597892,"identity":"1a469b78-d414-43a6-99d8-75c1d302a48d","added_by":"auto","created_at":"2024-07-18 15:51:41","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":564671,"visible":true,"origin":"","legend":"\u003cp\u003eTEMPO projection and data augmentation results of single-sampled data.\u003c/p\u003e","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-4602847/v1/6af0cd4b65a93496ca32e9f3.png"},{"id":60597893,"identity":"44eeb62a-4912-4eb3-bec2-0bef6dc324a3","added_by":"auto","created_at":"2024-07-18 15:51:50","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":390003,"visible":true,"origin":"","legend":"\u003cp\u003eA DL framework for identifying perinatal cortical morphology fingerprint. (a) a contrastive learning framework for data augmentation, feature extraction, and individual recognition; (b) the workflow of the green feature extraction module in (a), which extract feature representation from brain maps; (c) the workflow within the grey excitation module in (b), which calculates channel weights for attention-based feature learning and fusion.\u003c/p\u003e","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-4602847/v1/69589fc24d539ba251a4976b.png"},{"id":60597896,"identity":"643afafb-2851-43ad-90f8-e61be9d85481","added_by":"auto","created_at":"2024-07-18 15:51:52","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":167061,"visible":true,"origin":"","legend":"\u003cp\u003eA two-stage learning strategy for model parameters\u003c/p\u003e","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-4602847/v1/ce8a65e422b0787669563e27.png"},{"id":60597919,"identity":"2f8d0370-068b-40ac-a20b-a3f5547d552f","added_by":"auto","created_at":"2024-07-18 15:51:53","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":124727,"visible":true,"origin":"","legend":"\u003cp\u003eWeights of brain regions when using all three morphological features. The red bars in (a) are the brain regions with the top ten highest weights, and the green dashed line indicates the average weight of the whole brain. (b) Visualization of the weight of each brain region.\u003c/p\u003e","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-4602847/v1/ad8ead32b5720f8a84d5fa59.png"},{"id":79604863,"identity":"c708ee95-f3fc-415d-adf5-9ac86f20a378","added_by":"auto","created_at":"2025-03-31 16:07:55","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1987783,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-4602847/v1/6c7c3ffb-fa1e-4710-97c5-645dd70ea64f.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"\u003cp\u003eIdentification of morphological fingerprint in perinatal brains using quasi-conformal mapping and contrastive learning \u003c/p\u003e","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eSignificant progress has been achieved in understanding brain development and organization due to advances in neuroimaging and neuroscience (Giedd et al., 2010 [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]; Mills et al. 2014 [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]). There is growing awareness of the importance of differences in brain structure and function among individuals. These inter-individual variations may underlie differences in cognitive abilities, emotional processing, and susceptibility to neurological and psychiatric disorders (Dannlowski et al., 2012 [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]; Karama et al., 2014 [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]; Schmaal et al., 2017 [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]). Some researchers have explored the potential of brain function and structure for the usage of biometric identification (Bassi et al., 2018 [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]; Chen et al., 2018 [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]). While prior investigations have revealed that the adult brain exhibited stable structural and functional fingerprints for representing individual differences (Finn et al., 2015 [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e]; Menon et al., 2019 [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]), it remains unclear whether the fingerprint in the brain emerges as early as during the third trimester\u0026mdash;a critical period marked by the explosive growth of cortical anatomy and rapid establishment of structural and functional connectome.\u003c/p\u003e \u003cp\u003eMagnetic Resonance Imaging (MRI) is a non-invasive imaging technique known for its capacity to yield high-resolution images for the examination of both brain structure and function. This technology has been widely applied to study infant brains, such as the early developmental patterns of morphology, microstructure, fiber tracts, and brain connectomes (Liu et al., 2021 [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]; Liu et al., 2021 [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e]; Zheng et al., 2023 [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e] ; Zheng et al., 2023 [\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e]). Studies have characterized individual differences in human brains through anatomical and functional connectivity. For example, the white matter tractography in adult brains could serve as an effective fingerprint for the identification of individuals (Yeh et al., 2016 [\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]), and the identification accuracy of functional connectivity increases with age from childhood to adulthood (Vanderwal et al., 2021 [\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]). Furthermore, recent studies have attempted to investigate the fingerprint in perinatal brains and achieved recognition rates of 62.22% and 78% based on structural (Ciarrusta et al., 2022 [\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e]) and functional (Hu et al., 2022 [\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e]) connectivity, respectively. These results demonstrated the fact that individual uniqueness of brain connectome emerges during early brain development. Similar as brain connectivity, cortical morphology could also serve as a valuable fingerprint for individual recognition (Wachinger et al., 2015 [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]; Aloui et al., 2018 [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e]), which achieved remarkable performance that superior to the functional connectivity in differentiating adult individuals (Tian et al, 2021 [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]). In recent years, some studies have begun to explore cortical folding patterns for infant subject identification. For instance, Duan et al. successfully identified 1- and 2-year-old infants by using cortical folding information (i.e., curvature, convexity, and sulcus depth) of the corresponding neonate (Duan et al.,2019 [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e]; Duan et al.,2020 [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]). Nevertheless, it remains unclear whether the individual variations in human brain morphology already appear as early as the beginning of the third trimester.