6 results about "Curve evolution" patented technology

The present application relates to the field of image segmentation, and discloses a rapid curve evolution morphology-based medical image segmentation method, which allows for interactive selection of a segmentation area and can improve segmentation precision. The method comprises: converting a three-dimensional medical image into a multi-layer two-dimensional slice image; calibrating for the first time at least two user interaction seed points selected by a user from multiple layers, adding all pixels in a target area of the user interaction seed points to a distance queue, popping out the pixels one by one, calculating the path distances between the pixels and all adjacent pixels and determining whether the path distances are less than the values of the adjacent pixels in a distance transform map, and if not, updating the distance transform map and continuing adding the adjacent pixels into the distance queue until the queue is empty; dynamically calculating the user interaction seed points of each layer and performing calibration for the second time; preprocessing the calibrated two-dimensional slice image; for the calibrated two-dimensional slice image, calculating in parallel a curvature-based curve evolution morphology; and converting the two-dimensional slice image into a segmented three-dimensional medical image.

The application provides a deep neural network image segmentation method based on smoothness constraint, adopts a smoothness loss calculation method based on curve evolution as a loss function of a neural network segmentation model, and the loss function of the application firstly converts a segmentation label into a signed distance function, proposes a content loss measurement based on distance penalty, the loss size of a pixel point with segmentation error is positively correlated with the distance from the label boundary, and small regions with missegmentation are effectively avoided; then, an unsupervised length loss measurement is proposed, the measurement minimizes the boundary length of a predicted mask based on a level set method, and the smoothness of a segmentation result is ensured; finally, in order to solve the situation that a segmentation target disappears in a predicted image due to excessive constraint of the length loss function, an area compensation loss measurement is proposed, so as to provide an outward "compensation" loss for the process of segmenting an image.

The embodiment of the application discloses a hippocampus segmentation method and device based on deep learning and a storage medium, wherein the hippocampus segmentation method based on deep learning comprises the following steps: S1, a first image generated by nuclear magnetic resonance imaging is acquired, a target shape in the first image is inferred by using a trained deep belief network, and a second image containing the target shape is obtained; S2, an energy function is constructed according to the second image, a deep belief network driven lattice Boltzmann model is obtained based on the energy function, curve evolution is performed on the first image by using the deep belief network driven lattice Boltzmann model, and a third image and a fourth image after segmentation are obtained, and the fourth image contains the target shape; S3, steps S1 to S2 are executed for multiple times, multiple groups of third images and fourth images obtained are fed back to an error correction convolutional neural network model, multiple fifth images are obtained, the multiple fifth images are averaged and fused, and a segmentation result is obtained.

The invention relates to the field of image segmentation, and discloses a fast curve evolution morphological medical image segmentation method, which can interactively select a segmentation region and improve the segmentation precision. The method comprises the following steps: converting a three-dimensional medical image into a multi-layer two-dimensional slice image; at least two user interaction seed points selected by a user in multiple layers are calibrated for the first time, all pixels in a target area of the user interaction seed points are added into a distance queue, the pixels are popped up one by one, whether the path distance between the pixel and each adjacent pixel is smaller than the value of the adjacent pixel in the distance transformation diagram or not is calculated and judged, and if yes, the user interaction seed points are calibrated; if not, updating the distance transformation diagram and continuously adding the adjacent pixels into the distance queue until the queue is empty; dynamically calculating user interaction seed points of each layer and carrying out secondary calibration; preprocessing the calibrated two-dimensional slice image; carrying out parallel calculation on curve evolution morphology based on curvature on the calibrated two-dimensional slice image; and converting the two-dimensional slice image into a segmented three-dimensional medical image.

The invention discloses a medical image segmentation method based on an area-preserving convex curve evolution model, and belongs to the field of left ventricle medical image segmentation, and the method comprises the steps: obtaining a heart magnetic resonance image, and obtaining a to-be-segmented image; segmenting the to-be-segmented image by using a region scalable fitting image segmentation method to obtain a preliminary segmentation result of the left ventricle; determining an initial level set function of the area-preserving convexity curve evolution model based on the preliminary segmentation result of the left ventricle; and driving the initial level set function to move towards the contour of the left ventricle through the area-preserving convex curve evolution model so as to determine a final segmentation result of the left ventricle of the current layer. And segmenting the next layer of left ventricle by using the final segmentation result of the current layer of left ventricle until the left ventricle regions in all layers of images of the to-be-segmented image are completely segmented. According to the method, the problems that the left ventricle is difficult to completely segment by a common segmentation method and the left ventricle structure is often lost are solved.

The invention discloses an image segmentation method of a regional active contour model based on a Laplacian operator. According to the method, an adaptive-Laplacian operator is introduced, and a corresponding energy functional is used as a length regular term of a level set function, so that effective regularization of the level set function is realized. The regularization mechanism not only can suppress false contours generated in the curve evolution process, but also can adaptively adjust the diffusion process, so that noise interference in the level set evolution process is effectively suppressed, image edge structure details are kept, and segmentation robustness and accuracy are improved. Experimental results show that the method has high segmentation precision and robustness in complex natural image segmentation.