Development and validation of a fully automatic tissue delineation model for brain metastasis using a deep neural...

Area of Science:

• Medical Imaging
• Artificial Intelligence in Oncology
• Neurosurgery

Background:

• Accurate brain metastasis delineation is crucial but time-consuming for stereotactic radiosurgery (SRS) planning.
• Previous studies using convolutional neural networks (CNNs) for automatic delineation lacked clinical evaluation.
• This study aimed to develop a clinically applicable AI tool for single brain metastasis delineation.

Purpose of the Study:

• Develop and clinically evaluate an AI tool for automatic single brain metastasis delineation.
• Integrate the AI tool into clinical practice for SRS treatment planning.
• Assess the AI tool's accuracy and efficiency compared to manual methods.

Main Methods:

• Retrospective analysis of 426 patients' postcontrast T1-weighted MRIs.
• Development of a novel 2.5D CNN for single brain metastasis delineation.
• Performance evaluation using Dice Similarity Coefficient (DSC) and Average Surface Distance (ASD), with clinical expert review.

Main Results:

• The AI tool achieved a mean DSC of 88.34%±5.00% and ASD of 0.35±0.21 mm.
• AI-generated contours were deemed acceptable for clinical use in 72.6% of slices.
• AI contouring time (74 seconds) was significantly less than manual contouring (196 seconds).

Conclusions:

• The AI tool demonstrates high accuracy in delineating single brain metastases, comparable to ground truth.
• The developed AI tool effectively reduces contouring time, aiding clinical workflow in Gamma Knife (GK) treatment planning.
• This AI tool shows promise for integration into clinical practice for brain metastasis SRS.