A Fully Automated Deep Learning Network for Brain Tumor Segmentation.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neuro-oncology

Background:

• Accurate brain tumor segmentation is crucial for diagnosis and treatment planning.
• Manual segmentation is time-consuming and subject to inter-observer variability.
• Deep learning offers a potential solution for automated and accurate segmentation.

Purpose of the Study:

• To develop and validate a fully automated deep learning method for segmenting brain tumors into subcomponents.
• To assess the performance of the developed method on multicentric datasets.

Main Methods:

• Developed three 3D-Dense-UNets for binary segmentation of whole tumor (WT), tumor core (TC), and enhancing tumor (ET).
• Utilized 285 multiparametric MRI cases from the BraTS2018 dataset for training and cross-validation.
• Evaluated the model using 3-fold cross-validation and tested on held-out, BraTS validation, and independent clinical datasets.

Main Results:

• Achieved mean cross-validation Dice scores of 0.92 (WT), 0.84 (TC), and 0.80 (ET).
• Demonstrated high accuracy on held-out test cases (0.90 WT, 0.84 TC, 0.80 ET) and external datasets (up to 0.90 WT, 0.82 TC, 0.80 ET).
• The automated method showed consistent performance across different datasets.

Conclusions:

• A fully automated deep learning method for brain tumor segmentation was successfully developed.
• The method achieves high prediction accuracy, comparable to manual segmentation, on both benchmark and clinical data.
• This automated segmentation tool holds significant potential for integration into clinical workflows.