Biology subjects

Yeh, F.-C.

Publications and source records attributed to Yeh, F.-C..

At least 19 recordsLinked to original sources

bioRxiv2026

White matter reflects the childhood exposome

The childhood environment is critical for brain development. However, most neuroimaging studies examine individual environmental measures (e.g., socioeconomic status) or a limited set of exposures, obscuring how the combination of complex, real-world exposures jointly influence brain development. Here we investigated how white matter shape and tissue properties are linked to the childhood exposome, a multidimensional measure capturing over 300 environmental exposures. Using multi-shell diffusion MRI from 8,183 children (ages 9-10) in the ABCD study, we quantified microstructural and macrostructural properties across 62 person-specific white matter tracts. The exposome showed widespread and highly replicable associations with both white matter microstructure and macrostructure: more advantaged environments were associated with larger tract macrostructure and lower orientation dispersion. Principal component analysis revealed that the dominant axis of exposome-white matter covariation aligns with the cortical sensorimotor-association hierarchy, such that tracts spanning this hierarchy exhibit the strongest associations with the exposome. Multivariate models demonstrated that patterns of white matter features explained 25% of the variance in the exposome in unseen individuals. Notably, white matter-based prediction of cognition was markedly reduced after accounting for the exposome ([~]82% reduction in explained variance), indicating that brain-cognition associations overlap substantially with variance captured by the exposome. These findings generalized to independent data from the Healthy Brain Network (n=869), which differs substantially from ABCD in MRI acquisition, participant selection, and childhood environments. Together, these results suggest that white matter architecture strongly reflects the childhood environment.

neuroscience↗

bioRxiv2026

Tractometry-Based Quantification of Along-Tract White-Matter Hemispheric Asymmetry in Alzheimer's Disease

White-matter hemispheric asymmetry is a fundamental property of human brain organization and is known to change in aging, neurodevelopment, and neurodegenerative disorders. Tractometry analyzes diffusion-derived microstructural measures along the full length of tracts, localizing changes to specific tract-segments rather than collapsing tracts into a single value. Yet, existing frameworks lack a principled way to quantify left-right hemispheric asymmetries along homologous tracts. Here, we introduce an asymmetry-aware tractometry framework that integrates a symmetric white-matter atlas with BUAN (Bundle Analytics) to enable anatomically consistent, along-tract comparison of homologous pathways. By defining homologous bundles with a shared template and consistent orientation, each left-hemisphere segment is directly matched to its right-hemisphere counterpart, enabling principled, segment-wise comparison and revealing spatially localized asymmetries along-tract. Applying this framework to diffusion MRI data from the Alzheimers Disease Neuroimaging Initiative (ADNI) comprising 1,215 subjects, we demonstrate how this approach reveals systematic left-right asymmetries across major white-matter pathways and show how these patterns differentiate cognitively normal (CN) individuals from those with mild cognitive impairment (MCI) and dementia. This method provides a sensitive and anatomically grounded tool for studying hemispheric specialization and its disruption in aging and disease, and establishes a general approach for asymmetry-aware tractometry in population neuroimaging studies.

neuroscience↗

bioRxiv2025

Anatomical White Matter Tracts Span the Cortical Hierarchy to Support Cognitive Diversity

Long-range white matter (WM) tracts support cognition by enabling communication between distant cortical regions, which are organized along a hierarchy defined by the sensorimotor-to-association (S-A) axis. However, it remains unknown how WM tracts are positioned within the cortical hierarchy to support cognition. Here, we show that WM tracts are differentially positioned in the cortical hierarchy to support specific cognitive functions, and that tracts spanning the hierarchy connect regions with greater cognitive diversity. Moreover, tracts situated within the same hierarchical level connect biologically similar regions, while those crossing the hierarchy bridge distinct biological milieux to support diverse cognitive functions. The placement of tracts in the cortical hierarchy also reflects developmental variation in tract microstructure and individual differences in cognition. Together, these findings provide a framework that moves beyond conventional categories of association or projection tracts and links WM tract anatomy to cortical organization, cognitive function, cortical neurobiology, and neurodevelopment. We anticipate that this cortex-anchored framework for describing WM tracts may aid the interpretation of individual differences in WM structure related to development and behavior.

