JoVE - Journal of Visualized Experiments
JoVE Visualize
Contact Us

Development and Evaluation of Automated Artificial Intelligence-Based Brain Tumor Response Assessment in Patients with Glioblastoma

  • 0From the Departments of Electrical and Computer Engineering (J.Z., M.A.M.), Duke University, Durham, North Carolina.
Ajnr. American Journal of Neuroradiology +

|

November 14, 2024

|

November 14, 2024

View abstract on PubMed

Summary

This summary is machine-generated.

An AI-based volumetric brain tumor MRI response assessment algorithm showed moderate agreement with radiologists but slightly worse survival prediction. This AI-VTRA tool requires further development for clinical use in glioblastoma patients.

Area Of Science

  • Neuro-oncology
  • Artificial Intelligence in Medicine
  • Medical Imaging Analysis

Background

  • Glioblastoma treatment response assessment relies on standardized MRI criteria.
  • Accurate and consistent response evaluation is crucial for patient management and clinical trials.
  • Automated AI tools could potentially improve efficiency and objectivity in MRI-based assessments.

Purpose Of The Study

  • To develop and evaluate an AI-based volumetric brain tumor MRI response assessment algorithm (AI-VTRA).
  • To assess the agreement of AI-VTRA with radiologist-based assessments (BT-RADS).
  • To evaluate AI-VTRA's ability to stratify glioblastoma patients by overall survival.

Main Methods

  • Retrospective analysis of 3,403 brain MRI exams from 634 glioblastoma patients.
  • Development of an AI-VTRA algorithm using automated volumetric tumor segmentation.
  • Evaluation of AI-VTRA agreement with BT-RADS and survival stratification using Kaplan-Meier and Cox models.

Main Results

  • AI-VTRA demonstrated moderate agreement (F1 = 0.587-0.755) with radiologist response assessments.
  • Kaplan-Meier analysis showed statistically worse survival for patients assessed by humans alone compared to AI alone (log-rank P = .007).
  • Cox proportional hazard models indicated a disadvantage for AI-based assessments in overall survival prediction (P = .012).

Conclusions

  • The AI-VTRA algorithm shows moderate performance in replicating human MRI response assessments for glioblastoma.
  • AI-VTRA exhibited slightly worse stratification of patients by overall survival compared to human assessments.
  • Further refinement of AI-VTRA is necessary for robust clinical application in brain tumor response assessment.