\u003c/p\u003e \u003cp\u003eDeep learning methods can achieve more advanced feature representations in brain images. Multiple deep learning models have been applied to resolve challenges in brain image analysis in recent years. For example, the convolutional neural network (CNN) has been utilized for the detection of brain lesion (Chen et al., 2020 [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e]) and white matter abnormalities (McKinley et al., 2019 [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e]), as well as for the diagnosis of brain disorders (Esmaeilzadeh et al., 2018 [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e]; Qureshi et al., 2019 [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e]). Convolutional networks based on the whole-brain cortical surface (Mostapha et al. 2018 [\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e]) have offered new perspectives for studying human brain MRI and cortical morphology. However, 3D-CNN methods are often unsuitable for small-sample datasets due to the large number of parameters and high computational demands. On the other hand, 2D-CNN methods have lower computational requirements and can be more easily embedded into a mature model architecture. T. W. Meng et al. have introduced Teichm\u0026uuml;ller Extremal Mapping of POint clouds (TEMPO), a quasi-conformal mapping method for conformally mapping a simply-connected open triangle mesh to a 2D rectangle space [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e]. The TEMPO method can effectively preserve conformality, reducing the loss of local features and geometric structures in the mapping of the original point cloud. Leveraging deep learning methods enables the automatic acquisition of more high-dimensional features of cortical morphology through multilevel nonlinear transformations, thereby simplifying the model\u0026rsquo;s principles and procedures via end-to-end learning. Therefore, we utilized quasi-conformal mapping to project the 3D brain mesh onto a 2D plane and employed 2D-CNN to extract individual cortical morphological feature representations.\u003c/p\u003e \u003cp\u003eThe present study aims to validate the existence of morphological fingerprints in perinatal brains. Each hemispheric surface of an individual subject was inflated to a sphere and was then projected to a 2D plane through TEMPO. We propose a contrastive learning framework based on the pretrained ResNet18 encoder to recognize an individual at term-equivalent age by using his or her brain MRI acquired at birth. The attention mechanism is incorporated to fuse features from different partitions generated from the 3D-to-2D mapping. The effectiveness of each morphological feature is assessed to demonstrate their contributions to the individual fingerprint in perinatal brains.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e"},{"header":"2. Materials and Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003e2.1 Dataset\u003c/h2\u003e \u003cp\u003e\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eImaging data used in this study was obtained from the Developing Human Connectome Project (dHCP-release 3, \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://biomedia.github.io/dHCP-release-notes/acquire.html\u003c/span\u003e\u003cspan address=\"https://biomedia.github.io/dHCP-release-notes/acquire.html\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e).\u003c/p\u003e\u003cp\u003e The dHCP study was conducted in accordance with the Declaration of Helsinki and was approved by the United Kingdom Health Research Ethics Authority (reference no. 14/LO/1169). Written informed consent was obtained from all participating families before the scan.\u003c/p\u003e\u003cp\u003eAll participants received MRI scans performed using a 3T Philips Achieva system [\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e]. T2-weighted (T2w) multi-slice fast spin-echo images were obtained in both sagittal and axial slice stacks. These images had an in-plane resolution of 0.8x0.8mm\u0026sup2; and 1.6mm slices, with an overlap of 0.8mm. The imaging parameters were as follows: repetition time (TR)\u0026thinsp;=\u0026thinsp;12000ms, echo time (TE)\u0026thinsp;=\u0026thinsp;156ms, SENSE factor of 2.11 for axial and 2.60 for sagittal slices. Additionally, a 3D MPRAGE scan was performed with 0.8mm isotropic resolution. The dataset included 783 infants in total, comprising 682 infants who underwent MRI scanning one time and 101 infants who were scanned twice (scanned at approximately 1\u0026ndash;2 weeks after birth and term-equivalent age, respectively) or more. We selected 400 sets of once sampling data and 62 pairs of twice sampling data for the experiment. Participants who have low-quality imaging data or focal brain abnormality (radiology score\u0026thinsp;\u0026gt;\u0026thinsp;3, an evaluation by the dHCP) were discarded. Detailed demographic information of the selected infants are given in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e.\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eDemographic information of the infants included in this study.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"6\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGroups\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMedian birth age (weeks)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eBirth age range (weeks)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eMedian scan age (weeks)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eScan age range (weeks)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eMale/ Female\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCross-sectional\u003c/p\u003e \u003cp\u003ecohort\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e39.86\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e[24.71\u0026ndash;43.57]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e41.14\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e[26.86\u0026ndash;45.14]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e215/ 185\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLongitudinal cohort (1st scan)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e31.86\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e[23.86\u0026ndash;40.14]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e34.71\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e[26.71\u0026ndash;42.71]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e32/29\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLongitudinal cohort (2nd scan)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e31.86\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e[23.86\u0026ndash;40.14]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e41.29\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e[38.43\u0026ndash;44.86]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e32/29\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003e2.2 Dataset Processing\u003c/h2\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eImage preprocessing followed the pipeline proposed by the dHCP [\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e]. In summary, the process first involved bias correction and brain extraction of the motion-corrected T2-weighted image. This was followed by segmenting the brain into different tissue types using Draw-EM algorithm. White-matter mesh was then extracted and expanded to fit the pial surface. The cortical thickness was estimated based on the Euclidean distance between the white and pial surface. We generated the inflated surfaces from the white surface, which were then projected to a sphere for surface registration. The mean surface curvature and sulcal depth (mean surface convexity/concavity) were respectively estimated from the white surface and from inflation.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec5\" class=\"Section2\"\u003e \u003ch2\u003e2.3 Teichm\u0026uuml;ller Mapping\u003c/h2\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eDespite the promising spherical convolutional neural network (SCNN) models on (Cohen et al., 2018 [\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e]; C. Esteves et al., 2018 [\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e]; C. Jiang et al., 2019 [\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e]), the SCNNs often suffer from substantial computational complexity, e.g., extensive parameters, which make them difficult to converge on small sample datasets. Therefore, we projected the 3-dimensional spherical mesh of the brain surface to 2-dimensional and employed the convolutional neural network (CNN) model for analysis.