MultiEdge-ResDeepLab: A Deep Learning Architecture for Multi-Modal Medical Image Segmentation with Enhanced Edge Detection

In the field of automatic image segmentation, especially for multi-modal medical images, precise segmentation of key structures is essential for accurate disease diagnosis and treatment. This study presents a novel deep learning architecture-MultiEdge-ResDeepLab-that combines edge enhancement with multi-modal segmentation techniques to boost performance. By incorporating edge enhancement using the Sobel operator and multi-scale feature fusion, the model improves its ability to capture finer details. Additionally, the use of dilated convolutions and ASPP modules expands the receptive field, allowing the model to capture richer contextual information and better hanle images at various scales. Experiments on the brats20 multi-modal medical image dataset demonstrate that MultiEdge-ResDeepLab outperforms existing methods in segmenting whole tumor (WT), tumor core (TC), and enhancing tumor (ET) areas. After 300 training rounds, the model achieved a Dice coefficient reached 0.959, showing a significant improvement in segmentation accuracy. The results highlight that combining edge information with multi-scale features can improve segmentation performance in complex medical applications. MultiEdge-ResDeepLab not only improves segmentation precision but also provides new directions for image processing research.