Accurate brain age prediction with lightweight deep neural networks

Han, Peng; Gong, Weikang; Beckmann, Christian F.; Vedaldi, Andrea; Smith, Stephen M.

doi:10.1101/2019.12.17.879346

Cited by 43 publications

(104 citation statements)

References 49 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…As part of the revision process, we also performed a comparison to a state-ofthe-art 3D convolutional neural network 19 , which was trained on our full set of sMRI scans (n = 8000) using the identical cross-validation splits. We used Peng's computer code to apply their algorithm architecture on the here examined sMRI images.…”

Section: Methodsmentioning

confidence: 99%

“…Performance of 3D convolutional DNN models. As part of the revision process, we have added a direct comparison to a state-ofthe-art three-dimensional (3D) convolutional neural network 19 , which was estimated based on our full set of training sMRI scans (n = 8000) using our identical cross-validation splits. This neural network architecture has won the first place in the 2019 predictive analytics competition on brain-imaging data (https://www.…”

Section: Performances On Brain Images Scales Similar To Linear Modelsmentioning

confidence: 99%

See 1 more Smart Citation

Different scaling of linear models and deep learning in UKBiobank brain images versus machine-learning datasets

et al. 2020

View full text Add to dashboard Cite

Recently, deep learning has unlocked unprecedented success in various domains, especially using images, text, and speech. However, deep learning is only beneficial if the data have nonlinear relationships and if they are exploitable at available sample sizes. We systematically profiled the performance of deep, kernel, and linear models as a function of sample size on UKBiobank brain images against established machine learning references. On MNIST and Zalando Fashion, prediction accuracy consistently improves when escalating from linear models to shallow-nonlinear models, and further improves with deep-nonlinear models. In contrast, using structural or functional brain scans, simple linear models perform on par with more complex, highly parameterized models in age/sex prediction across increasing sample sizes. In sum, linear models keep improving as the sample size approaches~10,000 subjects. Yet, nonlinearities for predicting common phenotypes from typical brain scans remain largely inaccessible to the examined kernel and deep learning methods.

show abstract

Section: Methodsmentioning

confidence: 99%

Section: Performances On Brain Images Scales Similar To Linear Modelsmentioning

confidence: 99%

Different scaling of linear models and deep learning in UKBiobank brain images versus machine-learning datasets

et al. 2020

View full text Add to dashboard Cite

show abstract

“…Our convolutional encoder network follows a very similar architecture to Peng et al 4 The 2 major differences are that we replace BatchNorm with InstanceNorm and replace all 3D convolutions/pooling operations with their 2D counterparts.…”

Section: D Slice Sequence Lstmmentioning

confidence: 99%

“…Peng et al 4 framed brain age prediction as a soft-class classification problem. Here a 3D convolutional network is trained by fitting data to a discretized normal distribution with mean equal to true chronological age of the individual and a standard deviation of 1.…”

Section: D "Info" Cnnmentioning

confidence: 99%

Accurate brain age prediction using recurrent slice-based networks

Lam

Santhalingam

Suresh

et al. 2020

Preprint

View full text Add to dashboard Cite

BrainAge (a subject’s apparent age predicted from neuroimaging data) is an important biomarker of brain aging. The deviation of BrainAge from true age has been associated with psychiatric and neurological disease, and has proven effective in predicting conversion from mild cognitive impairment (MCI) to dementia. Conventionally, 3D convolutional neural networks and their variants are used for brain age prediction. However, these networks have a larger number of parameters and take longer to train than their 2D counterparts. Here we propose a 2D slice-based recurrent neural network model, which takes in an ordered sequence of sagittal slices as input to predict the brain age. The model consists of two components: a 2D convolutional neural network (CNN), which encodes the relevant features from the slices, and a recurrent neural network (RNN) that learns the relationship between slices. We compare our method to other recently proposed methods, including 3D deep convolutional regression networks, information theoretic models, and bag-of-features (BoF) models (such as BagNet) - where the classification is based on the occurrences of local features, without taking into consideration their global spatial ordering. In our experiments, our proposed model performs comparably to, or better than, the current state of the art models, with nearly half the number of parameters and a lower convergence time.

show abstract

“…We added a bias correction as previously described (Smith et al 2019;Peng et al 2019) to correct age dependency of the training residuals. Briefly, we used a linear model ′ = ( ) = + to obtain an unbiased estimate of ′ as ̂ = ′ − , where the parameters and are estimated during training (on both the combination of training and validation set) and are thus applied directly to the test set.…”

Section: Training and Evaluation Of The Modelsmentioning

confidence: 99%

Predicting age and clinical risk from the neonatal connectome

Taoudi-Benchekroun

Christiaens

Grigorescu

et al. 2020

Preprint

View full text Add to dashboard Cite

The development of perinatal brain connectivity underpins motor, cognitive and behavioural abilities in later life. With the rise of advanced imaging methods such as diffusion MRI, the study of brain connectivity has emerged as an important tool to understand subtle alterations associated with neurodevelopmental conditions. Brain connectivity derived from diffusion MRI is complex, multi-dimensional and noisy, and hence it can be challenging to interpret on an individual basis. Machine learning methods have proven to be a powerful tool to uncover hidden patterns in such data, thus opening an opportunity for early identification of atypical development and potentially more efficient treatment.In this work, we used Deep Neural Networks and Random Forests to predict neurodevelopmental characteristics from neonatal structural connectomes, in a large sample of neonates (N = 524) derived from the developing Human Connectome Project. We achieved a highly accurate prediction of post menstrual age (PMA) at scan on term-born infants (Mean absolute error (MAE) = 0.72 weeks, r = 0.83, p<<0.001). We also achieved good accuracy when predicting gestational age at birth on a cohort of term and preterm babies scanned at term equivalent age (MAE = 2.21 weeks, r = 0.82, p<<0.001). From our models of PMA at scan for infants born at term, we computed the brain maturation index (i.e. predicted minus actual age) of individual preterm neonates and found significant correlation of this index with motor outcome at 18 months corrected age. Our results suggest that the neural substrate for later neurological functioning is detectable within a few weeks after birth in the structural connectome.

show abstract

Accurate brain age prediction with lightweight deep neural networks

Cited by 43 publications

References 49 publications

Different scaling of linear models and deep learning in UKBiobank brain images versus machine-learning datasets

Different scaling of linear models and deep learning in UKBiobank brain images versus machine-learning datasets

Accurate brain age prediction using recurrent slice-based networks

Predicting age and clinical risk from the neonatal connectome

Contact Info

Product

Resources

About