A multicenter study on deep learning for glioblastoma auto-segmentation with prior knowledge in multimodal imaging

Area of Science:

• Radiotherapy and Medical Imaging
• Artificial Intelligence in Oncology
• Neurosurgery and Neuro-oncology

Background:

• Accurate glioblastoma (GBM) segmentation is critical for effective radiotherapy planning.
• Manual segmentation is time-consuming and depends on expert experience.
• Existing automated methods may lack accuracy and generalizability.

Purpose of the Study:

• To develop and evaluate a novel deep learning-based auto-segmentation method (PKMI-Net) for GBM.
• To leverage prior knowledge from multimodal imaging for improved segmentation accuracy.
• To assess the clinical applicability and multicenter generalizability of the proposed method.

Main Methods:

• Retrospective study using data from 148 patients across four multicenter datasets.
• Development of PKMI-Net, a deep learning model utilizing multimodal imaging (CT, MRI).
• Evaluation of segmentation performance using Dice similarity coefficient (DSC), Hausdorff distance (HD95), average surface distance (ASD), and relative volume difference (RVD).

Main Results:

• PKMI-Net achieved high accuracy in segmenting gross tumor volume (GTV), clinical target volume 1 (CTV1), and CTV2 (DSCs: 0.94, 0.95, 0.92).
• Segmentation results were largely clinically acceptable with minimal revisions needed.
• The method demonstrated robust generalizability and consistent performance across diverse multicenter datasets.

Conclusions:

• The proposed PKMI-Net effectively automates GBM segmentation using multimodal imaging prior knowledge.
• This AI-driven approach significantly improves contouring accuracy and efficiency for radiotherapy planning.
• PKMI-Net shows potential to enhance the overall quality and efficiency of glioblastoma radiotherapy.