A Deep Neural Network Algorithm for Automatic Segmentation of Colon Gland Images

Abstract

The invention belongs to the technical field of image analysis area segmentation, and discloses a deep neural network algorithm for automatic segmentation of colon gland images. 1. Construct the data set: the colon gland data is used as a training set to obtain the instance map and contour map; 2. Construct the model: the model network includes a dense convolutional neural network and Refined U-Net, and the dense convolutional neural network is used to extract images. Rich primary feature information, the Refined U‑Net network connected with the dense convolutional neural network, learns the feature information of instances and contours; 3. Determine the model loss function: the loss function is the sum of Jaccard and focal loss; 4. Information fusion. This technical solution adopts the characteristics of deep dense neural network feature reuse and high parameter efficiency, and uses the combination of low-level features and high-level features of the Refined U-net network to construct a deep learning network model; uses the focal loss function to solve the problem of data sets Class imbalance problem, effectively make the contour accurate segmentation; finally achieve fast, clear and accurate colon gland image segmentation.

Description

technical field [0001] The invention belongs to the technical field of image analysis area segmentation, and specifically relates to a deep learning-based modeling, a deep dense convolutional neural network and a Refined U-net network. The focus loss function is used as the loss function of the contour task, and the Jaccard distance (Jaccard) is used. As the loss function of the instance segmentation task, Boundary F1 score (Boundary F1, BF), object-level balanced F (Object balanced F, Object F1) score, object-level similarity coefficient (Object Dice), object-level Hausdorff distance (Object Hausdorff ) as a deep neural network algorithm for automatic segmentation of colonic gland images as a quantitative evaluation method. Background technique [0002] The problem of biomedical image segmentation is one of the important research topics for intelligent medical diagnosis tasks, and it is also a difficult point in clinical practice evaluation. In practice, segmentation is us...

AI Technical Summary

Problems solved by technology

Although these methods have achieved good performance on the gland segmentation task in colon tissue, they have not considered the contour feature information of the gland.

The CNN sliding window method is based on image blocks, resulting in jagged edges of the segmented image; the method based on the total variational neural network uses multiple models for prediction, which has the disadvantages of cumbersome tasks and non-sharing of feature information

Images