An Artificial Intelligence Framework for Measuring Human Spine Aging Using MRI

TL;DR

This work proposes an AI framework to quantify human spine aging from whole-spine sagittal T2-weighted MRI, predicting a spine age and deriving the Spine Age Gap (SAG) as a biomarker of spine health. It combines semantic segmentation via nnUnet, a deep convolutional neural network for age estimation, and a bias-correction step to produce SAG, trained on a large dataset of over 18,000 MRI series. A data-driven normal-spine definition using Canberra distance, UMAP, and HDBSCAN enables training on age-appropriate normal examples and supports meaningful SAG interpretation. The study shows strong spine-age prediction performance, demonstrates SAG’s associations with degenerative spine conditions and lifestyle factors, and suggests SAG as a useful clinical biomarker for spine health, with Grad-CAM analyses providing interpretability of model focus.

Abstract

The human spine is a complex structure composed of 33 vertebrae. It holds the body and is important for leading a healthy life. The spine is vulnerable to age-related degenerations that can be identified through magnetic resonance imaging (MRI). In this paper we propose a novel computer-vison-based deep learning method to estimate spine age using images from over 18,000 MRI series. Data are restricted to subjects with only age-related spine degeneration. Eligibility criteria are created by identifying common age-based clusters of degenerative spine conditions using uniform manifold approximation and projection (UMAP) and hierarchical density-based spatial clustering of applications with noise (HDBSCAN). Model selection is determined using a detailed ablation study on data size, loss, and the effect of different spine regions. We evaluate the clinical utility of our model by calculating the difference between actual spine age and model-predicted age, the spine age gap (SAG), and examining the association between these differences and spine degenerative conditions and lifestyle factors. We find that SAG is associated with conditions including disc bulges, disc osteophytes, spinal stenosis, and fractures, as well as lifestyle factors like smoking and physically demanding work, and thus may be a useful biomarker for measuring overall spine health.

Figures (7)

• Figure 1: Overview of the three steps of the model to produce biological spine age. In the first step, semantic segmentation of the spine is generated and masked using a nnUnet model Khallaghi2023quantitative. Next, the masked spine series are passed to a DCNN model consisting of multiple blocks of 3D convolution, batch-normalization, max-pooling layers, and a linear layer for the final prediction. The details of the model and layers are available in Table \ref{['tab_model']}. In the last step, we perform bias correction to produce bias-corrected spine age.
• Figure 2: Train, validation, and test set histogram based on gender across age brackets.
• Figure 3: Clusters based on UMAP-reduced spine conditions. Only clusters that were more than 15% of the population were kept and the rest merged into one. We labeled the clusters based on the dominant conditions in the cluster. The population percentage of each cluster is written in front of the label.
• Figure 4: Absolute error on the normal test set grouped based on gender and age bracket
• Figure 5: Grad-CAM heatmap on the middle frame of the MRI for four different patients. The values are adjusted based on $f(x) = max(ln(288x), 1)$ to have better contrast.