US-ATHC: Unsupervised Multi-Class Glioma Segmentation via Adaptive Thresholding and Clustering.

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Neuro-oncology

Background:

• Accurate glioma segmentation in 3D MRI is vital for clinical decision-making.
• Supervised learning for glioma segmentation is limited by the scarcity of expert annotations.
• Unsupervised methods are needed to overcome data limitations in medical image analysis.

Purpose of the Study:

• To develop a fully unsupervised pipeline for glioma segmentation.
• To achieve accurate global tumor detection and multi-class subregion segmentation.
• To provide an annotation-independent solution for glioma analysis.

Main Methods:

• US-ATHC employs a two-step unsupervised approach: adaptive thresholding and hierarchical clustering.
• Step 1: Global tumor mask generation using adaptive thresholding (Sauvola) and 3D consistency fusion.
• Step 2: Multi-class subregion segmentation (active tumor, edema, necrosis) via optimized affinity propagation clustering.

Main Results:

• US-ATHC demonstrated high accuracy in tumor and subregion segmentation on the BraTS 2021 dataset.
• The method outperformed classical clustering and state-of-the-art deep learning models.
• External validation on the Gliobiopsy dataset confirmed robustness and clinical applicability.

Conclusions:

• US-ATHC offers an unsupervised, accurate, and computationally efficient solution for glioma segmentation.
• Its annotation-independent nature is ideal for data-scarce scenarios.
• The method supports integration into clinical workflows and large-scale neuroimaging studies.