Automated Segmentation Techniques in Glioblastoma Imaging

Automated segmentation in glioblastoma imaging employs advanced computational models to delineate tumour boundaries and subcompartments on magnetic resonance images. These methods reduce observer variability, expedite volumetric assessment and enable longitudinal tracking of disease progression or response to therapy. Central to this field are deep learning frameworks—particularly convolutional neural networks—that learn from large, annotated datasets to identify contrast-enhancing regions, non-enhancing cores and peritumoural oedema. Recent innovations integrate transfer learning to adapt pre-trained models to institution-specific data, and spatial regularisation to improve robustness in post‐treatment scans where surgical cavities and blood products complicate image appearance. Performance is typically evaluated using overlap and distance metrics such as the Dice score and Hausdorff distance, with leading algorithms now achieving levels of agreement approaching expert manual delineation. Such tools are being incorporated into clinical workflows for surgical planning, radiation target definition and standardised reporting systems, with the ultimate aim of delivering objective and reproducible measures to guide multidisciplinary decision-making.

Research from Nature Portfolio

Recent studies have applied state-of-the-art neural architectures to early postoperative multi-modal MRI, training and validating models across multicentre cohorts of nearly 1 000 patients. These approaches achieved a residual tumour Dice score of around 0.61 and balanced classification accuracy of approximately 80 %, demonstrating generalisability across imaging protocols and institutions. A foundational investigation introduced a fully automated segmentation platform for longitudinal volumetry, comparing outputs with dual human raters over twelve-month follow-up scans. Automated volumetric trend curves for contrast-enhancing and non-enhancing compartments exhibited correlations equivalent to inter-rater agreement, highlighting the potential to replace manual delineation in prolonged monitoring scenarios.

Automated Segmentation Techniques in Glioblastoma Imaging publication trend

The graph below shows the total number of articles in automated segmentation techniques in glioblastoma imaging across all publications each year (not limited to Nature Index journals).

Technical terms

Automated segmentation: Computational delineation of tumour regions in medical images without manual input.

Magnetic resonance imaging (MRI): A non-invasive imaging modality using magnetic fields and radio waves to visualise soft tissues.

Deep learning: A subset of machine learning using layered neural networks to model complex patterns in data.

Convolutional neural network (CNN): A deep learning architecture tailored for image analysis, employing convolutional filters to extract features.

U-Net architecture: A CNN with symmetric encoding and decoding paths for precise image segmentation.

Transfer learning: The adaptation of a model pre-trained on one dataset to a related task with limited additional data.

Dice score: An overlap metric quantifying similarity between automated and reference segmentations, ranging from 0 (no overlap) to 1 (perfect overlap).

Hausdorff distance: A measure of the greatest boundary discrepancy between two segmented regions, used to assess spatial accuracy.

Resection cavity: The post-surgical void in brain tissue following tumour removal, which can complicate automated delineation.

Fluid-attenuated inversion recovery (FLAIR): An MRI sequence that suppresses cerebrospinal fluid signals to enhance lesion visibility.

References

1. Auto-segmentation of Adult-Type Diffuse Gliomas: Comparison of Transfer Learning-Based Convolutional Neural Network Model vs. Radiologists. Journal of Imaging Informatics in Medicine (2024).
2. A Fully Automated Post-Surgical Brain Tumor Segmentation Model for Radiation Treatment Planning and Longitudinal Tracking. Cancers (2023).
3. Improving the Generalizability of Deep Learning for T2-Lesion Segmentation of Gliomas in the Post-Treatment Setting. Bioengineering (2024).
4. Segmentation of glioblastomas in early post-operative multi-modal MRI with deep neural networks. Scientific Reports (2023).
5. Deep Learning for MRI Segmentation and Molecular Subtyping in Glioblastoma: Critical Aspects from an Emerging Field. Biomedicines (2024).
6. Clinical Evaluation of a Fully-automatic Segmentation Method for Longitudinal Brain Tumor Volumetry. Scientific Reports (2016).
7. Preoperative Brain Tumor Imaging: Models and Software for Segmentation and Standardized Reporting. Frontiers in Neurology (2022).