Fighting the scanner effect in brain MRI segmentation with a progressive level-of-detail network trained on multi-site data

TL;DR

This work addresses the scanner effect in brain MRI segmentation by introducing LOD-Brain, a progressive level-of-detail architecture trained on a massive multi-site T1-weighted MRI dataset. By learning a coarse anatomical prior at a lower level and refining site-specific details at higher levels, the model achieves strong cross-site generalization with minimal performance gaps between internal and external data, and operates with a compact parameter count suitable for fast inference. The paper validates the approach through extensive multi-site experiments, ablations, and robustness analyses, showing competitive or superior performance across diverse datasets, demographics, and scanner configurations, while remaining open-source. The practical impact lies in enabling reliable, scalable brain structure segmentation across hospitals and imaging centers without per-site retraining or fine-tuning, facilitating large-scale clinical and research studies.

Abstract

Many clinical and research studies of the human brain require an accurate structural MRI segmentation. While traditional atlas-based methods can be applied to volumes from any acquisition site, recent deep learning algorithms ensure very high accuracy only when tested on data from the same sites exploited in training (i.e., internal data). The performance degradation experienced on external data (i.e., unseen volumes from unseen sites) is due to the inter-site variabilities in intensity distributions induced by different MR scanner models, acquisition parameters, and unique artefacts. To mitigate this site-dependency, often referred to as the scanner effect, we propose LOD-Brain, a 3D convolutional neural network with progressive levels-of-detail (LOD) able to segment brain data from any site. Coarser network levels are responsible to learn a robust anatomical prior useful for identifying brain structures and their locations, while finer levels refine the model to handle site-specific intensity distributions and anatomical variations. We ensure robustness across sites by training the model on an unprecedented rich dataset aggregating data from open repositories: almost 27,000 T1w volumes from around 160 acquisition sites, at 1.5 - 3T, from a population spanning from 8 to 90 years old. Extensive tests demonstrate that LOD-Brain produces state-of-the-art results, with no significant difference in performance between internal and external sites, and robust to challenging anatomical variations. Its portability opens the way for large scale application across different healthcare institutions, patient populations, and imaging technology manufacturers. Code, model, and demo are available at the project website.

Figures (11)

• Figure 1: LOD-Brain is a level-of-detail (LOD) network, where each LOD is a U-net which processes 3D brain multi-data at a different scale. Lower levels learn a coarse and site-independent brain representation, while superior ones incorporate the learnt spatial context, and refine segmentation masks at finer scales. Examples of outputs (grey matter renderings) at different LODs are shown in blue at the bottom.
• Figure 2: Multi-site dataset: we collect and analyse with MRIQC esteban2017mriqc almost 27,000 volumes originating from around 160 different sites (26,169 volumes after the quality check). (a) A visualisation by t-SNE tsne of the 68 MRIQC features (different colour for each dataset). Note that one dataset (e.g., https://brain-development.org/ixi-dataset/ in yellow colour) may contain volumes from more than one site or acquired with different scanners, and thus separate in clusters in the t-SNE space. (b) Dataset cardinalities. (c) Details on data quality assessment and (d) pre-processing. From (e) to (i), different demographic features and scanner properties are reported.
• Figure 3: LOD-Brain architecture selected for the experiments on the brain MRI segmentation task ($L=2$, $d=4$).
• Figure 4: (a) Using 70 available training datasets, we trained 4 models with $[1, 5, 10, 15]$ volumes per dataset. The model is tested on INT (red) and EXT data (blue). (b) Using 15 volumes per dataset, we train models with an increasing number of sites $[1, 4, 8, 16, 32, 64, 70]$. Testing is done on INT and EXT data (red and blue respectively). Both graphs fit exponential curves.
• Figure 5: Ablation study. Performance of models trained with different architectural options are shown with respect to the best model (on the zero x-axis). Results (Dice coefficient differences) are computed on the validation set (those marked with * are statistically significant according to $t_{test}$ applying Bonferroni correction).