\u003c/p\u003e \u003cp\u003eTo achieve this goal, a quasi-conformal mapping method (Teichm\u0026uuml;ller Mapping) was employed, which is a geometric transformation method based on complex variational functions. Conformal mapping facilitates the mapping of one space to another while preserving local angles between intersecting curves or surfaces, thereby ensuring the preservation of shapes and angles within a small region of the original space. In general, its formula representation is as follows:\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Equa\" class=\"Equation\"\u003e \u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equa\" name=\"EquationSource\"\u003e\n$$f\\text{*}d{s}_{N}^{2}=\\lambda d{s}_{M}^{2}$$\u003c/div\u003e \u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003ewhere M and N are two Riemann surfaces, f is the mapping function: M\u0026rarr;N, and λ is a positive scalar function.\u003c/p\u003e \u003cp\u003eA generalization of conformal mapping is quasi-conformal mapping, which allows for a certain degree of angle and shape distortion to adapt to more complex data structures and transformations. Teichm\u0026uuml;ller Mapping is such a type of quasi-conformal mapping that can induce uniform conformality distortions in the target point cloud, thereby preserving stable relative positions and local shapes. The general formula for a Teichm\u0026uuml;ller mapping (T-map) can be expressed as follows: let M and N be two Riemann surfaces, and \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(f:M\\to N\\)\u003c/span\u003e\u003c/span\u003e be a quasi-conformal mapping. If it is associated with a quadratic differential \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(q=\\varphi d{z}^{2}\\)\u003c/span\u003e\u003c/span\u003e, where \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\varphi :M\\to \\mathbb{C}\\)\u003c/span\u003e\u003c/span\u003e is a holomorphic function, and its associated Beltrami coefficient is of the form \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\mu \\left(f\\right)=k\\frac{\\varphi }{\\left|\\varphi \\right|}\\)\u003c/span\u003e\u003c/span\u003e, where k\u0026lt;1 is a constant, and the quadratic differential q is non-zero and satisfies \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\left|\\right|q{\\left|\\right|}_{1}={\\int }_{{S}^{1}}\\left|\\varphi \\right|\u0026lt;{\\infty }\\)\u003c/span\u003e\u003c/span\u003e, then f is called a Teichm\u0026uuml;ller mapping (T-map) associated with the quadratic differential \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(q\\)\u003c/span\u003e\u003c/span\u003e.\u003c/p\u003e \u003cp\u003eIn the present study, we used the TEMPO method for projection. Specifically, we adopt an approach of segmenting the hemispherical plane along the x\u0026thinsp;=\u0026thinsp;0 plane in the coordinate system and subsequently applying the TEMPO method separately to the two hemispheres for mapping. Then, we interpolated the mapped mesh to obtain a two-dimensional matrix. Although this approach sacrificed the continuity of the data around the segmentation curve, it significantly reduced the area distortion from mapping deformation and preserved the overall continuity of other areas, offering a more desirable outcome for our study. The projected meshes on 2D plane are shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003e2.4 Data Augmentation\u003c/h2\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eData augmentation is a commonly employed technique to address the issue of insufficient data and lack of data diversity. Its primary objective is to expand the dataset by applying diverse transformations and processing to the original data. In the context of images, various augmentation methods have been established, including random cropping, rotation, flipping, random noise, Gaussian blur, and color transformations such as brightness, contrast, and saturation adjustments (Cubuk et al., 2018 [\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e]; Shorten et al., 2019 [\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e]; T. Chen et al., 2020 [\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e]). We used data augmentation to expand single sampling data into sample pairs, which can be used for model pre-training. We employed rotation, random noise, and Gaussian blur methods as shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec7\" class=\"Section2\"\u003e \u003ch2\u003e2.5 Model Architecture\u003c/h2\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eAs illustrated in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003ea, we present a contrastive learning-based feature extraction framework. We aligned the 3D grids of the left and right brain hemispheres onto spheres, and then partitioned each sphere into two hemispheres. Subsequently, we employed the TEMPO method to map the spherical grids onto planar grids and compute four 3\u0026times;224\u0026times;224 matrices through interpolation. These cortical morphological matrices were then fed into the feature extraction module to extract cortical morphological feature fingerprints of individual subjects (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eb). These fingerprints (vectors of length 512) were ultimately utilized for calculating pairwise similarities for individual identification. Notably, in the feature extraction process, we employed a channel-wise attention mechanism-based excitation module for fusing features from different brain partitions and weight allocation to feature map channels, as shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003ec.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003e2.6 Contrastive Learning\u003c/h2\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eContrastive learning is a self-supervised method for training models to learn meaningful data representations. It works by comparing the similarity between pairs of data samples. In essence, it creates pairs of positive and negative samples and uses a loss function to make the feature representations of positive pairs more similar while making those of negative pairs more dissimilar. A notable framework for contrastive learning is SimCLR (short for Simple Contrastive Learning of Representations) [\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e], which used the NT-Xent loss function to maximize the similarity of positive pairs and minimize the similarity of negative pairs in the feature space. The NT-Xent loss function is defined as:\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Equb\" class=\"Equation\"\u003e \u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equb\" name=\"EquationSource\"\u003e\n$$L=-\\frac{1}{N}\\sum _{i=1}^{N}\\left(\\text{l}\\text{o}\\text{g}\\left(\\frac{\\text{e}\\text{x}\\text{p}\\left(-\\parallel {z}_{i}-{z}_{i}^{+}\\parallel /\\tau \\right)}{\\sum _{j=1,j\\ne i}^{N}\\text{e}\\text{x}\\text{p}\\left(-\\parallel {z}_{i}-{z}_{j}^{+}\\parallel /\\tau \\right)}\\right)\\right)$$\u003c/div\u003e \u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003ewhere N is the batch size, \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\parallel {z}_{i}-{z}_{i}^{+}\\parallel\\)\u003c/span\u003e\u003c/span\u003eis the Euclidean distance between the \u003cem\u003ei\u003c/em\u003e-th sample and its positive sample, \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\parallel {z}_{i}-{z}_{j}^{+}\\parallel\\)\u003c/span\u003e\u003c/span\u003e is the Euclidean distance between the \u003cem\u003ei\u003c/em\u003e-th sample and the \u003cem\u003ej\u003c/em\u003e-th sample\u0026rsquo;s corresponding positive sample. τ denotes a temperature parameter.