0139

Quantitating Neuroanatomic Volumetry and White Matter Hyperintensity Lesion wrapped in AI Model in Aging Cohorts as a determinant of Brain Age
Neha Yadav1, Niraj Kumar Gupta1, and Vivek Tiwari1
1Department of Biological Sciences, Indian Institute of Science Education and Research Berhampur, Berhampur, India

Synopsis

Keywords: Aging, Aging, Brain Age, Volumetry, White Matter Hyperintensity

Motivation: In an aging population, a subset of individuals at any age group present with low white matter hyperintensity (WMH) volume in the brain, while another subset has intermediate to high WMH load.

Goal(s): To establish a Brain Age Estimation model involving WMH lesion quantification as a clinical indicator of Brain Health.

Approach: We have investigated the ‘Brain Health’ in terms of Brain Age using neuroanatomic volume, thickness together with WMH load across cognitively normal, impaired and Alzheimer’s Disease subjects.

Results: An increased Brain Age gap is observed for the subjects with elevated WMH load compared to the brains with low WMH.

Impact: Brain health is a composite representation of structural, fiber and vascular health. For the first time, a MR based quantitative platform with WMH load and comprehensive neuroanatomic volumetry is established, which estimates ‘Brain Age’ as an indicator of Brain Health.

INTRODUCTION

A hyperintense lesion in White-matter area observed on T2-FLAIR images is termed as white matter hyperintensity (WMH), arising due to small vessel infarcts. WMH load varies at large in an aging population, a subset of individuals at a given age have no WMH, while another subset harbors intermediate to high WMH load1,2. WMH load would lead to increased vascular insult to the brain regions but it is yet not established what is the threshold and WMH kinetics that would impact the neuroanatomical health cognitive status. Here, we have quantified WMH load, neuroanatomic-volume, and thickness in cognitively normal, MCI, and AD subjects from NACC and ADNI cohorts to establish a ‘RadioNeuroanatomic platform estimative of ‘Brain age’ as indicator of brain health. Additionally, we have also established a minimum Optimal number of Neuroanatomic features discriminative of cognitive status as CN, MCI, and AD for easy-to-use in clinical set-up.

METHODS

Longitudinal T1w and T2-Flair from 950 NACC subjects (CN=528, MCI=146, AD=276) were segmented using FreeSurfer3 for volumetry and thickness. Additionally, Periventricular WMH (PVWMH) and Deep WMH (DWMH) were quantified using an in-house optimized UBO-Detector4 pipeline. The chronological age, 178 neuroanatomical structure volume and two small vessel disease lesions: PVWMH and DWMH load from the Cognitively Normal subjects were used in the boosting algorithm method to obtain the optimal and accurate architecture model for Brain Age estimation (Fig.1). The training data only involved cognitively normal subjects who had NO PVWMH and DWMH to rule out any bias coming from WMH load. Bagging and error correction techniques were performed by dividing the training data into numerous training and validation sets. The training and validation processes were iterated for 50-times. Subsequently, the trained Brain Age model was used to predict Brain Age and Brain Age Gap5 (BAG) using the following equation.
BAG = Chronological Age – Estimated Brain Age
The top ten important features of the brain age prediction model were derived using permutation importance from scikit-learn6 Python library. The Brain Age model was also cross-validated for cognitively normal subjects, from the ADNI3 cohort (N=92), with Low (0-1.5ml) and High (5-10ml) WMH.

RESULTS

The Brain Age measure using neuroanatomic volume together with WMH load correlated significantly with chronological age as depicted by the average correlation coefficient (r=0.89±0.03) for 50 iterations (Fig.2A). Additionally, the Brain Age Gap (BAG) correlates positively with the total WMH load (r=0.24, p<0.0001) (Fig.2B). The cognitively normal subjects who had high WMH volume (5-10ml) load were estimated to have a significantly higher Brain Age Gap of 2.4±2.9 years at the early age group and 2.2±3.3 years at the intermediate age groups compared to the subjects with NO detectable WMH (Fig.3 A,B). BAG estimation from the ADNI3 cohort also revealed similar increased BAG for the subjects with and without WMH (Fig.3C). Permutation importance analyses for the top ten brain volumetry contributors towards Brain age estimation for the subjects with no WMH revealed that the volume of 3rd-ventricle is the topmost feature with an importance factor of 0.14 (Fig.4A) while, for the subjects with high WMH, PVWMH was the most important feature, with an importance factor of 0.083 (Fig.4B).

