Deep Learning Based Automatic Fibroglandular Tissue Segmentation in Breast Magnetic Resonance Imaging Screening

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Increasing global breast cancer incidence necessitates improved risk assessment tools.
• Fibroglandular tissue (FGT) density in MRI is critical for breast cancer risk stratification and prognosis.
• Accurate measurement of FGT presents a significant challenge in diagnostic imaging.

Purpose of the Study:

• To develop and evaluate an automated deep learning model for segmenting breast glandular tissue in MRI scans.
• To establish a foundation for precise quantification of fibroglandular tissue (FGT).
• To enhance the accuracy and efficiency of FGT measurement in breast MRI.

Main Methods:

• Utilized the publicly available 'Duke Breast Cancer MRI' dataset.
• Trained a deep neural network using the nnU-Net framework for automated segmentation.
• Performed quantitative evaluation using metrics such as Dice Similarity Coefficient, accuracy, sensitivity, and specificity.

Main Results:

• The deep learning model achieved high performance in FGT segmentation.
• Macro-averaged metrics included Dice Similarity Coefficient 0.827 ± 0.152, accuracy 0.997 ± 0.003, sensitivity 0.825 ± 0.158, and specificity 0.999 ± 0.001.
• High metric values demonstrate the precision and reliability of the automated segmentation.

Conclusions:

• The developed deep learning model effectively segments fibroglandular tissue in breast MRI.
• This automated approach provides a reliable method for precise FGT quantification.
• Findings support the development of advanced tools for breast cancer risk assessment and management.