MRI-Based Breast Cancer Classification and Localization by Multiparametric Feature Extraction and Combination Using...

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Healthcare
• Oncology

Background:

• Deep learning (DL) shows promise in breast MRI but its effectiveness in multi-parametric MRI (mpMRI) combinations for breast cancer detection is under-investigated.
• Investigating DL for breast cancer detection using combined MRI sequences is crucial for improving diagnostic accuracy.

Purpose of the Study:

• To implement a DL method for breast cancer classification and detection by extracting and combining features from multiple MRI sequences.
• To evaluate the performance of DL in breast cancer detection using various mpMRI combinations.

Main Methods:

• A retrospective study involving 569 internal and 125 external cohort cases.
• A cascaded convolutional neural network and long short-term memory network was utilized for lesion classification and localization.
• Performance was assessed using sensitivity, specificity, AUC, and comparison with radiologist readings.

Main Results:

• The DL method achieved high classification AUCs of 0.98 (internal) and 0.91 (external).
• Without contrast-enhanced MRI (DCE-MRI), the DL method outperformed radiologists (AUC 0.96 vs. 0.90).
• Lesion localization sensitivity reached 0.97 with DCE-MRI and 0.93 with T2WI alone.

Conclusions:

• The DL method demonstrates high accuracy for breast cancer lesion detection and classification in both internal and external cohorts.
• Contrast-agent-free mpMRI combinations with DL offer comparable performance to DCE-MRI and radiologists, suggesting potential for reduced contrast agent use.