Automatic lesion detection, segmentation and characterization via 3D multiscale morphological sifting in breast MRI

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Radiology

Background:

• Computer-aided detection/diagnosis (CAD) systems for 4D breast MRI often treat tasks separately and require manual input.
• Existing methods necessitate user intervention for slice or region selection, limiting efficiency.

Purpose of the Study:

• To develop an integrated, automated CAD system for 4D multimodal breast MRI.
• To enhance lesion detection, segmentation, and characterization without user intervention.

Main Methods:

• A novel 3D multiscale morphological sifting (MMS) for accurate and efficient lesion candidate generation and segmentation.
• Extraction of analytical features from multimodal sequences (T1, T2, DCE) for signal intensity, texture, morphology, and kinetics.
• Classification using random under-sampling boost (RUSboost) for lesion/normal tissue differentiation and random forest for malignancy assessment.

Main Results:

• Achieved a true positive rate (TPR) of 0.90 at 3.19 false positives per patient (FPP) for lesion detection.
• Achieved a TPR of 0.91 at 2.95 FPP for malignant lesion identification.
• Obtained an average Dice Similarity Index (DSI) of 0.72±0.15 for lesion segmentation.

Conclusions:

• The proposed integrated CAD system demonstrates favorable performance in breast lesion detection and characterization.
• The system's ability to handle 4D multimodal data and perform tasks without user intervention represents a significant advancement.
• This automated approach offers improved efficiency and accuracy for breast MRI analysis.