\u003c/p\u003e \u003cp\u003eIn this study, we used the SimCLR as the main contrastive learning framework to extract morphological feature representations of individual differences in the perinatal cerebral cortex. We constructed positive sample pairs by using multiple data augmentation methods on single sampling data and used the other samples in the same batch to construct negative pairs. Specifically, we adopted a contrastive loss similar to the classical Siamese network [\u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e]. We constructed multiple negative sample pairs while adopting a non-softmax computation of the loss function. The cross-entropy loss that was separate for the positive and negative sample pairs was calculated and summed up. The loss function was defined as follows:\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Equc\" class=\"Equation\"\u003e \u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equc\" name=\"EquationSource\"\u003e\n$$L=\\sum _{i=1}^{N}\\left(\\left(1-{y}_{i}\\right)\\text{*}{\\text{dist}}_{i}^{2}+y\\text{*}{\\text{clamp}}_{\\text{min}}{\\left(m-{\\text{dist}}_{i},0\\right)}^{2}\\right)$$\u003c/div\u003e \u003c/div\u003e \u003cdiv id=\"Equd\" class=\"Equation\"\u003e \u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equd\" name=\"EquationSource\"\u003e\n$${\\text{dist}}_{i}=\\parallel {x}_{1i}-{x}_{2i}\\parallel ,{\\text{clamp}}_{\\text{min}}\\left(a,b\\right)=\\left\\{\\begin{array}{ll}a,\u0026amp; \\text{if }a\\ge b\\\\ b,\u0026amp; \\text{if }a\u0026lt;b\\end{array}\\right.$$\u003c/div\u003e \u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003ewhere N is the batch size, \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({y}_{i}\\)\u003c/span\u003e\u003c/span\u003e is the label indicating whether the \u003cem\u003ei\u003c/em\u003e-th sample pair was a positive or negative sample, \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({x}_{1i}\\)\u003c/span\u003e\u003c/span\u003eand \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\({x}_{2i}\\)\u003c/span\u003e\u003c/span\u003erepresent the two samples of the i-th sample pair respectively, and m was an artificially set margin.\u003c/p\u003e \u003cp\u003eIn this framework, we used the ResNet18 backbone (He et al., 2016 [\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e]) as the brain map encoder, as shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e-(b). For each morphological feature, four sets of projection maps (medial and lateral projections of the left and right brain, respectively) per sample were fed into the parameter-sharing ResNet18 to extract four feature maps.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec9\" class=\"Section2\"\u003e \u003ch2\u003e2.7 Excitation Module\u003c/h2\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eThe excitation module aimed to enhance the overall representation capability of the feature vectors to the morphological features of the samples through the channel attention mechanism, as shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003ec. It achieved the fusion of features from four different partitions. Similar to the SE-Nets (Hu et al., 2018 [\u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e]), the channel weights were learned by two fully connected layers that mapped the feature vector to a channel attention vector. This channel attention vector was applied to the input feature map to weigh the channels. However, different from the previous literature, our excitation module optimized the assignment of the weight to three channels of morphological features and four channels (partitions) of projection maps simultaneously, which was arranged before the final layer of the neckbone network. Adding a hyperparameter weight_scale was intended to restore the weights to a distribution with a mean of 1 or approximately 1 to maintain the consistency of the data scale. The feature vector of the sample was obtained by pooling and a simplified projection through a fully connected layer. In addition to the excitation method, we employed other two fusion techniques: a decision-level fusion approach based on a voting mechanism and a feature-level fusion method using a two-layer Multilayer Perceptron (MLP). The former averaged similarity matrices calculated on four partitions and then made identification decisions based on the averaged similarity matrix. The latter utilized a two-layer MLP to map the input features.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec10\" class=\"Section2\"\u003e \u003ch2\u003e2.8 Training Procedure and Evaluation Metrics\u003c/h2\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eFor the training process, we initially excluded the excitation module and employed the original structure during the training of the ResNet18 backbone. The network parameters were trained by using primarily augmented sample pairs for pre-training and several two-shot data for fine-tuning, as illustrated in the stage1 of Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e. After that, we froze the ResNet18 and updated the parameters in the excitation module and the FC layer (stage2 of Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e). Finally, the trained feature extraction module was applied to the twice scanned samples. We calculated and compared the similarity of the feature vectors between two time points of the same subject (self-self similarity) and the feature similarity between a neonate at birth and other infants scanned at term-equivalent age (self-other similarity).\u003c/p\u003e \u003cp\u003eThe Euclidean distance was used to measure the similarity between individual brains (Eickhoff et al.,2005 [\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e]; Al-Saffar et al., 2020 [\u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e]). We utilized the Top 1 accuracy (where self-similarity surpasses all self-other similarities) and Top 5 accuracy (where self-similarity ranks among the top five similarities between itself and all other samples) as metrics to assess the effectiveness of the model.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e"},{"header":"3. Results","content":"\u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003e3.1. Experimental Setup\u003c/h2\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eWe used the ResNet18 backbone network and its pre-trained network parameters provided by Pytorch on the ImageNet dataset for initialization. To train the ResNet18 network, we augmented all single-sampled data to create positive and negative sample pairs. These pairs were then used for the initial training of ResNet18. Subsequently, we performed fine-tuning of ResNet18 and trained the feature fusion module. The training process employed a learning rate of 5e-4 with a momentum of 0.9 and a weight decay of 5e-5. Each training session comprised 8 epochs and the SGD optimizer was used. For all experiments, we conducted triple-fold cross-validation using twice sampled data (i.e., the real sample pairs). In each fold, we used 40 and 20 real sample pairs for training and testing, respectively. We repeated the experiments for 30 rounds to ensure robustness and obtained weighted accuracies for identification.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003e3.2. Experiment Results\u003c/h2\u003e \u003cdiv id=\"Sec14\" class=\"Section3\"\u003e \u003ch2\u003e3.2.1. Individual Recognition\u003c/h2\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eOur model achieved a notable Top 1 accuracy of 85.90% and a Top 5 accuracy of 92.20%. These results suggested that the self-similarity of the prenatal brains between birth and term-equivalent ages tended to be higher than self-other similarities. We also compared the accuracies derived from different fusion strategies for fusing features from left lateral, left medial, right lateral and right medial brain. (e.g., excitation, voting and MLP). The excitation method outperformed other fusion methods. The voting strategy that integrated judgments from different partitions also demonstrated high Top 1 accuracy of 85.25% and Top 5 accuracy of 91.75%. The method using a MLP for feature fusion achieved the Top 1 accuracy of 77.78% and Top 5 accuracy of 87.39%. Overall, the excitation method exhibited superior performance. The detailed model performance is shown in Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eModel performance comparison between different fusion methods and different feature channels selected.