Malignant Bone Tumors Diagnosis Using Magnetic Resonance Imaging Based on Deep Learning Algorithms

Area of Science:

• Oncology
• Radiology
• Artificial Intelligence

Background:

• Malignant bone tumors are aggressive with low survival rates.
• Timely diagnosis and treatment are crucial for improving patient prognosis.
• Current diagnostic timelines can be lengthy, impacting outcomes.

Purpose of the Study:

• To predict bone tumor malignancy using deep learning algorithms.
• To analyze magnetic resonance imaging (MRI) data for tumor classification.
• To reduce diagnostic time for bone tumors.

Main Methods:

• Utilized two pretrained ResNet50 deep learning models for T1 and T2 weighted MRI classification.
• Developed a feedforward neural network incorporating clinical data and MRI classifier outputs.
• Trained and validated the models on a cohort of 23 patients.

Main Results:

• Achieved high accuracies for T1 (93.67% training, 95.00% validation) and T2 (86.67% training, 95.00% validation) classifiers.
• The clinical model demonstrated an accuracy of 80.84% (training) and 80.56% (validation).
• Receiver operating characteristic (ROC) analysis confirmed the clinical model's ability for class separation.

Conclusions:

• The proposed deep learning method accurately predicts bone tumor malignancy from MRI.
• This approach bypasses the need for manual image segmentation, saving time.
• The algorithms can accelerate diagnosis and treatment initiation, improving patient prognosis.