MR Image Fusion-Based Parotid Gland Tumor Detection

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Differentiating benign and malignant parotid gland tumors is critical for effective treatment planning and diagnosis.
• Current methods for tumor differentiation can be challenging, time-consuming, and labor-intensive.

Purpose of the Study:

• To develop and evaluate an automated system for differentiating benign and malignant parotid gland tumors using Magnetic Resonance (MR) image fusion and deep learning.
• To compare the performance of different deep learning architectures and MR image fusion combinations.

Main Methods:

• Utilized MR images from 114 patients with parotid gland tumors.
• Applied a two-dimensional Discrete Wavelet Transform (DWT)-based image fusion technique to combine Apparent Diffusion Coefficient (ADC), T1-weighted contrast-enhanced (T1C-w), and T2-weighted sequences.
• Trained and tested ResNet18, GoogLeNet, and DenseNet-201 architectures on four fused image datasets (IF (ADC, T1C-w), IF (ADC, T2-w), IF (T1C-w, T2-w), and IF (ADC, T1C-w, T2-w)).
• Developed a Graphical User Interface (GUI) application for image fusion and classification.

Main Results:

• DenseNet-201 models achieved high accuracies: 95.45% for IF (ADC, T1C-w), 95.96% for IF (ADC, T2-w), and 92.93% for IF (ADC, T1C-w, T2-w).
• ResNet18 achieved 94.95% accuracy for IF (T1C-w, T2-w).
• The developed GUI application facilitates image fusion and prediction of tumor malignancy.

Conclusions:

• Image fusion combined with deep learning models offers a promising approach for accurate and efficient differentiation of parotid gland tumors.
• DenseNet-201 demonstrated superior performance across multiple fused MR image datasets.
• The GUI application provides a user-friendly tool to aid clinicians in diagnosis and treatment planning.