neuroscience↗

bioRxiv2024

A sensorimotor-association axis of thalamocortical connection development

Human cortical development follows a sensorimotor-to-association sequence during childhood and adolescence1-6. The brains capacity to enact this sequence over decades indicates that it relies on intrinsic mechanisms to regulate inter-regional differences in the timing of cortical maturation, yet regulators of human developmental chronology are not well understood. Given evidence from animal models that thalamic axons modulate windows of cortical plasticity7-12, here we evaluate the overarching hypothesis that structural connections between the thalamus and cortex help to coordinate cortical maturational heterochronicity during youth. We first introduce, cortically annotate, and anatomically validate a new atlas of human thalamocortical connections using diffusion tractography. By applying this atlas to three independent youth datasets (ages 8-23 years; total N = 2,676), we reproducibly demonstrate that thalamocortical connections develop along a maturational gradient that aligns with the cortexs sensorimotor-association axis. Associative cortical regions with thalamic connections that take longest to mature exhibit protracted expression of neurochemical, structural, and functional markers indicative of higher circuit plasticity as well as heightened environmental sensitivity. This work highlights a central role for the thalamus in the orchestration of hierarchically organized and environmentally sensitive windows of cortical developmental malleability.

neuroscience↗

bioRxiv2024

An Open Resource: MR and light sheet microscopy stereotaxic atlas of the mouse brain

We have combined MR histology and light sheet microscopy (LSM) of five postmortem C57BL/6J mouse brains in a stereotaxic space based on micro-CT yielding a multimodal 3D atlas with the highest spatial and contrast resolution yet reported. Brains were imaged in situ with multi gradient echo (mGRE) and diffusion tensor imaging (DTI) at 15 m resolution ([~] 2.4 million times that of clinical MRI). Scalar images derived from the average DTI and mGRE provide unprecedented contrast in 14 complementary 3D volumes, each highlighting distinct histologic features. The same tissues scanned with LSM and registered into the stereotaxic space provide 17 different molecular cell type stains. The common coordinate framework labels (CCFv3) complete the multimodal atlas. The atlas has been used to correct distortions in the Allen Brain Atlas and harmonize it with Franklin Paxinos. It provides a unique resource for stereotaxic labeling of mouse brain images from many sources.

neuroscience↗

bioRxiv2023

Probabilistic Coverage of the Frontal Aslant Tract in Young Adults: Insights into Individual Variability, Lateralization, and Language Functions

The frontal aslant tract (FAT) is a crucial neural pathway of language and speech, but little is known about its connectivity and segmentation differences across populations. In this study, we utilized diffusion MRI automatic tractography to investigate the probabilistic coverage of the FAT in a large sample of 1065 young adults. Our primary goal was to reveal individual variability and lateralization of FAT and its structure-function correlations in language processing. Our results showed that the left anterior FAT exhibited the most substantial individual differences, particularly in the superior and middle frontal gyrus, with greater variability in the superior than the inferior region. Furthermore, we found significant left lateralization in FAT, with a greater difference in innervation coverage in the inferior and posterior portions. Additionally, our analysis revealed a significant correlation between the size of left FAT inferior innervation areas and Picture Vocabulary function, highlighting the structural and functional importance of the left FAT in language processing. In comparison, the anisotropy of FAT did not show significant correlation. Overall, our study provides valuable insights into individual and population differences in FAT connectivity and segmentation and sheds light on its critical role in language functions.

neuroscience↗

bioRxiv2022

Short superficial white matter and aging: a longitudinal multi-site study of 1,293 subjects and 2,711 sessions

It is estimated that short association fibers, or "U-shaped" fibers running immediately beneath the cortex, may make up as much as 60% of the total white matter volume. However, these have been understudied relative to the long-range association, projection, and commissural fibers of the brain. This is largely because of limitations of diffusion MRI fiber tractography, which is the primary methodology used to non-invasively study the white matter connections. Inspired by recent anatomical considerations and methodological improvements in U-fiber tractography, we aim to characterize changes in these fiber systems in cognitively normal aging, which provide insight into the biological foundation of age-related cognitive changes, and a better understanding of how age-related pathology differs from healthy aging. To do this, we used three large, longitudinal and cross-sectional datasets (N = 1293 subjects, 2711 sessions) to quantify microstructural features and length/volume features of several U-fiber systems. We find that axial, radial, and mean diffusivities show positive associations with age, while fractional anisotropy has negative associations with age in superficial white matter throughout the entire brain. These associations were most pronounced in the frontal, temporal, and temporoparietal regions. Moreover, measures of U-fiber volume and length decrease with age in a heterogenous manner across the brain, with prominent effects observed for pre- and post-central gyri. These features, and their variations with age, provide the background for characterizing normal aging, and, in combination with larger association pathways and gray matter microstructural features, may provide insight into fundamental mechanisms associated with aging and cognition.

