DeepCERES: A Deep learning method for cerebellar lobule segmentation using ultra-high resolution multimodal MRI

TL;DR

This work tackles the challenge of accurate cerebellar lobule segmentation by moving beyond mono-modal, 1 mm^3 data to multimodal MRI at ultrahigh resolution of $0.125~\text{mm}^3$, combining a two-stage cascade with memory-efficient architectures (DPN) and an atlas-guided prior. It constructs a ground-truth library via semi-automatic labeling on high-resolution data, augments it with a large lifespan-derived dataset, and integrates deep learning with multi-atlas label fusion to improve accuracy and robustness. An ensemble of U-Net and DPN architectures, enhanced by subject-specific atlases and extensive data augmentation, yields superior lobule Dice scores and robustness across variations, with a fully online DeepCERES pipeline capable of processing standard-resolution inputs. The pipeline, which includes super-resolution, T2 synthesis, and an automated reporting system, is designed for public release and broad applicability to lifespan and disease studies, supporting more precise cerebellar analysis in large-scale datasets. Overall, the combination of ultra-high-resolution multimodal data, atlas priors, model ensembles, and robust augmentation advances enables more reliable cerebellar lobule segmentation with practical clinical and research impact.

Abstract

This paper introduces a novel multimodal and high-resolution human brain cerebellum lobule segmentation method. Unlike current tools that operate at standard resolution ($1 \text{ mm}^{3}$) or using mono-modal data, the proposed method improves cerebellum lobule segmentation through the use of a multimodal and ultra-high resolution ($0.125 \text{ mm}^{3}$) training dataset. To develop the method, first, a database of semi-automatically labelled cerebellum lobules was created to train the proposed method with ultra-high resolution T1 and T2 MR images. Then, an ensemble of deep networks has been designed and developed, allowing the proposed method to excel in the complex cerebellum lobule segmentation task, improving precision while being memory efficient. Notably, our approach deviates from the traditional U-Net model by exploring alternative architectures. We have also integrated deep learning with classical machine learning methods incorporating a priori knowledge from multi-atlas segmentation, which improved precision and robustness. Finally, a new online pipeline, named DeepCERES, has been developed to make available the proposed method to the scientific community requiring as input only a single T1 MR image at standard resolution.

Figures (6)

• Figure 1: Left: 3D reconstruction of WM label and a coronal view of segmentation before semiautomatic correction. Right: 3D reconstruction of WM label and a coronal segmentation view after semiautomatic correction. The WM “arbor vitae” is better defined in the corrected version.
• Figure 2: Scheme of the proposed segmentation method. The first network segments left-right cerebellum and background and the second network segments the lobules using the T1 and T2 images and the output of the first network.
• Figure 3: Proposed DPN architecture.
• Figure 4: Scheme of the DeepCERES pipeline.
• Figure 5: Temporal profile of DeepCERES method