MNet++: Extended 2D/3D Networks for Anisotropic Medical Image Segmentation
Researchers have successfully reproduced and extended MNet, a hybrid 2D/3D convolutional network for medical image segmentation. The study verified MNet's performance on prostate MRI and liver CT datasets, achieving high Dice similarity coefficients. Two new extensions were introduced: a learned Fusion Gating mechanism for adaptive feature blending and a VMamba module for improved long-range modeling, both of which maintained robustness to anisotropic conditions. AI