Click here to

Session: Advancing Science to Expand Access to State-of-the-Art Applications in Medical Physics: I [Return to Session]

An AI-Based Stereotactic Radiosurgery Management Web Platform for Brain Metastases

Z Yang1*, M Chen1, H Jiang2, Z Wardak1, S Stojadinovic1, R Timmerman1, T Dan1, W Lu1, X Gu1,3, (1) UT Southwestern Medical Center, Dallas, TX, (2) NeuralRad LLC., Madison, WI, (3) Stanford University, Palo Alto, CA


TU-D1030-IePD-F5-4 (Tuesday, 7/12/2022) 10:30 AM - 11:00 AM [Eastern Time (GMT-4)]

Exhibit Hall | Forum 5

Purpose: Stereotactic radiosurgery (SRS) has demonstrated improved neurocognitive preservation intreating multiple brain metastases (mBMs) compared to whole brain radiotherapy. However, treatment planning and follow-up could be labor-intensive for mBMs patients while there’s no available tool for SRS automation and management. Here we proposed an AI-based mBMs web platform integrated with Gamma Knife (GK) treatment for SRS management.

Methods: The proposed web platform provides streamlined mBMs SRS management with four major modules: Contour, Group, Plan Review and Follow-up. For incoming new data, the Contour module firstly auto-segments BMs with embedded in-house CNN network. The segmentations is further adjusted by an AI-based false-positive (FP) reduction model and then auto-labeled via registration with Talairach atlas. Once the users approve the contours, the Group module auto-sorts BMs into different sessions for distributed GK-SRS treatments to minimize radiation toxicity and delivery time using an in-house spatiotemporal planning algorithm. The Plan Review module allows review of plan properties such as prescription, shots and grouping information. The Follow-up module provides access of treatment follow up image comparison and supports multi-course treatment dose tracking.

Results: The Contour module takes about 4 minutes to segment new data. The Hausdorff distance was 2.98mm and center of mass shift was 1.55mm for segmentation accuracy assessment. The initial BMs detection sensitivity was 0.93. After FP removal, the detection sensitivity was 0.87 and the FP rate was 0.09 with 100% label accuracy. The Group module takes less than 1 minute to generate BMs spatiotemporal distributions for distributed GK-SRS treatments. Compared to the manual distributions in current clinic workflow, the auto-distributions provided more uniformly distributed treatment volumes (p=0.013) and lower overall V12 (p=0.052). All processes run automatically in background.

Conclusion: We developed an AI-based SRS management web platform to improve clinical workflow efficiency for mBMs patients, with a potential generalization to other intracranial procedures.

Funding Support, Disclosures, and Conflict of Interest: This work is supported by NIH R01 CA235723


Gamma Knife, Stereotactic Radiosurgery, Quality Control


IM/TH- Formal Quality Management Tools: General (most aspects)

Contact Email