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"6\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eFusion methods\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eAccuracy\u003c/p\u003e \u003cp\u003e(%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"4\" nameend=\"c6\" namest=\"c3\"\u003e \u003cp\u003eMorphological Features\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eCortical thickness\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eMean curvature\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eSulcal depth\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eAll\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eExcitation\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTOP 1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e25.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e82.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e73.34\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e\u003cb\u003e85.90\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTOP 5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e60.35\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e87.65\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e85.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e\u003cb\u003e92.20\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eMLP\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTOP 1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e29.31\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e75.08\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e71.45\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e77.78\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTOP 5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e49.81\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e85.08\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e82.13\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e87.39\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eDecision Level\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTOP 1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e37.15\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e81.67\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e73.65\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e85.25\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTOP 5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e61.17\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e87.67\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e84.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e91.75\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eConcatenation\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTOP 1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e23.65\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e81.65\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e73.34\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e16.70\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTOP 5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e59.35\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e87.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e83.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e55.70\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec15\" class=\"Section3\"\u003e \u003ch2\u003e3.2.2. Contributions of Morphological Features and Brain Regions\u003c/h2\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eTo explore the contributions of each morphological feature to the recognition task, we conducted single-channel comparison experiments for the three morphological features: curvature, thickness, and sulcus. Mean curvature achieved the highest Top 1 and Top 5 accuracies of 82.00% and 87.65%, respectively, which outperformed the other two features. Cortical thickness did not exhibit any discriminative power.\u003c/p\u003e \u003cp\u003eWe then explored the contributions of different brain regions in the recognition task. We utilized the weights derived from the excitation module as the contribution weights of brain regions shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003ea, and the weights of brain regions were mapped back to the cortical surface (as illustrated in Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eb. The parietal and occipital cortices of the right hemisphere and the antero-medial temporal cortices of the right hemisphere showed higher weights relative to other regions, suggesting these regions may have evident morphological differences among individuals.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003c/div\u003e"},{"header":"4. Discussion","content":"\u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eIn this study, we proved the existence of morphological fingerprints in neonatal cerebral cortex through a deep learning model. We achieved a remarkable Top 1 accuracy of 85.90% in the individual recognition. This suggested that certain cortical morphological fingerprints are already formed as early as the beginning of the third trimester and maintained stable during the perinatal period, which can serve as an effective fingerprint for recognizing individual neonate.\u003c/p\u003e \u003cp\u003eTraining the ResNet18 encoder with a contrastive learning approach significantly improved the recognition accuracy, demonstrating the effectiveness of the framework in learning cortical morphological features. Additionally, the incorporation of the excitation module based on attention mechanisms also greatly enhanced the model performance, outperforming both the comparative voting and MLP methods. We speculated that the improvement might be attributed to the excitation module enabling the model to prioritize brain regions with significant individual variability, reducing the impact of other regions on individual recognition and thus enhancing identification accuracy (Bodapati et al., 2021 [\u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e41\u003c/span\u003e]). Moreover, the excitation module enabled the direct assessment of the contribution of different brain regions, providing a convenient way to analyze the recognition contribution rates of different brain areas.\u003c/p\u003e \u003cp\u003eOur findings indicated that both cortical curvature and sulcal depth exhibited individual variations, and the former achieved the best recognition rate, emphasizing the crucial role of the cortical folding morphology in characterizing individual differences at early developmental stage. On the other hand, cortical thickness did not exhibit discriminative power to individual recognition. Previous studies have suggested that the individual variability of cortical folding patterns has been established at term age (Duan et al., 2019 [\u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e42\u003c/span\u003e]), while the cortical thickness showed relative longer maturation period. Specifically, the cortical thickness typically undergoes rapid growth shortly after birth, peaking around 14 months after birth and subsequently decreases (Wang et al., 2019 [\u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e43\u003c/span\u003e]). Therefore, we hypothesized that the earlier maturation of folding patterns imparts greater individual variability to cortical morphology, leading to higher accuracy in individual recognition. Additionally, sensitivity to noise may be another potential factor leading to lower recognition efficiency of thickness features. Moreover, we observed that primary cortex (e.g., somatosensory and visual) carried higher attention weights, indicating greater inter-individual differences in these regions. Given that the primary regions experienced more pronounced development than the high-order cortex in the second trimester (Gilmore et al., 2018 [\u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e44\u003c/span\u003e]; Duan et al., 2019 [\u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e42\u003c/span\u003e]), we speculated that the high attention weights assigned to the primary cortex may be attributed to their stable morphology maturity throughout the third trimester.