Automatic Classification of Focal Liver Lesions Based on Multi-Sequence MRI

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence

Background:

• Accurate diagnosis of focal liver lesions is crucial for patient treatment planning.
• Current methods may lack the precision needed for complex cases.
• Multi-sequence Magnetic Resonance Imaging (MRI) offers rich data for lesion characterization.

Purpose of the Study:

• To develop and evaluate a deep learning model for automated classification of focal liver lesions.
• To assess the model's performance across eight different MRI sequences.
• To improve diagnostic accuracy and efficiency in radiological practice.

Main Methods:

• A deep learning model incorporating feature extraction, feature fusion attention, and attention-guided data augmentation was developed.
• The model was trained and validated on multi-sequence MRI data for focal liver lesion classification.
• Classification was performed across seven distinct lesion categories.

Main Results:

• The model achieved a patient-wise classification accuracy of 0.9302 and a lesion-wise accuracy of 0.8592.
• An F1-score of 0.8395, recall of 0.8296, and precision of 0.8551 were obtained.
• The results demonstrate significant performance in classifying focal liver lesions.

Conclusions:

• Combining multi-sequence MRI with deep learning provides a powerful tool for liver lesion diagnosis.
• The proposed model shows effectiveness in supporting radiologists for accurate classification.
• This approach enhances radiological practice and patient treatment planning.