Multidimensional Dual Encoding Network For Liver Lesion Classification From Multi-Phase Magnetic Resonance Imaging.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Liver cancer poses a significant mortality risk, necessitating advanced diagnostic tools.
• Current automated liver cancer analysis methods often overlook inter-modal correlations in magnetic resonance imaging (MRI).
• Existing approaches show limitations in liver lesion classification and prediction accuracy.

Purpose of the Study:

• To develop an automated method that leverages information correlation across multiple MRI modalities for liver cancer analysis.
• To enhance the classification and prediction performance of liver lesions using a novel network architecture.

Main Methods:

• A multidimensional dual encoding network was designed, incorporating multidimensional information extraction and a dual encoder.
• The network processes eight MRI modalities, extracting both 2D and 3D information.
• A classification structure with two differently connected networks was employed for joint prediction.

Main Results:

• The proposed method achieved a balanced F1 score of 0.781, Cohen_Kappa of 0.731, accuracy of 0.779, and AUC of 0.944 on a dataset of 498 multiphase MRI images.
• Ablation studies and comparisons with state-of-the-art methods validated the model's effectiveness.
• The network successfully utilized information from eight modalities to improve lesion analysis.

Conclusions:

• The multidimensional dual encoding network effectively integrates multimodal MRI data for improved liver lesion classification and prediction.
• This approach addresses limitations in existing methods by considering inter-modal information correlation.
• The developed method shows promise for enhancing automated liver cancer diagnosis and patient outcomes.