Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

SST-DUNet: Automated preclinical functional MRI skull stripping using Smart Swin Transformer and Dense UNet

Sima Soltanpour, Rachel Utama, Arnold Chang, Md Taufiq Nasseef, Dan Madularu, Praveen Kulkarni, Craig Ferris, Chris Joslin

TL;DR

This work addresses the key bottleneck of manual skull stripping in preclinical fMRI by introducing SST-DUNet, a lightweight 3D Dense UNet with a Smart Swin Transformer encoder. The approach leverages channel-specific attention via SSW-MSA and a combined Dice–Focal loss to handle class imbalance and boundary accuracy in low-resolution rat fMRI data. The method achieves state-of-the-art Dice scores around 0.98 across multiple in-house datasets and generalizes to unseen data, while maintaining fast inference suitable for practical preprocessing pipelines. Across seed-based functional connectivity and ICA analyses, SST-DUNet-produced skull-stripped data yield connectivity results highly concordant with manual processing, underscoring its potential to replace manual skull-stripping in preclinical studies.

Abstract

Skull stripping is a common preprocessing step that is often performed manually in Magnetic Resonance Imaging (MRI) pipelines, including functional MRI (fMRI). This manual process is time-consuming and operator dependent. Automating this process is challenging for preclinical data due to variations in brain geometry, resolution, and tissue contrast. While existing methods for MRI skull stripping exist, they often struggle with the low resolution and varying slice sizes in preclinical fMRI data. This study proposes a novel method called SST-DUNet, that integrates a dense UNet-based architecture with a feature extractor based on Smart Swin Transformer (SST) for fMRI skull stripping. The Smart Shifted Window Multi-Head Self-Attention (SSW-MSA) module in SST is adapted to replace the mask-based module in the Swin Transformer (ST), enabling the learning of distinct channel-wise features while focusing on relevant dependencies within brain structures. This modification allows the model to better handle the complexities of fMRI skull stripping, such as low resolution and variable slice sizes. To address the issue of class imbalance in preclinical data, a combined loss function using Focal and Dice loss is utilized. The model was trained on rat fMRI images and evaluated across three in-house datasets with a Dice similarity score of 98.65%, 97.86%, and 98.04%. The fMRI results obtained through automatic skull stripping using the SST-DUNet model closely align with those from manual skull stripping for both seed-based and independent component analyses. These results indicate that the SST-DUNet can effectively substitute manual brain extraction in rat fMRI analysis.

SST-DUNet: Automated preclinical functional MRI skull stripping using Smart Swin Transformer and Dense UNet

TL;DR

This work addresses the key bottleneck of manual skull stripping in preclinical fMRI by introducing SST-DUNet, a lightweight 3D Dense UNet with a Smart Swin Transformer encoder. The approach leverages channel-specific attention via SSW-MSA and a combined Dice–Focal loss to handle class imbalance and boundary accuracy in low-resolution rat fMRI data. The method achieves state-of-the-art Dice scores around 0.98 across multiple in-house datasets and generalizes to unseen data, while maintaining fast inference suitable for practical preprocessing pipelines. Across seed-based functional connectivity and ICA analyses, SST-DUNet-produced skull-stripped data yield connectivity results highly concordant with manual processing, underscoring its potential to replace manual skull-stripping in preclinical studies.

Abstract

Skull stripping is a common preprocessing step that is often performed manually in Magnetic Resonance Imaging (MRI) pipelines, including functional MRI (fMRI). This manual process is time-consuming and operator dependent. Automating this process is challenging for preclinical data due to variations in brain geometry, resolution, and tissue contrast. While existing methods for MRI skull stripping exist, they often struggle with the low resolution and varying slice sizes in preclinical fMRI data. This study proposes a novel method called SST-DUNet, that integrates a dense UNet-based architecture with a feature extractor based on Smart Swin Transformer (SST) for fMRI skull stripping. The Smart Shifted Window Multi-Head Self-Attention (SSW-MSA) module in SST is adapted to replace the mask-based module in the Swin Transformer (ST), enabling the learning of distinct channel-wise features while focusing on relevant dependencies within brain structures. This modification allows the model to better handle the complexities of fMRI skull stripping, such as low resolution and variable slice sizes. To address the issue of class imbalance in preclinical data, a combined loss function using Focal and Dice loss is utilized. The model was trained on rat fMRI images and evaluated across three in-house datasets with a Dice similarity score of 98.65%, 97.86%, and 98.04%. The fMRI results obtained through automatic skull stripping using the SST-DUNet model closely align with those from manual skull stripping for both seed-based and independent component analyses. These results indicate that the SST-DUNet can effectively substitute manual brain extraction in rat fMRI analysis.
Paper Structure (21 sections, 6 equations, 9 figures, 4 tables)

This paper contains 21 sections, 6 equations, 9 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Proposed Method
  3. Smart Swin Transformer
  4. Network Architecture
  5. Combo Loss Function
  6. Implementation Details
  7. Datasets
  8. Data Pre-processing
  9. Data for Training and Test
  10. Data Post-processing
  11. Experimental results
  12. MRI Image Analysis
  13. Evaluation Metrics
  14. Results
  15. Seed-based functional connectivity analysis
  16. ...and 6 more sections

Figures (9)

  • Figure 1: On the left side of the figure, the Smart Swin Transformer, composed of a W-MSA and an SSM-MSA, is shown. The right side illustrates the detailed computation process of the SSW-MSA, with the varying colors in the Smart Mask representing the distinct masks used across different channels.
  • Figure 2: The shifted window and Smart Mask within the SSW-MSA (Smart Shifted Window Multi-Head Self-Attention) mechanism of the Smart Swin Transformer (SST), specifically for one slice and one channel.
  • Figure 3: A Framework of the proposed model, SST-DUNet.
  • Figure 4: This figure presents the skull-stripping results across various CTNI datasets (A, B: dataset 1- C: dataset 2, D: dataset 3). The red region indicates the manually annotated ground truth, while the areas outlined in different colors correspond to the results produced by different methods.
  • Figure 5: This figure presents the skull stripping results for two fMRI samples from dataset 1 in three differnt views (Axial, Sagittal, and Coronal). The light green region indicates the manually annotated ground truth, while the areas outlined with different colored lines correspond to the outcomes from different methods.
  • ...and 4 more figures