Comparative Analysis of Image Processing Techniques for Enhanced MRI Image Quality: 3D Reconstruction and...

Area of Science:

• Oncology
• Medical Imaging
• Artificial Intelligence

Background:

• Osteosarcoma is a prevalent bone tumor in children and adolescents.
• Manual segmentation of osteosarcoma in MRI is time-consuming and subjective.
• Automated segmentation can improve diagnostic efficiency and reproducibility.

Purpose of the Study:

• To develop and evaluate a convolutional neural network (CNN) for automatic osteosarcoma segmentation in MRI.
• To compare the performance of the CNN across different MRI sequences (T1W, T2W, T1W + Gd).

Main Methods:

• Acquired 3692 DICOM MRI images from 46 patients.
• Applied contrast stretching and median filtering for image pre-processing.
• Utilized a 3D U-Net architecture for deep learning-based segmentation.

Main Results:

• Achieved high segmentation accuracy with mean Dice Similarity Coefficients (DSC) of 83.75% (T1W), 85.45% (T2W), and 87.62% (T1W + Gd).
• Demonstrated the potential of the 3D U-Net model for segmenting osteosarcoma in various MRI modalities.

Conclusions:

• The developed CNN-based method offers an automated approach for osteosarcoma segmentation in MRI.
• Limitations such as poorly defined borders and missed lesions require further refinement for clinical application.