A web-based brain metastases segmentation and labeling platform for stereotactic radiosurgery.

Area of Science:

• Medical imaging analysis
• Computational oncology
• Radiotherapy workflow optimization

Background:

• Stereotactic radiosurgery (SRS) is a standard treatment for brain metastases (BMs).
• Manual delineation of multiple BMs is time-consuming and creates workflow inefficiencies.
• There is a need for automated tools for BMs delineation and quantitative evaluation.

Purpose of the Study:

• To develop a web-based platform for automated brain metastases (BMs) segmentation and labeling.
• To enhance the efficiency of the SRS clinical workflow by automating delineation tasks.
• To build upon previous deep learning-based segmentation algorithms.

Main Methods:

• Developed a web-based platform using the Django framework with a web client and back-end server.
• Implemented automated tasks: skull stripping, deep learning-based BMs segmentation, and affine registration-based BMs labeling.
• Enabled postprocessing features including contour adjustment, false-positive removal, and DICOM RTStruct export.

Main Results:

• Evaluated on 10 clinical cases (12-81 BMs per case), with an average operation time of 4-5 minutes per patient.
• Achieved high segmentation accuracy with averaged metrics: center of mass shift (1.55 ± 0.36 mm), Hausdorff distance (2.98 ± 0.63 mm), and mean SSD (1.06 ± 0.31 mm).
• Reduced averaged false-positive over union (FPoU) from 0.43 to 0.19 and false-negative rate (FNR) remained at 0.15 after postprocessing.

Conclusions:

• The web-based platform substantially improves clinical efficiency compared to manual brain metastases contouring.
• This automated tool is valuable for assisting in SRS treatment planning and follow-up.
• The platform offers a practical solution for streamlining the SRS workflow.