AUCseg: An Automatically Unsupervised Clustering Toolbox for 3D-Segmentation of High-Grade Gliomas in Multi-Parametric MR Images

Botao Zhao, Yan Ren, Ziqi Yu, Jinhua Yu, Tingying Peng, Xiao‐Yong ZhangFrontiers in Oncology 2021 · DOI

Abstract

The segmentation of high-grade gliomas (HGG) using magnetic resonance imaging (MRI) data is clinically meaningful in neurosurgical practice, but a challenging task. Currently, most segmentation methods are supervised learning with labeled training sets. Although these methods work well in most cases, they typically require time-consuming manual labeling and pre-trained models. In this work, we propose an automatically unsupervised segmentation toolbox based on the clustering algorithm and morphological processing, named AUCseg. With our toolbox, the whole tumor was first extracted by clustering on T2-FLAIR images. Then, based on the mask acquired with whole tumor segmentation, the enhancing tumor was segmented on the post-contrast T1-weighted images (T1-CE) using clustering methods. Finally, the necrotic regions were segmented by morphological processing or clustering on T2-weighted images. Compared with K-means, Mini-batch K-means, and Fuzzy C Means (FCM), the Gaussian Mixture Model (GMM) clustering performs the best in our toolbox. We did a multi-sided evaluation of our toolbox in the BraTS2018 dataset and demonstrated that the whole tumor, tumor core, and enhancing tumor can be automatically segmented using default hyper-parameters with Dice score 0.8209, 0.7087, and 0.7254, respectively. The computing time of our toolbox for each case is around 22 seconds, which is at least 3 times faster than other state-of-the-art unsupervised methods. In addition, our toolbox has an option to perform semi-automatic segmentation via manually setup hyper-parameters, which could improve the segmentation performance. Our toolbox, AUCseg, is publicly available on Github.

Introduction

High-grade gliomas (HGG) are aggressive brain tumors characterized by irregular shapes and indistinct boundaries, presenting a significant challenge for accurate segmentation in clinical practice. While supervised deep learning models, such as U-Net, have achieved high performance, they are constrained by the "labeling bottleneck"—the requirement for vast amounts of expert-annotated data. Furthermore, these models often struggle with generalization across different clinical settings. To address these limitations, we present AUCseg (Automatically Unsupervised Clustering segmentation), an open-source toolbox designed for fast, label-free 3D brain tumor segmentation.

Methods

AUCseg utilizes a hierarchical, unsupervised pipeline that mimics the logical progression of radiological analysis. The process is divided into four primary stages:

  1. Preprocessing: Includes skull-stripping, co-registration of multi-parametric MRI sequences (T2-FLAIR, T1-CE, T2), and intensity normalization.
  2. Whole Tumor (WT) Extraction: The entire affected area, including edema, is isolated by applying clustering algorithms to T2-FLAIR images.
  3. Enhancing Tumor (ET) Segmentation: Within the WT mask, the system identifies intensely enhancing regions using post-contrast T1-weighted (T1-CE) images.
  4. Necrotic Core (NC) Identification: Necrotic regions are segmented either through morphological "hole-filling" of the ET mask or via clustering on T2-weighted images.
Schematic diagram of the AUCseg automatic unsupervised pipeline.

The toolbox supports several clustering algorithms, including K-means, Mini-batch K-means, Fuzzy C Means (FCM), and Gaussian Mixture Models (GMM). Evaluation on 210 HGG cases from the BraTS2018 dataset identified GMM as the most robust method due to its probabilistic handling of "fuzzy" tissue boundaries.

Results and Discussion

The AUCseg toolbox demonstrates high efficiency and competitive accuracy without the need for training data. Using GMM clustering, the system achieved Dice scores of 0.8209 for the Whole Tumor, 0.7087 for the Tumor Core, and 0.7254 for the Enhancing Tumor.

Visual comparison of AUCseg segmentation results against ground truth labels for HGG.

A key advantage of AUCseg is its speed; the average processing time is approximately 22 seconds per case, which is significantly faster than other unsupervised methods like GHMRF. Additionally, the toolbox offers a semi-automatic mode allowing for manual hyperparameter adjustment, which can further improve Dice scores (e.g., WT accuracy increasing to 0.84).

While AUCseg provides a transparent and fast alternative to "black-box" deep learning models, it remains sensitive to image artifacts and incomplete skull-stripping. Future work will focus on improving robustness to pathological variability and narrowing the performance gap with state-of-the-art supervised models.

Conclusion

AUCseg offers a "ready-to-use" solution for brain tumor segmentation that eliminates the need for manual labeling and pre-trained models. Its speed and interpretability make it a promising tool for real-time clinical assistance. The toolbox is publicly available on GitHub (https://github.com/Haifengtao/AUCseg).

Summary updated

Abstract

The segmentation of high-grade gliomas (HGG) using magnetic resonance imaging (MRI) data is clinically meaningful in neurosurgical practice, but a challenging task. Currently, most segmentation methods are supervised learning with labeled training sets. Although these methods work well in most cases, they typically require time-consuming manual labeling and pre-trained models. In this work, we propose an automatically unsupervised segmentation toolbox based on the clustering algorithm and morphological processing, named AUCseg. With our toolbox, the whole tumor was first extracted by clustering on T2-FLAIR images. Then, based on the mask acquired with whole tumor segmentation, the enhancing tumor was segmented on the post-contrast T1-weighted images (T1-CE) using clustering methods. Finally, the necrotic regions were segmented by morphological processing or clustering on T2-weighted images. Compared with K-means, Mini-batch K-means, and Fuzzy C Means (FCM), the Gaussian Mixture Model (GMM) clustering performs the best in our toolbox. We did a multi-sided evaluation of our toolbox in the BraTS2018 dataset and demonstrated that the whole tumor, tumor core, and enhancing tumor can be automatically segmented using default hyper-parameters with Dice score 0.8209, 0.7087, and 0.7254, respectively. The computing time of our toolbox for each case is around 22 seconds, which is at least 3 times faster than other state-of-the-art unsupervised methods. In addition, our toolbox has an option to perform semi-automatic segmentation <i>via</i> manually setup hyper-parameters, which could improve the segmentation performance. Our toolbox, AUCseg, is publicly available on Github. (https://github.com/Haifengtao/AUCseg).
BibTeX
@inproceedings{zhao2021aucseg,
  author = {Botao Zhao and Yan Ren and Ziqi Yu and Jinhua Yu and Tingying Peng and Xiao‐Yong Zhang},
  title = {{AUCseg: An Automatically Unsupervised Clustering Toolbox for 3D-Segmentation of High-Grade Gliomas in Multi-Parametric MR Images}},
  booktitle = {Frontiers in Oncology},
  year = {2021},
  doi = {10.3389/fonc.2021.679952},
}
Figure