HAL will be down for maintenance from Friday, June 10 at 4pm through Monday, June 13 at 9am. More information
Skip to Main content Skip to Navigation
Master Thesis

Prédiction de l'âge cérébral chez le sujet sain en IRM anatomique par le deep learning

Abstract : Objectives: to define a clinically usable preprocessing pipeline for MRI data, to verify by interpretability methods whether knowledge is learned by algorithms are those that are related to the mechanism of cerebral aging, test the validity of the model on an independent cohort.
Data and methods: we used 1597 open-access T1 weighted MRI from 24 hospitals. Preprocessing consisted in applying: N4 bias field correction, registration to MNI152 space, white and grey stripe intensity normalization, skull stripping and brain tissue segmentation. Prediction of brain age was done with growing complexity of data input (histograms, grey matter from segmented MRI, raw data) and models for training (linear models, non linear model such as gradient boosting over decision trees, and 2D and 3D convolutional neural networks). Work on the interpretability of models involved (i) visualizing, displaying maps, maps to corrode, displaying values, and (ii) generating heat maps which permitted to identify regions of interest used by the algorithm in its learning. Finally, we tested the validity of this biomarker on an independent cohort of healthy subjects and Alzheimer's patients.
Results: processing time seemed feasible in a radiological workflow : 5 min for one 3D T1 MRI. We found a significant correlation between age and gray matter volume with a correlation r = -0.74. Our best model obtained a mean absolute error (MAE) of 3.60 years, with fine tuned convolution neural network (CNN) pretrained on ImageNet. The MAE was of 5.5 years with a method correcting the center effect. Our work on interpretability on simpler models permitted to observe heterogeneity of prediction depending on brain regions known for being involved in ageing (grey matter, ventricles). Occlusion method of CNN showed the importance of Insula and deep grey matter (thalami, caudate nuclei) in predictions.
Conclusions: predicting the brain age using deep learning could define a biomarker of cerebral aging, usable in daily neuroradiological practice. Our work on interpretability shows that models learn the most in brain regions known to be affected by aging. (grey matter, seat of atrophy, white matter, affected by leukoaraiosis, and ventricles, which tend to dilate with ageing). This methods give more confidence in their applicability of CNN in clinical practice, previously considered as a " Black Box ". Finally, the use of the model on another cohort of patients showed (i) its ability to generalize on independent data, and on the other hand its application in diseases to which it was not trained: the estimated age by the algorithm is higher in Alzheimer's patients. This constitutes a pathway for earlier detection of this disease, via imaging methods. The use of similar methods in multimodal MRI, including functional MRI, could be an interesting avenue. This biomarker could be used as a surrogate for Grey matter volumetry in studies aimed at improving the performance of morphological imaging in the diagnosis of degenerative diseases and in the clinical trials of treatments for these diseases.
Document type :
Master Thesis
Complete list of metadata

Contributor : Université Paris Descartes - Scd Connect in order to contact the contributor
Submitted on : Thursday, February 17, 2022 - 3:58:36 PM
Last modification on : Saturday, April 2, 2022 - 3:34:23 AM
Long-term archiving on: : Wednesday, May 18, 2022 - 7:12:25 PM


Files produced by the author(s)


Distributed under a Creative Commons Attribution - NonCommercial - NoDerivatives 4.0 International License


  • HAL Id : dumas-03578858, version 1


Paul Herent. Prédiction de l'âge cérébral chez le sujet sain en IRM anatomique par le deep learning. Médecine humaine et pathologie. 2019. ⟨dumas-03578858⟩



Record views


Files downloads