FT-FEDTL: A fine-tuned feature-extracted deep transfer learning model for multi-class microwave-based brain tumor...

Area of Science:

• Biomedical Engineering
• Medical Imaging Analysis
• Artificial Intelligence in Healthcare

Background:

• Microwave brain imaging (MBI) is an emerging technology for early brain tumor detection.
• Manual classification of microwave-based brain tumors (MBT) is time-consuming and challenging for physicians.
• Deep transfer learning (DTL) offers potential for efficient brain tumor classification.

Purpose of the Study:

• To propose an efficient DTL model, FT-FEDTL, for multi-class MBT classification.
• To enhance automated identification and categorization of MBT images for improved brain tumor diagnosis.
• To provide a superior pathway for brain tumor diagnosis using advanced AI.

Main Methods:

• Utilized the InceptionV3 architecture as a base for feature extraction in the FT-FEDTL model.
• Applied fine-tuning to five additional layers with hyperparameters for enhanced classification performance.
• Created a balanced dataset of 4200 MBT images using augmentation techniques from two sources.

Main Results:

• The FT-FEDTL model achieved superior classification performance on a balanced dataset compared to traditional and pre-trained models.
• Attained high performance metrics: 99.65% accuracy, 99.16% recall, 99.48% precision, 99.10% specificity, and 99.23% F-score.
• Demonstrated effectiveness in classifying six classes of brain tumors.

Conclusions:

• The FT-FEDTL model shows significant potential for assisting radiologists in multi-class MBT image classification.
• The proposed model offers a reliable and efficient tool for biomedical applications in brain tumor diagnosis.
• Experimental outcomes validate the model's capability for accurate and automated tumor classification.