DISCUSSION

Our brain age model is the first report wherein we clearly present that periventricular and deep white matter hyperintensity load incurs increased Brain Age which is highr than the chronological age, thus indicating a faster brain health abrogation with Aging. Subjects with WMH >5ml had brain age gaps of more than ~3 years compared to the subjects with low or minuscule WMH. This further establishes that in order to understand the normal aging and pathological aging trajectory, one cannot treat the subjects with WMH and without WMH together. Owing to the vascular insults due to WMH load, the structural atrophy and hypertrophy kinetics will be distinct depending upon the threshold of WMH load.
It is intriguing to note that a unique set of brain features; PVWMH, Inferior-lateral-ventricle, Amygdala contributes towards the Brain Age for the subjects with high WMH load which is distinct from the low/nil WMH group. We also report that only CSF volume, WMH load and Brain volume serve as unique feature discriminative of CN, CI and AD (Fig.5).

CONCLUSION

The Brain Age model, which combines neuroanatomic volumetry and WMH load, clearly indicates significant contributions from WMH load towards Brain Age estimation, wherein elevated WMH load results in an increased Brain Age gap at a given chronological age. The Brain Age using Neuroanatomic volumetry and WMH load would serve as a precise non-invasive clinical marker of Brain Health and Cognitive status; CN CI and AD.

Acknowledgements

  1. The study is funded by Indian Council of Medical Research (ICMR).
  2. MRI and Cognitive data were obtained from the National Alzheimer’s Coordinating Center (NACC) database (funded by NIA/NIH Grant U24 AG072122) and Alzheimer's Disease Neuroimaging Initiative (ADNI) (National Institutes of Health Grant U01 AG024904) and DOD ADNI (Department of Defense award number W81XWH-12-2-0012).

References

  1. Yadav N, Majumdar A & Tiwari V. MRI measured Neuroanatomic volume, Cortical Thinning and WMH load with Aging: The Early, Intermediate and Late events of Cognitive status. International Society for Magnetic Resonance in Medicine.(2023).
  2. Gupta N K, Yadav N, Aman A, & Tiwari V. Brain WMH Load Kinetics & Regional Distribution with Aging: A Signature of Structural & Cognitive Health. International Society for Magnetic Resonance in Medicine.
  3. Fischl B, FreeSurfer. Neuroimage 62, 774-781 (2012).
  4. Jiang J, et al. UBO Detector - A cluster-based, fully automated pipeline for extracting white matter hyperintensities. Neuroimage 174, 539-549 (2018).
  5. Baecker L, Garcia-Dias R, Vieira S, Scarpazza C & Mechelli A. Machine learning for brain age prediction: Introduction to methods and clinical applications. EBioMedicine 72, 103600 (2021).
  6. Pedregosa F, et al. Scikit-learn: Machine Learning in Python. Journal of Machine Learning Research 12, 5 (2011).

Figures

Fig.1 Schematics illustrating the workflow of the Brain age prediction model

Fig.2 (A) The average association between Brain Age (BA) and Chronological Age (CA) was obtained from the ensemble average prediction for 50 iterations. (B) The linear regressions show the association of white matter hyperintensity with the Brain Age Gap (BAG) for the NACC cohort.

Fig.3 (A) Voxel-wise probability map depicting the occurrence of total WMH load (PVWMH +DWMH) on the coronal, sagittal, and axial slice for cognitively normal subjects in the early age group and intermediate age group with mean Brain Age Gap (BAG) and standard deviation. (B) Brain Age Gap (BAG) of the cognitively normal NACC and (C) ADNI3 subjects in the early age group (50-64), intermediate age group (65-79), and late age group (≥80) with no or low WMH (blue) and high WMH (orange).

Fig.4 The top ten brain MRI-determined neuroanatomic quantities ranked using permutation importance in decreasing order of contribution towards Brain Age estimation for subjects with (A) no WMH (PVWMH and DWMH < 1.5 ml) and subjects with (B) High WMH (PVWMH + DWMH > 5ml). The Coronal, sagittal, and axial MRI slices depicting segmented masks of important brain features overlaid on T1-weighted and T2-FLAIR images.

Fig.5 Cognitive status prediction accuracy of different ML Models for a combination of different MRI obtains neuroanatomic volumes and thicknesses. MRI features are added one by one to check for the increase in average accuracy of the XGB classifier (blue circle), Random Forest (green triangle), Bagging classifier (yellow diamond), and Simple Classification Tree (orange square) ML models. The result shows the mean accuracy ± SD (standard deviation). The highest accuracy for all ML models was obtained for the combination of 3 MRI features: total Brain volume, CSF, and WMH.

Proc. Intl. Soc. Mag. Reson. Med. 32 (2024)
0139
DOI: https://doi.org/10.58530/2024/0139