The invention provides a medical image segmentation method and device, which can resolve the problem of accurate image segmentation under the condition that the boundary of medical image objects is fuzzy and the prospect difference is little. The method comprises the steps of conducting data enhancing and pre-processing on medical images; establishing multichannel nerve network; classifying the medical images after data enhancing and pre-processing through the multichannel nerve network, obtaining the classifying result of the prospect and background, and conducting boundary detection on the structure of the medical image after data enhancing and pre-processing, obtaining the boundary result; blending the classifying result of the prospect and the background and the boundary result throughmixing network, and segmenting real examples in the medical images, obtaining the final segmentation result.