\u003c/p\u003e \u003cp\u003eOur model may hold potential in clinical applications. While previous studies revealed differential patterns of development across brain measures or regions, very few examine heterogeneity in the developmental patterns between persons. Examining within-subject change over time is necessary to depict typical development of an individual as well as identifying possible atypical alterations [\u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e45\u003c/span\u003e]. For example, a significant dissimilarity observed in an individual brain compared to its earlier stage may indicate abnormal development. Furthermore, our model largely reduced training parameters by projecting cortical spherical surface to two-dimensional space, which would be suitable for the computation on small dataset, such as infant data. In addition, by accounting for the individual variability in alteration patterns associated with brain disorders, our approach has the potential to correct inter-subject discrepancies in brain morphology and enable clinical models to focus on disease-specific changes.\u003c/p\u003e \u003cp\u003eOur work still has some limitations. First, the longitudinal dataset was relatively small. The generalizability of this model and the reproducibility of our findings should be examined on a larger independent dataset. Second, the conformal transformation may also lead to area distortion during projection, especially near the split curve. Such distortion may potentially influence the recognition accuracy. Some novel methods for addressing these challenges should be developed in the future. In addition, our study is still at experimental stage, it is necessary to validate the results from both methodological and clinical perspectives.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e"},{"header":"5. Conclusions","content":"\u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eIn conclusion, we introduce a deep learning framework that maps the three-dimensional spherical surface onto a two-dimensional image for computational simplification, facilitating the investigation of individual differences in neonatal brain structures. Our study presents the initial evidence of morphological fingerprints emerging during the perinatal period and highlights the potential for identifying atypical brain development of an individual.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e \u003ch2\u003eConflicts of Interest\u003c/h2\u003e \u003cp\u003eThe authors declare no conflict of interest.\u003c/p\u003e \u003c/p\u003e\u003ch2\u003eFunding\u003c/h2\u003e \u003cp\u003eThis work was funded by the National Natural Science Foundation of China (62202212) and the STI-2030 major project (2021ZD0200800).\u003c/p\u003e\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eConceptualization, B.W., Y.S., D.Z. and W.Z.; methodology, W.Z., and Y.W.; validation, B.W. and W.Z.; formal analysis, B.W. and Y.W.; data curation, Y.W.; writing---original draft preparation, B.W. and W.Z.; writing---review and editing, W.Z, D.Z., and Y.S.; supervision, W.Z, and Y.S.; funding acquisition, W.Z.. All authors have reviewed and agreed to the published version of the manuscript.\u003c/p\u003e\u003ch2\u003eData Availability\u003c/h2\u003e\u003cp\u003eWe thank all the contributors to the dHCP dataset that support our study. This dataset is openly available at http://www.developingconnectome.org.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eGiedd, J. N., \u0026amp; Rapoport, J. L. (2010). Structural MRI of pediatric brain development: what have we learned and where are we going? \u003cem\u003eNeuron 67(5)\u003c/em\u003e, 728\u0026ndash;734.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMills, K. L., Goddings, A. L., Clasen, L. S., Giedd, J. N., \u0026amp; Blakemore, S. J. (2014). The developmental mismatch in structural brain maturation during adolescence. \u003cem\u003eDevelopmental neuroscience\u003c/em\u003e, \u003cem\u003e36\u003c/em\u003e(3\u0026ndash;4), 147\u0026ndash;160.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDannlowski, U., Stuhrmann, A., Beutelmann, V., Zwanzger, P., Lenzen, T., Grotegerd, D., Domschke, K., Hohoff, C., Ohrmann, P., Bauer, J., \u0026amp; Lindner, C. (2014). Limbic scars: long-term consequences of childhood maltreatment revealed by functional and structural magnetic resonance imaging. \u003cem\u003eBiological psychiatry\u003c/em\u003e, \u003cem\u003e71\u003c/em\u003e(4), 286\u0026ndash;293.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKarama, S., Bastin, M. E., Murray, C., Royle, N. A., Penke, L., Mu\u0026ntilde;oz Maniega, S., Gow, A. J., Corley, J., Vald\u0026eacute;s Hern\u0026aacute;ndez, M., Lewis, J. D., \u0026amp; Rousseau (2014). M.\u0026Eacute;. Childhood cognitive ability accounts for associations between cognitive ability and brain cortical thickness in old age. \u003cem\u003eMolecular psychiatry\u003c/em\u003e, \u003cem\u003e19\u003c/em\u003e(5), 555\u0026ndash;559.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSchmaal, L., Hibar, D. P., S\u0026auml;mann, P. G., Hall, G. B., Baune, B. T., Jahanshad, N., Cheung, J. W., van Erp, T. G., Bos, D., Ikram, M. A., \u0026amp; Vernooij, M. W. (2017). Cortical abnormalities in adults and adolescents with major depression based on brain scans from 20 cohorts worldwide in the ENIGMA Major Depressive Disorder Working Group. \u003cem\u003eMolecular psychiatry\u003c/em\u003e, \u003cem\u003e22\u003c/em\u003e(6), 900\u0026ndash;909.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBassi, M., \u0026amp; Triverbi, P. Human biometric identification through brain print. In 2018 Second International Conference on Electronics, Communication and Aerospace Technology (ICECA) 2018, 1514\u0026ndash;1518.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen, Shiyang, Xiaoping, \u0026amp; Hu (2018). Individual identification using the functional brain fingerprint detected by the recurrent neural network. \u003cem\u003eBrain connectivity\u003c/em\u003e, \u003cem\u003e8\u003c/em\u003e(4), 197\u0026ndash;204.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFinn, E. S., Shen, X., Scheinost, D., Rosenberg, M. D., Huang, J., \u0026amp; Chun, M. M. (2015). Xenophon Papademetris, and R. Todd Constable. Functional connectome fingerprinting: identifying individuals using patterns of brain connectivity. \u003cem\u003eNature neuroscience\u003c/em\u003e, \u003cem\u003e18\u003c/em\u003e(11), 1664\u0026ndash;1671.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMenon, S. S., \u0026amp; Krishnamurthy, K. (2019). A comparison of static and dynamic functional connectivities for identifying subjects and biological sex using intrinsic individual brain connectivity. \u003cem\u003eScientific reports\u003c/em\u003e, \u003cem\u003e9\u003c/em\u003e(1), 5729.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiu, T., Gao, F., Zheng, W., You, Y., Zhao, Z., Lv, Y., Chen, W., Zhang, H., Ji, C., \u0026amp; Wu, D. (2021). Diffusion MRI of the infant brain reveals unique asymmetry patterns during the first-half-year of development. \u003cem\u003eNeuroimage\u003c/em\u003e, \u003cem\u003e242\u003c/em\u003e, 118465.