Deep Learning Segmentation of Infiltrative and Enhancing Cellular Tumor at Pre- and Posttreatment Multishell
Louis Gagnon1, Diviya Gupta1, George Mastorakos1
1From the Departments of Radiology (L.G., D.G., C.C., T.M.S., U.N., N.F., A.M.D., J.D.R.), Pathology (V.G.), Radiation Medicine and Applied Sciences (C.R.M., T.M.S., J.H.G.), Neurologic Surgery (T.B.), Bioengineering (T.M.S.), and Neurosciences (J.D.S., D.P., A.M.D.), University of California San Diego, 9500 Gillman Dr, La Jolla, CA 92093; Cortechs.ai, San Diego, Calif (G.M., N.W.); Department of Translational Neurosciences, Pacific Neuroscience Institute and Saint John's Cancer Institute at Providence Saint John's Health Center, Santa Monica, Calif (S.K.); and Department of Biophysics, Medical College of Wisconsin, Milwaukee, Wis (K.M.S.).
A deep learning model accurately segments glioblastoma tumor on MRI scans and predicts patient survival. This advanced imaging method aids in distinguishing tumor recurrence from treatment effects, improving patient outcome predictions.
Area of Science:
- Radiology and Medical Imaging
- Artificial Intelligence in Medicine
- Oncology
Background:
- Glioblastoma poses significant challenges in treatment monitoring due to complex tumor characteristics.
- Accurate segmentation of enhancing and nonenhancing tumor is crucial for treatment assessment and survival prediction.
- Distinguishing recurrent tumor from posttreatment changes on MRI remains a clinical challenge.
Purpose of the Study:
- To develop and validate a deep learning (DL) method for segmenting cellular tumor in glioblastoma.
- To assess the DL model's ability to predict overall survival (OS) and progression-free survival (PFS).
- To differentiate recurrent glioblastoma from posttreatment effects using advanced MRI techniques.
Main Methods:
- Retrospective analysis of 1397 MRI scans from 1297 glioblastoma patients.
- Utilized multimodal MRI data including perfusion and multishell diffusion imaging.
- Employed a nnU-Net deep learning model for cellular tumor segmentation and survival prediction analysis.
Main Results:
- The DL model achieved a median Dice score of 0.79 for segmentation.
- The area under the receiver operating characteristic curve (AUC) for detecting recurrent tumor was 0.84.
- Estimated cellular tumor volume was significantly associated with OS and PFS in both internal and external test sets.
Conclusions:
- A deep learning model effectively segments glioblastoma tumor and predicts patient survival.
- The model accurately distinguishes tumor recurrence from treatment-related changes.
- Advanced imaging integrated with DL offers a powerful tool for glioblastoma management and outcome prediction.