neuroscience↗

bioRxiv2022

Aging and white matter microstructure and macrostructure: a longitudinal multi-site diffusion MRI study of 1,184 participants

Quantifying the microstructural and macrostructural geometrical features of the human brains connections is necessary for understanding normal aging and disease. Here, we examine brain white matter diffusion magnetic resonance imaging data from one cross-sectional and two longitudinal datasets totaling in 1184 subjects and 2236 sessions of people aged 50-97 years. Data was drawn from well-established cohorts, including the Baltimore Longitudinal Study of Aging dataset, Cambridge Centre for Ageing Neuroscience dataset, and the Vanderbilt Memory & Aging Project. Quantifying 4 microstructural features and, for the first time, 11 macrostructure-based features of volume, area, and length across 120 white matter pathways, we apply linear mixed effect modeling to investigate changes in pathway-specific features over time, and document large age associations within white matter. Conventional diffusion tensor microstructure indices are the most age-sensitive measures, with positive age associations for diffusivities and negative age associations with anisotropies, with similar patterns observed across all pathways. Similarly, pathway shape measures also change with age, with negative age associations for most length, surface area, and volume-based features. A particularly novel finding of this study is that while trends were homogeneous throughout the brain for microstructure features, macrostructural features demonstrated heterogeneity across pathways, whereby several projection, thalamic, and commissural tracts exhibited more decline with age compared to association and limbic tracts. The findings from this large-scale study provide a comprehensive overview of the age-related decline in white matter and demonstrate that macrostructural features may be more sensitive to heterogeneous white matter decline. Therefore, leveraging macrostructural features may be useful for studying aging and could have widespread implications for a variety of neurodegenerative disorders.

neuroscience↗

bioRxiv2022

HiDiver: A Suite of Methods to Merge Magnetic Resonance Histology, Light Sheet Microscopy, and Complete Brain Delineations

We have developed new imaging and computational workflows to produce accurately aligned multimodal 3D images of the mouse brain that exploit high resolution magnetic resonance histology (MRH) and light sheet microscopy (LSM) with fully rendered 3D reference delineations of brain structures. The suite of methods starts with the acquisition of geometrically accurate (in-skull) brain MRIs using multi-gradient echo (MGRE) and new diffusion tensor imaging (DTI) at an isotropic spatial resolution of 15 m. Whole brain connectomes are generated using over 100 diffusion weighted images acquired with gradients at uniformly spaced angles. Track density images are generated at a super-resolution of 5 m. Brains are dissected from the cranium, cleared with SHIELD, stained by immunohistochemistry, and imaged by LSM at 1.8 m/pixel. LSM channels are registered into the reference MRH space along with the Allen Brain Atlas (ABA) Common Coordinate Framework version 3 (CCFv3). The result is a high-dimensional integrated volume with registration (HiDiver) that has a global alignment accuracy of 10-50 m. HiDiver enables 3D quantitative and global analyses of cells, circuits, connectomes, and CNS regions of interest (ROIs). Throughput is sufficiently high that HiDiver is now being used in comprehensive quantitative studies of the impact of gene variants and aging on rodent brain cytoarchitecture.

neuroscience↗

bioRxiv2020

MASiVar: Multisite, Multiscanner, and Multisubject Acquisitions for Studying Variability in Diffusion Weighted Magnetic Resonance Imaging

PurposeDiffusion weighted imaging (DWI) allows investigators to identify structural, microstructural, and connectivitybased differences between subjects, but variability due to session and scanner biases is a challenge. MethodsTo investigate DWI variability, we present MASiVar, a multisite dataset consisting of 319 diffusion scans acquired at 3T from b = 1000 to 3000 s/mm2 across 14 healthy adults, 83 healthy children (5 to 8 years), three sites, and four scanners as a publicly available, preprocessed, and de-identified dataset. With the adult data, we demonstrate the capacity of MASiVar to simultaneously quantify the intrasession, intersession, interscanner, and intersubject variability of four common DWI processing approaches: (1) a tensor signal representation, (2) a multi-compartment neurite orientation dispersion and density model, (3) white matter bundle segmentation, and (4) structural connectomics. Respectively, we evaluate region-wise fractional anisotropy (FA), mean diffusivity, and principal eigenvector; region-wise cerebral spinal fluid volume fraction, intracellular volume fraction, and orientation dispersion index; bundle-wise shape, volume, FA, and length; and whole connectome correlation and maximized modularity, global efficiency, and characteristic path length. ResultsWe plot the variability in these measures at each level and find that it consistently increases with intrasession to intersession to interscanner to intersubject effects across all processing approaches and that sometimes interscanner variability can approach intersubject variability. ConclusionsThis study demonstrates the potential of MASiVar to more globally investigate DWI variability across multiple levels and processing approaches simultaneously and suggests harmonization between scanners for multisite analyses should be considered prior to inference of group differences on subjects.

