Medical image segmentation method based on full perceived dynamic network

In clinical practice, the information access about tissues and lesion areas from the medical images is an important basis for disease diagnosis, treatment planning, and tracking detection. Accurate and automatic medical image segmentation is highly effective in improving diagnostic efficiency and quantitatively evaluating the treatment process. However, due to the diversity of biological tissues and the uncontrollable imaging situation, it is difficult to achieve stable and high-precision segmentation in medical images by the existing methods. In this work, anew medical image segmentation model, full perceived dynamic network (FPD-Net), is proposed to address this issue. First, amore efficient convolution method is designed to efficiently extract feature information by performing comprehensive multidimensional differential convolution operations on the input. Then, anew Transformer block called composite sensing Transformer (CST) is designed, which enhances the perceptual extraction of image detail features and multi-scale information through local detail enhancement and multi-scale encoding. In addition, a deep interaction encoding structure is built that efficiently integrates the advantages of convolutional encoding and autocorrelation encoding, enabling the model to fully perceive and utilize global and local information in the image. Finally, extensive experiments are conducted to validate our method in both tasks: retinal vessel segmentation and liver tumor segmentation, and make comparison with several state-of-the art methods. The results indicate that our FPD-Net outperforms other advanced methods in comprehensive segmentation performance. This shows the great potential of our method in medical assisted diagnosis.