A Lightweight Deep Learning Based Microwave Brain Image Network Model for Brain Tumor Classification Using...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Biomedical Engineering

Background:

• Computerized classification of brain tumors from reconstructed microwave brain (RMB) images is crucial for monitoring brain disease progression.
• Existing methods require robust and efficient classification models for diagnostic accuracy.

Purpose of the Study:

• To propose an eight-layered lightweight classifier, the microwave brain image network (MBINet), for classifying RMB images into six distinct classes.
• To evaluate the performance of MBINet against other neural network models for brain tumor classification.

Main Methods:

• An experimental antenna sensor-based microwave brain imaging (SMBI) system was used to collect 1320 RMB images.
• Image preprocessing involved resizing and normalization, followed by augmentation to create 13,200 training images per fold for 5-fold cross-validation.
• The MBINet model, utilizing a self-organized operational neural network (Self-ONN), was trained and validated.

Main Results:

• MBINet achieved high performance metrics: 96.97% accuracy, 96.93% precision, 96.85% recall, 96.83% F1-score, and 97.95% specificity for six-class classification.
• MBINet outperformed four Self-ONNs, two vanilla CNNs, ResNet50, ResNet101, and DenseNet201, demonstrating superior classification outcomes (nearly 98%).

Conclusions:

• The MBINet model offers reliable and accurate classification of tumors from RMB images within the SMBI system.
• This lightweight model shows significant potential for clinical applications in brain tumor diagnosis and monitoring.