bioengineering↗

bioRxiv2020

Tractography dissection variability: what happens when 42 groups dissect 14 white matter bundles on the same dataset?

White matter bundle segmentation using diffusion MRI fiber tractography has become the method of choice to identify white matter fiber pathways in vivo in human brains. However, like other analyses of complex data, there is considerable variability in segmentation protocols and techniques. This can result in different reconstructions of the same intended white matter pathways, which directly affects tractography results, quantification, and interpretation. In this study, we aim to evaluate and quantify the variability that arises from different protocols for bundle segmentation. Through an open call to users of fiber tractography, including anatomists, clinicians, and algorithm developers, 42 independent teams were given processed sets of human whole-brain streamlines and asked to segment 14 white matter fascicles on six subjects. In total, we received 57 different bundle segmentation protocols, which enabled detailed volume-based and streamline-based analyses of agreement and disagreement among protocols for each fiber pathway. Results show that even when given the exact same sets of underlying streamlines, the variability across protocols for bundle segmentation is greater than all other sources of variability in the virtual dissection process, including variability within protocols and variability across subjects. In order to foster the use of tractography bundle dissection in routine clinical settings, and as a fundamental analytical tool, future endeavors must aim to resolve and reduce this heterogeneity. Although external validation is needed to verify the anatomical accuracy of bundle dissections, reducing heterogeneity is a step towards reproducible research and may be achieved through the use of standard nomenclature and definitions of white matter bundles and well-chosen constraints and decisions in the dissection process.

neuroscience↗

bioRxiv2020

White matter neuroanatomical predictors of aphasic verb retrieval

BackgroundCurrent neurocognitive models of language function have been primarily built from evidence regarding object naming, and their hypothesized white matter circuit mechanisms tend to be coarse-grained. MethodsIn this cross-sectional, observational study, we used novel correlational tractography to assess the white matter circuit mechanism behind verb retrieval, measured via action picture-naming performance in adults with chronic aphasia. ResultsThe analysis identified tracts implicated in current neurocognitive dual-stream models of language function, including the left inferior fronto-occipital fasciculus, inferior longitudinal fasciculus, and arcuate fasciculus. However, the majority of tracts associated with verb retrieval were not ones included in dual-stream models of language function. Instead, they were projection pathways that connect frontal and parietal cortices to subcortical regions associated with motor functions, including the left corticothalamic pathway, frontopontine tract, parietopontine tract, corticostriatal pathway, and corticospinal tract. ConclusionsThese results highlight that cortico-subcortical projection pathways implicated in motor functions may be importantly related to language function. This finding is consistent with grounded accounts of cognition and may furthermore inform neurocognitive models. Impact StatementThis study suggests that in addition to traditional dual-stream language fiber tracts, the integrity of projection pathways that connect frontal and parietal cortices to subcortical motor regions may be critically associated with verb-retrieval impairments in adults with aphasia. This finding challenges neurological models of language function.

neuroscience↗

bioRxiv2020

QSIPrep: An integrative platform for preprocessing and reconstructing diffusion MRI

Diffusion-weighted magnetic resonance imaging (dMRI) has become the primary method for non-invasively studying the organization of white matter in the human brain. While many dMRI acquisition sequences have been developed, they all sample q-space in order to characterize water diffusion. Numerous software platforms have been developed for processing dMRI data, but most work on only a subset of sampling schemes or implement only parts of the processing workflow. Reproducible research and comparisons across dMRI methods are hindered by incompatible software, diverse file formats, and inconsistent naming conventions. Here we introduce QSIPrep, an integrative software platform for the processing of diffusion images that is compatible with nearly all dMRI sampling schemes. Drawing upon a diverse set of software suites to capitalize upon their complementary strengths, QSIPrep automatically applies best practices for dMRI preprocessing, including denoising, distortion correction, head motion correction, coregistration, and spatial normalization. Throughout, QSIPrep provides both visual and quantitative measures of data quality as well as "glass-box" methods reporting. Taken together, these features facilitate easy implementation of best practices for processing of diffusion images while simultaneously ensuring reproducibility.

