Speaker
Description
Breast cancer remains one of the leading causes of mortality among women, making early detection and accurate diagnosis essential. Magnetic resonance imaging (MRI) plays a key role in this, offering detailed insights into tumor characteristics. However, manual tumor segmentation is labor-intensive and prone to variability, prompting the exploration of automated deep learning techniques. These methods hold promise for improving diagnostic precision, treatment planning, and monitoring therapeutic response.
In this study, the performance of 2D and 3D deep learning techniques for breast tumor segmentation on diffusion-weighted MRI (DW-MRI) is evaluated. While 3D techniques are often considered superior, several factors challenge this assumption. Breast tissue is highly heterogeneous, complicating segmentation. Additionally, the 1 cm gap between DW-MRI slices can result in unrealistic 3D volume reconstructions, and the sensitivity of DW imaging to water movement may not accurately reflect anatomical structures across slices, raising further segmentation challenges.
DW-MRI, despite being underutilized compared to dynamic contrast-enhanced (DCE) MRI, offers advantages for patients who cannot undergo contrast-based scans, as it is faster and more cost-effective. This study focuses on DW-MRI and its associated ADC maps, which provide valuable information about tissue characteristics.
Through extensive testing on a complete dataset, it has been demonstrated that, despite the challenges posed by DW imaging and breast tissue, 3D segmentation techniques outperformed 2D methods. This research validates the use of 3D deep learning models for DW-MRI, highlighting the potential of DW imaging as a viable alternative to DCE-MRI in breast cancer diagnosis.