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiu, T., Zhang, H., You, Y., Zheng, W., Zhao, Z., Liu, T., Su, X., Tian, F., Zhang, Y., \u0026amp; Wu, D. (2021). Brain developmental differences between preterm-born twins and singletons: a multi-modal MRI Study. \u003cem\u003eJpr\u003c/em\u003e, \u003cem\u003e8\u003c/em\u003e, 276\u0026ndash;285.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZheng, W., Wang, X., Liu, T., Hu, B., \u0026amp; Wu, D. (2023). Preterm-birth alters the development of nodal clustering and neural connection pattern in brain structural network at term‐equivalent age. \u003cem\u003eHuman Brain Mapping\u003c/em\u003e, \u003cem\u003e44\u003c/em\u003e(16), 5372\u0026ndash;5386.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZheng, W., Zhao, L., Zhao, Z., Liu, T., Hu, B., \u0026amp; Wu, D. (2023). Spatiotemporal Developmental Gradient of Thalamic Morphology, Microstructure, and Connectivity from the Third Trimester to Early Infancy. \u003cem\u003eJournal of Neuroscience\u003c/em\u003e, \u003cem\u003e43\u003c/em\u003e(4), 559\u0026ndash;570.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYeh, F. C., Vettel, J. M., Singh, A., Poczos, B., Grafton, S. T., Kirk, I., Erickson, Wen-Yih, I., Tseng, \u0026amp; Timothy, D. (2016). Verstynen. Quantifying differences and similarities in whole-brain white matter architecture using local connectome fingerprints. \u003cem\u003ePLoS computational biology 12(11)\u003c/em\u003e, e1005203.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eVanderwal, T., Eilbott, J., Kelly, C., Frew, S. R., Todd, S., Woodward, M. P., \u0026amp; Milham (2021). and F. Xavier Castellanos. Stability and similarity of the pediatric connectome as developmental measures. \u003cem\u003eNeuroImage 226\u003c/em\u003e, 117537.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCiarrusta, J., Christiaens, D., Fitzgibbon, S. P., Dimitrova, R., Hutter, J., Hughes, E., Duff, E., et al. (2022). The developing brain structural and functional connectome fingerprint. \u003cem\u003eDevelopmental Cognitive Neuroscience\u003c/em\u003e, \u003cem\u003e55\u003c/em\u003e, 101117.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHu, D., Wang, F., Zhang, H., Wu, Z., Zhou, Z., Li, G., Wang, L., Lin, W., Li, G., \u0026amp; UNC/UMN Baby Connectome Project Consortium. (2022). Existence of functional connectome fingerprint during infancy and its stability over months. \u003cem\u003eJournal of Neuroscience\u003c/em\u003e, \u003cem\u003e42\u003c/em\u003e(3), 377\u0026ndash;389.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWachinger, C., Golland, P., Kremen, W., \u0026amp; Fischl, B. (2015). Martin Reuter, and Alzheimer\u0026rsquo;s Disease Neuroimaging Initiative. BrainPrint: A discriminative characterization of brain morphology. \u003cem\u003eNeuroimage\u003c/em\u003e, \u003cem\u003e109\u003c/em\u003e, 232\u0026ndash;248.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAloui, K., Nait-Ali, A., \u0026amp; Naceur, M. S. (2018). Using brain prints as new biometric feature for human recognition. Pattern \u003cem\u003eRecognition Letters\u003c/em\u003e 2018, \u003cem\u003e113\u003c/em\u003e, 38\u0026ndash;45.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTian, Y., Yeo, B. T., Cropley, V., \u0026amp; Zalesky, A. (2021). High-resolution connectomic fingerprints: Mapping neural identity and behavior. \u003cem\u003eNeuroimage\u003c/em\u003e, \u003cem\u003e229\u003c/em\u003e, 117695.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDuan, D., Xia, S., Wu, Z., Wang, F., Wang, L., Lin, W., \u0026amp; Gilmore, J. H. (2019). Dinggang Shen, and Gang Li. Cortical Foldingprints for Infant Identification. In 2019 IEEE 16th International Symposium on Biomedical ImagingISBI 396\u0026ndash;399.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDuan, D., Xia, S., Rekik, I., Wu, Z., Wang, L., Lin, W., \u0026amp; Gilmore, J. H. (2020). Dinggang Shen, and Gang Li. Individual identification and individual variability analysis based on cortical folding features in developing infant singletons and twins. \u003cem\u003eHuman brain mapping\u003c/em\u003e, \u003cem\u003e41\u003c/em\u003e(8), 1985\u0026ndash;2003.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen, H., Qin, Z., Ding, Y., Tian, L., \u0026amp; Qin, Z. (2020). Brain tumor segmentation with deep convolutional symmetric neural network. \u003cem\u003eNeurocomputing\u003c/em\u003e, \u003cem\u003e392\u003c/em\u003e, 305\u0026ndash;313.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMcKinley, R., Wepfer, R., Aschwanden, F., Grunder, L., Muri, R., Rummel, C., Verma, R., et al. (2019). Simultaneous lesion and neuroanatomy segmentation in multiple sclerosis using deep neural networks. \u003cem\u003eScientific reports\u003c/em\u003e, \u003cem\u003e11\u003c/em\u003e(1), 1087.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eEsmaeilzadeh, S., Yang, Y., \u0026amp; Adeli, E. (2018). End-to-end parkinson disease diagnosis using brain mr-images by 3d-cnn. \u003cem\u003earXiv preprint arXiv\u003c/em\u003e:180605233.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eQureshi, M. N., Iqbal, J., Oh, \u0026amp; Lee, B. (2019). 3D-CNN based discrimination of schizophrenia using resting-state fMRI. \u003cem\u003eArtificial intelligence in medicine\u003c/em\u003e, \u003cem\u003e98\u003c/em\u003e, 10\u0026ndash;17.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMostapha, M., Kim, S. H., Wu, G., Zsembik, L., Pizer, S., \u0026amp; Styner, M. (2018). ISBI. Non-euclidean, convolutional learning on cortical brain surfaces. In 2018 IEEE 15th International Symposium on Biomedical Imaging 527\u0026ndash;530.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMeng, T., Wei, G., \u0026amp; Choi, P. T. (2016). Lok Ming Lui. Tempo: feature-endowed Teichmuller extremal mappings of point clouds. \u003cem\u003eSIAM Journal on Imaging Sciences\u003c/em\u003e, \u003cem\u003e9\u003c/em\u003e(4), 1922\u0026ndash;1962.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHughes, E. J., Winchman, T., Padormo, F., Teixeira, R., Wurie, J., Sharma, M., Fox, M., Hutter, J., Cordero-Grande, L., Price, A. N., \u0026amp; Allsop, J. (2017). A dedicated neonatal brain imaging system. \u003cem\u003eMagnetic resonance in medicine\u003c/em\u003e, \u003cem\u003e78\u003c/em\u003e(2), 794\u0026ndash;804.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMakropoulos, A., Robinson, E. C., Schuh, A., Wright, R., Fitzgibbon, S., Bozek, J., Counsell, S. J., Steinweg, J., Vecchiato, K., Passerat-Palmbach, J., \u0026amp; Lenz, G. (2018). The developing human connectome project: A minimal processing pipeline for neonatal cortical surface reconstruction. \u003cem\u003eNeuroimage\u003c/em\u003e, \u003cem\u003e173\u003c/em\u003e, 88\u0026ndash;112.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCohen, T. S., \u0026amp; Geiger, M. (2018). Jonas K\u0026ouml;hler, and Max Welling. Spherical cnns. arXiv preprint arXiv:1801.10130.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eEsteves, C., Allen-Blanchette, C., Makadia, A., \u0026amp; Daniilidis, K. (2018). Learning so (3) equivariant representations with spherical cnns. In Proceedings of the European Conference on Computer Vision (ECCV). 52\u0026ndash;68.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChiyu Max Jiang, Huang, J., \u0026amp; Kashinath, K. (2019). Prabhat, Philip Marcus, and Matthias Nie\u0026szlig;ner. Spherical CNNs on Unstructured Grids. In ICLR (Poster).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCubuk, E. D., Zoph, B., Mane, D., Vasudevan, V., \u0026amp; Quoc, V. (2018). Le. Autoaugment: Learning augmentation policies from data. \u003cem\u003earXiv preprint arXiv\u003c/em\u003e:180509501.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eShorten, C., \u0026amp; Taghi, M. (2019). Khoshgoftaar. A survey on image data augmentation for deep learning. \u003cem\u003eJournal of big data\u003c/em\u003e, \u003cem\u003e6\u003c/em\u003e(1), 1\u0026ndash;48.