Automatic Intracranial Segmentation: Is the Clinician Still Needed?

Area of Science:

• Radiation oncology
• Medical imaging
• Radiotherapy planning

Background:

• Stereotactic hypofractionated radiotherapy is effective for brain metastases.
• Manual segmentation of organs at risk is time-consuming.
• Automated segmentation aims to improve efficiency.

Purpose of the Study:

• To evaluate the performance of two commercial automated segmentation software.
• To compare automated segmentation with manual contouring for radiotherapy planning.
• To assess the suitability of automated tools for clinical use.

Main Methods:

• Twenty patients with brain metastases undergoing radiotherapy were included.
• CT scans were registered with MRI, and manual segmentation was performed.
• Two software (iPlan, Smart Segmentation) generated automated segmentations.
• Comparison included common segmented volume ratio, Dice-Sørensen coefficient, overlap ratio, and additional segmented volume.

Main Results:

• Agreement between manual and automated segmentation was poor across all metrics.
• Common segmented volumes ranged from 7.71% to 82.54%.
• Dice-Sørensen coefficients ranged from 0.0745 to 0.8398.
• Additional segmented volumes varied significantly (9.80% to 92.25%).
• Neither software consistently outperformed the other.

Conclusions:

• Current automated segmentation software is not sufficient for routine clinical use in stereotactic hypofractionated brain radiotherapy.
• Significant manual editing is still required to achieve accurate organ segmentation.
• Further development is needed to improve the reliability of automated segmentation tools.