Implicit Fuzzification via Bounded Noise Injection for Robust Medical Image Segmentation
Researchers have developed NoiseUNet, a novel framework designed to enhance the robustness of medical image segmentation. This method injects bounded noise into skip connections of encoder-decoder architectures, like U-Net, to improve feature fusion across different scales. The technique implicitly fuzzifies segmentation boundaries, leading to more accurate and reliable results, particularly on datasets with ambiguous edges such as the newly introduced ThyR thyroid ultrasound dataset. AI
IMPACT Enhances robustness in medical image segmentation, potentially improving diagnostic accuracy for ambiguous cases.