bioinformatics↗

bioRxiv2020

Integrating across neuroimaging modalities boosts prediction accuracy of cognitive ability

Variation in cognitive ability arises from subtle differences in underlying neural architectural properties. Understanding and predicting individual variability in cognition from the differences in brain networks requires harnessing the unique variance captured by different neuroimaging modalities. Here we adopted a multi-level machine learning approach that combines diffusion, functional, and structural MRI data from the Human Connectome Project (N=1050) to provide unitary prediction models of various cognitive abilities: global cognitive function, fluid intelligence, crystallized intelligence, impulsivity, spatial orientation, verbal episodic memory and sustained attention. Out-of-sample predictions of each cognitive score were first generated using a sparsity-constrained principal component regression on individual neuroimaging modalities. These individual predictions were then aggregated and submitted to a LASSO estimator that removed redundant variability across channels. This stacked prediction led to a significant improvement in accuracy, relative to the best single modality predictions (approximately 1% to 4% boost in variance explained), across a majority of the cognitive abilities tested. Further analysis found that diffusion and brain surface properties contribute the most to the predictive power. Our findings establish a lower bound to predict individual differences in cognition using multiple neuroimaging measures of brain architecture, both structural and functional, quantify the relative predictive power of the different imaging modalities, and reveal how each modality provides unique and complementary information about individual differences in cognitive function. Author summaryCognition is a complex and interconnected process whose underlying mechanisms are still unclear. In order to unravel this question, studies usually look at one neuroimaging modality (e.g. functional MRI) and associate the observed brain properties with individual differences in cognitive performance. However, this approach is limiting because it fails to incorporate other sources of brain information and does not generalize well to new data. Here we tackled both problems by using out-of-sample testing and a multi-level learning approach that can efficiently integrate across simultaneous brain measurements. We tested this scenario by evaluating individual differences across several cognitive domains, using five measures that represent morphological, functional and structural aspects of the brain network architecture. We predicted individual cognitive differences using each brain property group separately and then stacked these predictions, forming a new matrix with as many columns as separate brain measurements, that was then fit using a regularized regression model that isolated unique information among modalities and substantially helped enhance prediction accuracy across most of the cognitive domains. This holistic approach provides a framework for capturing non-redundant variability across different imaging modalities, opening a window to easily incorporate more sources of brain information to further understand cognitive function.

neuroscience↗

bioRxiv2019

The Impact of Edema and Fiber Crossing on Diffusion MRI Metrics: DBSI vs. Diffusion ODF

PurposeDiffusion tensor imaging (DTI) has been employed for over two decades to noninvasively quantify central nervous system (CNS) diseases/injuries. However, DTI is an inadequate simplification of diffusion modeling in the presence of co-existing inflammation, edema, and crossing nerve fibers.\n\nMethodsWe employed a tissue phantom using fixed mouse trigeminal nerves coated with various amounts of agarose gel to mimic crossing fibers in the presence of vasogenic edema. Diffusivity measures derived by DTI and diffusion basis spectrum imaging (DBSI) were compared at increasing levels of simulated edema and degrees of fiber crossing. Further, we assessed the ability of DBSI, diffusion kurtosis imaging (DKI), generalized q-sampling imaging (GQI), q-ball imaging (QBI), and neurite orientation dispersion and density imaging (NODDI) to resolve fiber crossing, in reference to the gold standard angles measured from structural images.\n\nResultsDTI-computed diffusivities and fractional anisotropy (FA) were significantly confounded by gelmimicked edema and crossing fibers. Conversely, DBSI calculated accurate diffusivities of individual fibers regardless of the extent of simulated edema and degrees of fiber crossing angles. Additionaly, DBSI accurately and consistently estimated crossing angles in various conditions of gel-mimicked edema when comparing with gold standard (r2=0.92, p=1.9x10-9, bias=3.9{degrees}). Small crossing angles and edema sinficantly impact dODF, making DKI, GQI and QBI less accurate in detecting and estimating fibers corrsing angles. Lastly, we demonstrate DBSIs superiority over DTI for recovering and delineating white matter tracts in peritumoral edema for preoperative planning of surgical resection.\n\nConclusionsDBSI is able to separate two crossing fibers and accurately recover their diffusivities in a complex environment characterized by increasing crossing angles and amounts of gel-mimicked edema. DBSI also indicated better angular resolution capability compared with DKI, QBI and GQI.

