Tumor Segmentation in Breast DCE- MRI Slice Using Deep Learning Methods

Precise tumor segmentation on DCE-MRI images is critical for the diagnosis and treatment of breast cancer. Automatic segmentation methods help specialists reduce the heavy manual workload in analyzing breast lesions. This article presents a methodology for breast tumor segmentation in DCE-MRI images. Our tests are performed on the QIN Breast DCE-MRI set, which has 20 exams. The set of images was converted into 2D slices for data augmentation, making 2416 slices. Then, trained UNet and SegNet architectures for tumor segmentation in DCE- MRI slices. After segmentation, we apply a false positive reduction step using morphological operations. The proposed methodology showed promising results, with a dice of 97.55% and IoU of 95.30% using the SegNet architecture. The segmentation obtained can provide the specialist with a more detailed analysis of the breast tumor.