Related Experiment Video
Updated: Aug 26, 2025
Role of Diffusion MRI Tractography in Endoscopic Endonasal Skull Base Surgery
Published on: July 5, 2021
Longitudinal Assessment of Posttreatment Diffuse Glioma Tissue Volumes with Three-dimensional Convolutional Neural
Jeffrey D Rudie1, Evan Calabrese1, Rachit Saluja1
1Department of Radiology and Biomedical Imaging, University of California, San Francisco, 513 Parnassus Ave, Suite S-261D, Box 0628, San Francisco, CA 94143 (J.D.R., E.C., D.W., J.B.C., S.C., C.P.H., A.M.R., L.P.S., J.E.V.M.); and Department of Radiology, Hospital of the University of Pennsylvania, Philadelphia, Pa (R.S.).
Artificial intelligence using neural networks can segment and track changes in diffuse glioma after treatment. These AI models show accuracy comparable to neuroradiologists for longitudinal assessment, highlighting their clinical potential.
Area of Science:
- Neuro-Oncology
- Medical Imaging
- Artificial Intelligence
Background:
- Diffuse gliomas require accurate posttreatment assessment for effective management.
- Longitudinal monitoring of treatment response is crucial for diffuse glioma patients.
- Current methods for longitudinal assessment can be time-consuming and subjective.
Purpose of the Study:
- To develop and evaluate artificial intelligence (AI) based neural networks for segmentation and longitudinal assessment of posttreatment diffuse glioma.
- To quantify changes in tumor tissues (edematous, infiltrated, or treatment-changed tissue [ED] and active tumor or enhancing tissue [AT]) over time.
- To compare the performance of AI networks against neuroradiologists in classifying longitudinal changes.
Main Methods:
- A retrospective cohort of 298 patients with diffuse glioma was used for training and testing AI models.
- Three-dimensional nnU-Net convolutional neural networks were trained for posttreatment tumor segmentation using multimodal MRI (T1, T2, T1 postcontrast, FLAIR).
- Separate nnU-Nets were trained on longitudinal data to quantify and classify changes in ED and AT, comparing AI performance to neuroradiologists.
Main Results:
- AI segmentation achieved Dice scores of 0.72–0.89 and volume similarities of 0.90–0.96.
- Longitudinal change networks accurately classified changes in ED and AT with 90%-91% accuracy.
- AI network accuracy for longitudinal changes was not significantly different from that of three neuroradiologists (90%-92% accuracy).
Conclusions:
- AI-based automated segmentation and longitudinal assessment of posttreatment diffuse glioma show significant potential.
- Neural networks can accurately quantify and classify changes in tumor tissues, aiding in treatment response evaluation.
- This AI approach offers a promising tool for objective and efficient neuro-oncology patient management.
