Fully Automated Enhanced Tumor Compartmentalization: Man vs. Machine Reloaded
Nicole Porz1,2, Simon Habegger1, Raphael Meier3
1Support Center for Advanced Neuroimaging-Institute for Diagnostic and Interventional Neuroradiology, University Hospital Inselspital and University of Bern, Bern, Switzerland.
Plos One
|November 3, 2016
Summary
A new automated brain tumor segmentation software (BraTumIA, BT) shows comparable results to a semi-automatic, FDA-approved method (SmartBrush®, SB) for glioblastoma. BT offers independence from human interaction, ideal for large datasets and future clinical studies.
Area of Science:
- Neuro-oncology
- Medical imaging analysis
- Artificial intelligence in medicine
Background:
- Accurate segmentation of glioblastoma (GBM) from MRI is crucial for treatment planning and outcome assessment.
- Existing methods include manual, semi-automatic, and fully-automatic segmentation techniques, each with varying accuracy and efficiency.
- Comparison of novel automated methods against established semi-automatic, FDA-approved tools is essential for clinical adoption.
Purpose of the Study:
- To compare the performance of a fully-automated segmentation software (BraTumIA, BT) with a semi-automatic, user-guided, FDA-approved technique (SmartBrush®, SB) for glioblastoma.
- To evaluate segmentation accuracy using metrics such as Dice coefficient, positive predictive value, sensitivity, and volume error.
- To assess the clinical utility and efficiency of both segmentation methods.
Main Methods:
- Nineteen patients with newly diagnosed glioblastoma underwent MRI on a 1.5 T scanner.
- Manual segmentation served as ground truth (GT).
- Semi-automatic segmentation was performed by four experts using SmartBrush® (SB), and fully-automatic segmentation used BraTumIA (BT).
Main Results:
- SmartBrush® (SB) achieved a mean Dice coefficient of 0.72–0.77, while BraTumIA (BT) achieved 0.68.
- Both BT and SB showed high correlation with GT for contrast-enhancing volumes (Pearson correlation 0.8 for BT).
- BT identified additional non-enhancing tumor tissue in 16/19 cases; analysis time was faster for SB (1:47–3:39 min) than BT (5 min).
Conclusions:
- BraTumIA (BT) and SmartBrush® (SB) offer comparable segmentation results for glioblastoma in a clinical context.
- BT's independence from human input makes it suitable for large datasets and integration with clinical/molecular data ('-omics').
- BT's multi-compartment segmentation may provide valuable insights into GBM subcompartments for future adjuvant therapy studies.
