Fully Convolutional DenseNet with Multiscale Context for Automated Breast Tumor Segmentation.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Accurate breast tumor segmentation is vital for cancer diagnosis.
• Current methods often require manual interaction and pre-processing.
• Tumor variability in shape and size presents segmentation challenges.

Purpose of the Study:

• To develop a fully convolutional network for automated, end-to-end breast tumor segmentation.
• To improve segmentation accuracy for diverse malignant tumor characteristics in digital mammograms.

Main Methods:

• A fully convolutional dense network architecture was employed.
• Multiscale image information was integrated using atrous convolutions with varied sampling rates.
• A weighted loss function was utilized to address class imbalance during training.

Main Results:

• The proposed algorithm achieved automatic segmentation of breast tumors.
• High segmentation precision was demonstrated across various tumor sizes and shapes.
• The method requires no preprocessing or postprocessing steps.

Conclusions:

• The developed deep learning model offers an effective solution for automatic breast tumor segmentation.
• The approach shows robustness in handling diverse tumor morphologies.
• This automated method has the potential to streamline breast cancer diagnosis workflows.