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen, T., Kornblith, S., Norouzi, M., \u0026amp; Hinton, G. A simple framework for contrastive learning of visual representations. In International conference on machine learning 2020, 1597\u0026ndash;1607.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChopra, S., Hadsell, R., Yann, \u0026amp; LeCun (2005). Learning a similarity metric discriminatively, with application to face verification. In 2005 IEEE computer society conference on computer vision and pattern recognition (CVPR\u0026rsquo;05). 539\u0026ndash;546.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHe, K., Zhang, X., Ren, S., \u0026amp; Sun, J. (2016). Deep residual learning for image recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition. 770\u0026ndash;778.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eEickhoff, S., Walters, N. B., Schleicher, A., Kril, J., Egan, G. F., Zilles, K., Watson, J. D. G., \u0026amp; Amunts, K. (2005). High-resolution MRI reflects myeloarchitecture and cytoarchitecture of human cerebral cortex. \u003cem\u003eHuman brain mapping\u003c/em\u003e, \u003cem\u003e24\u003c/em\u003e(3), 206\u0026ndash;215.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAl-Saffar, Z. A., T\u0026uuml;lay, \u0026amp; Yildirim (2020). A novel approach to improving brain image classification using mutual information-accelerated singular value decomposition. \u003cem\u003eIeee Access : Practical Innovations, Open Solutions\u003c/em\u003e, \u003cem\u003e8\u003c/em\u003e, 52575\u0026ndash;52587.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBodapati, J. D., Shareef, S. N., Naralasetti, V., \u0026amp; Mundukur, N. B. (2021). Msenet: Multi-modal squeeze-and-excitation network for brain tumor severity prediction. \u003cem\u003eInternational Journal of Pattern Recognition and Artificial Intelligence\u003c/em\u003e 2021, \u003cem\u003e35(07)\u003c/em\u003e, 2157005.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDuan, D., Xia, S., Rekik, I., Meng, Y., Wu, Z., Wang, L., Lin, W., Gilmore, J. H., Shen, D., \u0026amp; Li, G. (2019). Exploring folding patterns of infant cerebral cortex based on multi-view curvature features: Methods and applications. \u003cem\u003eNeuroimage\u003c/em\u003e, \u003cem\u003e185\u003c/em\u003e, 575\u0026ndash;592.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWang, F., Lian, C., Wu, Z., Zhang, H., Li, T., Meng, Y., Wang, L., Lin, W., Shen, D., \u0026amp; Li, G. (2019). Developmental topography of cortical thickness during infancy. \u003cem\u003eProceedings of the National Academy of Sciences 116(32)\u003c/em\u003e, 15855\u0026ndash;15860.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGilmore, J. H., Knickmeyer, R. C., \u0026amp; Gao, W. (2018). Imaging structural and functional brain development in early childhood. \u003cem\u003eNature Reviews Neuroscience\u003c/em\u003e, \u003cem\u003e19\u003c/em\u003e(3), 123\u0026ndash;137.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBecht, A. I., Kathryn, L., \u0026amp; Mills (2020). \u003cem\u003eModeling individual differences in brain development Biological Psychiatry\u003c/em\u003e 88.1 : 63\u0026ndash;69.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"brain-imaging-and-behavior","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"bior","sideBox":"Learn more about [Brain Imaging and Behavior](https://www.springer.com/journal/11682)","snPcode":"11682","submissionUrl":"https://submission.nature.com/new-submission/11682/3","title":"Brain Imaging and Behavior","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"morphological fingerprint, MRI, conformal mapping, deep learning, contrastive learning, perinatal period","lastPublishedDoi":"10.21203/rs.3.rs-4602847/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-4602847/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eThe morphological fingerprint in the brain is capable of identifying the uniqueness of an individual. However, whether such individual patterns are present in perinatal brains, and which morphological attributes or cortical regions better characterize the individual differences of neonates remain unclear. In this study, we proposed a deep learning framework that projected three-dimensional spherical meshes of three morphological features (i.e., cortical thickness, mean curvature, and sulcal depth) onto two-dimensional planes through quasi-conformal mapping, and employed the ResNet18 and contrastive learning for individual identification. We used the cross-sectional structural MRI data of 461 infants, incorporating with data augmentation, to train the model and fine-tuned the parameters based on 40 infants who had longitudinal scans. The model was validated on a fold of 20 longitudinal scanned infant data, and remarkable Top1 and Top5 accuracies of 85.90% and 92.20% were achieved, respectively. The sensorimotor and visual cortices were recognized as the most contributive regions in individual identification. Moreover, the folding morphology demonstrated greater discriminative capability than the cortical thickness. These findings provided evidence for the emergence of morphological fingerprints in the brain at the beginning of the third trimester, which may hold promising implications for understanding the formation of individual uniqueness in the brain during early development.\u003c/p\u003e","manuscriptTitle":"Identification of morphological fingerprint in perinatal brains using quasi-conformal mapping and contrastive learning","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-07-18 15:51:11","doi":"10.21203/rs.3.rs-4602847/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2024-10-31T18:00:47+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2024-09-22T20:36:15+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2024-09-19T23:54:48+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"277566730650796465901809453627371375127","date":"2024-09-19T05:08:59+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"24405632791331271500878437458560945129","date":"2024-09-17T09:47:14+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"232055774020535082979506580668299913825","date":"2024-08-11T19:25:59+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"103380918452654132244195121551351861779","date":"2024-08-08T18:12:51+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2024-08-06T13:57:00+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2024-08-06T13:50:18+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2024-06-20T06:01:53+00:00","index":"","fulltext":""},{"type":"submitted","content":"Brain Imaging and Behavior","date":"2024-06-19T03:22:06+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"brain-imaging-and-behavior","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"bior","sideBox":"Learn more about [Brain Imaging and Behavior](https://www.springer.com/journal/11682)","snPcode":"11682","submissionUrl":"https://submission.nature.com/new-submission/11682/3","title":"Brain Imaging and Behavior","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"c97f9b51-b9f2-4be4-8cbb-33d20aa34fd4","owner":[],"postedDate":"July 18th, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[],"tags":[],"updatedAt":"2025-03-31T16:00:56+00:00","versionOfRecord":{"articleIdentity":"rs-4602847","link":"https://doi.org/10.1007/s11682-025-00998-8","journal":{"identity":"brain-imaging-and-behavior","isVorOnly":false,"title":"Brain Imaging and Behavior"},"publishedOn":"2025-03-27 15:57:22","publishedOnDateReadable":"March 27th, 2025"},"versionCreatedAt":"2024-07-18 15:51:11","video":"","vorDoi":"10.1007/s11682-025-00998-8","vorDoiUrl":"https://doi.org/10.1007/s11682-025-00998-8","workflowStages":[]},"version":"v1","identity":"rs-4602847","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-4602847","identity":"rs-4602847","version":["v1"]},"buildId":"qtupq5eGEP_6zYnWcrvyt","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}