Related Experiment Video
Updated: Oct 14, 2025
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
Spherical coordinates transformation pre-processing in Deep Convolution Neural Networks for brain tumor segmentation
Carlo Russo1, Sidong Liu2,3, Antonio Di Ieva2
1Computational NeuroSurgery (CNS) Lab, Macquarie Medical School, Faculty of Medicine, Health and Human Science, Macquarie University, 1st floor, 75 Talavera Rd, Macquarie Park, Sydney, NSW, 2109, Australia. carlo.russo@mq.edu.au.
A new 3D spherical coordinate transform improves deep convolutional neural network (DCNN) accuracy for brain tumor segmentation in Magnetic Resonance Imaging (MRI). This method enhances model generalizability across different datasets and imaging settings.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Neuro-oncology
Background:
- Magnetic Resonance Imaging (MRI) is crucial for brain tumor assessment.
- Deep Convolutional Neural Networks (DCNNs) show promise for brain tumor segmentation.
- DCNN performance degrades with variations in imaging data (e.g., resolution, machine settings).
Purpose of the Study:
- To introduce a 3D spherical coordinate transform for MRI data pre-processing.
- To enhance the accuracy and generalizability of DCNN models for brain tumor segmentation.
- To address data standardization issues in DCNN training and application.
Main Methods:
- Applied a 3D spherical coordinate transform to MRI data during pre-processing.
- Trained and evaluated DCNN models using both standard Cartesian and spherical coordinate inputs.
- Compared segmentation performance on glioma datasets, focusing on Tumor Core and Enhancing Tumor regions.
Main Results:
- The DCNN model trained with spherical transform pre-processed data outperformed the Cartesian-input model.
- Segmentation accuracy improved for gliomas, particularly for Tumor Core and Enhancing Tumor classes.
- Merging models trained on both input types yielded further accuracy enhancements.
Conclusions:
- The 3D spherical coordinate transform improves DCNN segmentation accuracy for brain tumors.
- This method offers resolution independence, enhancing model generalizability and potentially solving domain shift issues in transfer learning.
- The proposed pre-processing technique facilitates more robust and reliable brain tumor segmentation using DCNNs.
