Search bioRxiv⌕ Search

bioRxiv · 10.1101/2021.04.13.439659

Training Data Distribution Significantly Impacts the Estimation of Tissue Microstructure with Machine Learning

Abstract

PurposeSupervised machine learning (ML) provides a compelling alternative to traditional model fitting for parameter mapping in quantitative MRI. The aim of this work is to demonstrate and quantify the effect of different training strategies on the accuracy and precision of parameter estimates when supervised ML is used for fitting. MethodsWe fit a two-compartment biophysical model to diffusion measurements from in-vivo human brain, as well as simulated diffusion data, using both traditional model fitting and supervised ML. For supervised ML, we train several artificial neural networks, as well as random forest regressors, on different distributions of ground truth parameters. We compare the accuracy and precision of parameter estimates obtained from the different estimation approaches using synthetic test data. ResultsWhen the distribution of parameter combinations in the training set matches those observed in similar data sets, we observe high precision, but inaccurate estimates for atypical parameter combinations. In contrast, when training data is sampled uniformly from the entire plausible parameter space, estimates tend to be more accurate for atypical parameter combinations but may have lower precision for typical parameter combinations. ConclusionThis work highlights the need to consider the choice of training data when deploying supervised ML for estimating microstructural metrics, as performance depends strongly on the training-set distribution. We show that high precision obtained using ML may mask strong bias, and visual assessment of the parameter maps is not sufficient for evaluating the quality of the estimates.

Source connections

Explore related subjects

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Gyori, N. G., Palombo, M., Clark, C. A., Zhang, H., Alexander, D. C.. 2021-04-14. Training Data Distribution Significantly Impacts the Estimation of Tissue Microstructure with Machine Learning. https://doi.org/10.1101/2021.04.13.439659

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related preprints

E2F4 gene therapy reverses established synaptic and cognitive deficits and remodels learning-associated signaling in Alzheimer's disease mice

Alzheimer's disease (AD) causes progressive synaptic dysfunction, neurodegeneration, and cognitive decline, yet strategies capable of reversing established deficits remain limited. Here, we show that neuronal expression of a p38MAPK phosphorylation-resistant human E2F4 variant (hE2F4DN), delivered by adeno-associated virus, restores hippocampal long-term potentiation and memory in homozygous 5xFAD mice after functional impairment is established. Treatment also reduces amyloid-{beta} and phospho-tau pathology and prevents hippocampal neuronal loss. Integrated proteomic and phosphoproteomic analyses reveal remodeling of metabolic, redox, neuroimmune, synaptic-vesicle, and GSK3{beta}-associated signaling pathways. Following contextual fear conditioning, hE2F4DN-treated mice exhibit broader learning-associated synaptic proteomic signatures and distinct phosphoregulatory responses involving synaptic machinery. These findings demonstrate that E2F4 gene therapy can reverse established synaptic and cognitive deficits while preserving neuronal integrity and reshaping molecular responses to learning, identifying E2F4 phosphorylation as a potential therapeutic target for advanced AD states.

neuroscience↗

Complex Spatial and Temporal Organization of Functional Brain Networks

Recent research exploring brain-behavior associations has relied on the language of complex systems to understand how stable behavioral individual differences emerge from brain function. At the same time, advances in neuroimaging have provided greater insight into both the topological and temporal properties of macroscale cortical networks. These new approaches offer a potential avenue for understanding how functional activity in the brain might behave in a manner that aligns with complex-systems interpretations of psychological traits. The present study used resting-state data from the Human Connectome project (N = 994) to explore associations between the spatial and temporal features of functional brain networks. Using an adaptive frequency approach to explore brain function at different timescales together with a family of methods from complex systems theory, we identified patterns of brain state organization reflecting a state-trait continuum, with more stable trait-like dynamics and greater functional coupling between networks in lower frequency oscillations. Additionally, we found evidence of greater coupling across timescales in brain networks associated with higher-order cognition. This research supports interpretations of brain organization as a spatiotemporal hierarchy and provides future avenues for exploring brain-behavior associations from a complex system perspective.

neuroscience↗

A Mechanistic Account of Dedifferentiation in the Aging Visual System

To uncover the mechanisms of age-related dedifferentiation, we simulated neural degradation in a computational model of the ventral visual stream. Severing connections or adding noise, whether in inferotemporal cortex (IT) or upstream, reduced IT differentiation, indicating that dedifferentiation need not originate locally. Degrading peripheral V1, which preferentially drove place-selective IT neurons, selectively reduced place differentiation, a candidate mechanism for the stronger place dedifferentiation reported in aging.

neuroscience↗