What question did this study set out to answer?

To develop and evaluate a deep-learning pipeline for subject-specific segmentation and anatomical mapping of glioblastoma sub-compartments using multiparametric MRI.

August 16, 2026Journal of Medical Engineering & TechnologyOpen Access

High precision segmentation of glioma and surroundings – a feasibility study using multiparametric MRI and deep learning

View Full Paper

Ask AI

Bookmark

Authors

ALArvid Lundervold ALArvid Lundervold SASaruar Alam

Discussion

Member takes

Overview

Feasibility study demonstrates high-precision deep learning segmentation of glioblastoma sub-compartments in MRI scans, suggesting potential utility for personalized radiotherapy planning.

Key Points

To develop and evaluate a deep-learning pipeline for subject-specific segmentation and anatomical mapping of glioblastoma sub-compartments using multiparametric MRI.
Designed an automated pipeline combining deep-learning segmentation with anatomical parcellation on multiparametric MRI (T1, T1-Gadolinium, T2, FLAIR) to extract non-enhancing/necrotic, enhancing, and edema compartments.
Evaluated segmentation accuracy against reference masks in n=50 UCSF-PDGM glioblastoma subjects and assessed longitudinal performance across six timepoints in one LUMIERE subject.
Achieved median Dice similarity coefficients of 0.90 for whole tumor, 0.94 for tumor core, and 0.86 for enhancing tumor, with 95th-percentile Hausdorff distances of 4.1 mm, 2.2 mm, and 2.0 mm, respectively (n=50).
Longitudinal tracking over six timepoints in a single subject yielded volumetric trajectories and regional burden patterns concordant with independent reference segmentations.

Cite This Study

Lundervold et al. (2026) studied this question.

synapsesocial.com/papers/6a817a01f2fb91fc834ad8c5 https://doi.org/10.1080/03091902.2026.2713703

View Full Paper

Ask AI

Bookmark

August 16, 2026Journal of Medical Engineering & TechnologyOpen Access

High precision segmentation of glioma and surroundings – a feasibility study using multiparametric MRI and deep learning

View Full Paper

Ask AI

Bookmark

Authors

ALArvid Lundervold ALArvid Lundervold SASaruar Alam

Discussion

Member takes

Overview

Feasibility study demonstrates high-precision deep learning segmentation of glioblastoma sub-compartments in MRI scans, suggesting potential utility for personalized radiotherapy planning.

Key Points

To develop and evaluate a deep-learning pipeline for subject-specific segmentation and anatomical mapping of glioblastoma sub-compartments using multiparametric MRI.
Designed an automated pipeline combining deep-learning segmentation with anatomical parcellation on multiparametric MRI (T1, T1-Gadolinium, T2, FLAIR) to extract non-enhancing/necrotic, enhancing, and edema compartments.
Evaluated segmentation accuracy against reference masks in n=50 UCSF-PDGM glioblastoma subjects and assessed longitudinal performance across six timepoints in one LUMIERE subject.
Achieved median Dice similarity coefficients of 0.90 for whole tumor, 0.94 for tumor core, and 0.86 for enhancing tumor, with 95th-percentile Hausdorff distances of 4.1 mm, 2.2 mm, and 2.0 mm, respectively (n=50).
Longitudinal tracking over six timepoints in a single subject yielded volumetric trajectories and regional burden patterns concordant with independent reference segmentations.

Cite This Study

Lundervold et al. (2026) studied this question.

synapsesocial.com/papers/6a817a01f2fb91fc834ad8c5 https://doi.org/10.1080/03091902.2026.2713703

View Full Paper

Ask AI

Bookmark