Medical Image Segmentation Using Discretized nmODE
Quansong He, Tao He, Yi Zhang · 2023
In recent years, deep learning has been widely used in medical image segmentation, and many high-performance models have emerged. However, these models usually sacrifice simplicity in order to obtain more accurate results, which make them unsuitable for mobile health application that requires convenience. To address this issue, we propose a more efficient method. Using discretized Neural memory Ordinary Different Equation (NmODE) to replace the decoders of UNet and its variants. We validate the efficacy of the nmODE decoder through comprehensive evaluations on two distinct datasets-LiTS2017 in the 3D context and ISIC2017 in the 2D context. The results demonstrated the efficacy of nmODE in diminishing network parameters while maintaining, or in some cases, enhancing performance levels.