FS-UNet: Mass segmentation in mammograms using an encoder-decoder architecture with feature strengthening

Area of Science:

• Medical Imaging
• Computer-Aided Diagnosis
• Artificial Intelligence in Radiology

Background:

• Accurate breast mass segmentation in mammograms is crucial for early cancer detection.
• Existing segmentation models often face challenges with accuracy and computational efficiency.

Purpose of the Study:

• To develop an effective and lightweight convolutional neural network model for automated breast mass segmentation.
• To improve the accuracy and robustness of breast mass segmentation in mammographic images.

Main Methods:

• Feature strengthening modules to enhance mass-related information.
• Parallel dilated convolution for multi-scale feature extraction.
• Mutual information loss function for optimizing prediction accuracy.

Main Results:

• The proposed model demonstrated excellent segmentation performance on the INbreast and CBIS-DDSM datasets.
• Achieved high scores in Dice coefficient, Intersection over Union, and sensitivity metrics.
• The lightweight design ensures clinical applicability and efficiency.

Conclusions:

• The developed model offers a promising solution for automated breast mass segmentation.
• It effectively enhances relevant features and captures multi-scale information for precise segmentation.
• The model's performance indicates its potential for improving mammogram analysis and patient outcomes.