neuroscience↗

bioRxiv2019

Differential Tractography as a Track-Based Biomarker for Neurodegeneration

Diffusion MRI tractography has been used to map the axonal structure of human brain, but its ability to detect neuronal injury is yet to be explored. Here we report differential tractography, a new type of tractography that utilizes repeat MRI scans and a novel tracking strategy to map the exact segment of fiber pathways with a neuronal injury. We examined differential tractography on multiple sclerosis, Huntington disease, amyotrophic lateral sclerosis, and epileptic patients. The results showed that the affected pathways shown by differential tractography matched well with the unique clinical symptoms of the patients, and the false discovery rate of the findings could be estimated using a sham setting to provide a reliability measurement. This novel approach enables a quantitative and objective method to monitor neuronal injury in individuals, allowing for diagnostic and prognostic evaluation of brain diseases.

neuroscience↗

bioRxiv2018

Machine learning versus logistic regression methods for 2-year mortality prognostication in a small, heterogeneous glioma database

BackgroundMachine learning (ML) is the application of specialized algorithms to datasets for trend delineation, categorization or prediction. ML techniques have been traditionally applied to large, highly-dimensional databases. Gliomas are a heterogeneous group of primary brain tumors, traditionally graded using histopathological features. Recently the World Health Organization proposed a novel grading system for gliomas incorporating molecular characteristics. We aimed to study whether ML could achieve accurate prognostication of 2-year mortality in a small, highly-dimensional database of glioma patients.\n\nMethodsWe applied three machine learning techniques: artificial neural networks (ANN), decision trees (DT), support vector machine (SVM), and classical logistic regression (LR) to a dataset consisting of 76 glioma patients of all grades. We compared the effect of applying the algorithms to the raw database, versus a database where only statistically significant features were included into the algorithmic inputs (feature selection).\n\nResultsRaw input consisted of 21 variables, and achieved performance of (accuracy/AUC): 70.7%/0.70 for ANN, 68%/0.72 for SVM, 66.7%/0.64 for LR and 65%/0.70 for DT. Feature selected input consisted of 14 variables and achieved performance of 73.4%/0.75 for ANN, 73.3%/0.74 for SVM, 69.3%/0.73 for LR and 65.2%/0.63 for DT.\n\nConclusionsWe demonstrate that these techniques can also be applied to small, yet highly-dimensional datasets. Our ML techniques achieved reasonable performance compared to similar studies in the literature. Though local databases may be small versus larger cancer repositories, we demonstrate that ML techniques can still be applied to their analysis, though traditional statistical methods are of similar benefit.

neuroscience↗

bioRxiv2018

Small Data Challenge: Structural Analysis and Optimization of Convolutional Neural Networks with a Small Sample Size

Deep neural networks have gained immense popularity in the Big Data problem; however, the availability of training samples can be relatively limited in certain application domains, particularly medical imaging, and consequently leading to overfitting problems. This \"Small Data\" challenge may need a mindset that is entirely different from the existing Big Data paradigm. Here, under the small data setting, we examined whether the network structure has a substantial influence on the performance and whether the optimal structure is predominantly determined by sample size or data nature. To this end, we listed all possible combinations of layers given an upper bound of the VC-dimension to study how structural hyperparameters affected the performance. Our results showed that structural optimization improved accuracy by 27.99%, 16.44%, and 13.11% over random selection for a sample size of 100, 500, and 1,000 in the MNIST dataset, respectively, suggesting that the importance of the network structure increases as the sample size becomes smaller. Furthermore, the optimal network structure was mostly determined by the data nature (photographic, calligraphic, or medical images), and less affected by the sample size, suggesting that the optimal network structure is data-driven, not sample size driven. After network structure optimization, the conventional convolutional neural network could achieve 91.13% in accuracy with only 500 samples, 93.66% in accuracy with only 1000 samples for the MNIST dataset and 94.10% in accuracy with only 3300 samples for the Mitosis (microscopic) dataset. These results indicate the primary importance of the network structure and the nature of the data in facing the Small Data challenge.

bioengineering↗