Application of Deep Learning Technology in Glioma

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neuro-oncology

Background:

• Glioma is a dangerous brain tumor requiring accurate segmentation for diagnosis and treatment.
• Magnetic resonance imaging (MRI) is crucial for visualizing gliomas.
• Deep learning methods, particularly convolutional neural networks (CNNs), are widely used for glioma segmentation.

Purpose of the Study:

• To develop an efficient and accurate deep learning model for glioma segmentation.
• To address limitations of existing 3D CNNs, such as insufficient spatial data acquisition and high memory consumption.
• To improve the analysis of glioma appearance for both indoor and outdoor patient cases.

Main Methods:

• A DM-DA-enabled cascading approach using a 2DResUnet model was developed.
• Multiscale fusion, attention mechanisms, and DenseBlocks were employed for segmentation.
• Fixed-region sampling and multisequence glioma image data were used to reduce model dimensionality.

Main Results:

• The DM-DA-Unet model demonstrated excellent performance in segmenting glioma edema, enhancement, and core areas on the BraTS17 dataset.
• The model achieved high average sensitivity, comparable to the best segmentation models on the BraTS1 dataset.
• The proposed CNN model effectively segments tumors with minimal memory consumption.

Conclusions:

• The developed DM-DA-Unet model offers a promising solution for accurate and memory-efficient glioma segmentation.
• This approach can aid clinicians in examining, analyzing, and diagnosing gliomas more effectively.
• The model's performance on benchmark datasets validates its potential for clinical application in neuro-oncology.