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

Stud Health Technol Inform. 2024 Aug 22:316:1115-1119. doi: 10.3233/SHTI240606.

Abstract

In light of the global increase in breast cancer cases and the crucial importance of the density of fibroglandular tissue (FGT) in assessing risk and predicting the course of the disease, the accurate measurement of FGT emerges as a significant challenge in diagnostic imaging. The current study focuses on the automatic segmentation of breast glandular tissue in MRI scans using a deep learning model. The aim is to establish a solid foundation for the development of methods for the precise quantification of fibroglandular tissue. For this purpose, the publicly available 'Duke Breast Cancer MRI' dataset was systematically processed to train a deep neural network model utilizing the nnU-Net ('no-new-Net') framework, which was then subjected to a quantitative evaluation. The results show the following macro-averaged metrics with standard deviation: Dice Similarity Coefficient 0.827 ± 0.152, accuracy 0.997 ± 0.003, sensitivity 0.825 ± 0.158, and specificity 0.999 ± 0.001. The effectiveness of our model in segmenting FGT is underscored by the high values of the Dice coefficient, Accuracy, Sensitivity, and Specificity, which reflect the precision and reliability of our results. The findings of this study lay a solid foundation for developing automated methods to quantify FGT. Our research efforts, especially driven by clinical studies at the University Hospital Augsburg, are focused on further exploring and validating these potentials.

Keywords: Deep Learning; Fibroglandular Tissue; Gynecology; Image Segmentation; Magnetic Resonance Imaging.

MeSH terms

  • Breast / diagnostic imaging
  • Breast Neoplasms* / diagnostic imaging
  • Deep Learning*
  • Female
  • Humans
  • Image Interpretation, Computer-Assisted / methods
  • Magnetic Resonance Imaging* / methods
  • Neural Networks, Computer
  • Reproducibility of Results
  • Sensitivity and Specificity