Cell segmentation method based on boundary uncertainty estimation

Abstract

The invention discloses a cell segmentation method based on boundary uncertainty estimation, and the method comprises the following steps: carrying out the denoising and enhancement preprocessing of an original cell image, and determining a plurality of to-be-processed pixel positions; extracting a depth feature code of the preprocessed image, and judging whether a cell exists in the pixel position or not through a classification model; on the basis of a classification result, a regression model is adopted for pixel positions with cells to generate a closed cell contour; applying a boundary uncertainty estimation regression model to carry out local refinement on the cell contour; and screening the cell bounding box with the highest confidence coefficient as a final segmentation result by adopting a non-maximum suppression technology.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular to a cell segmentation method based on boundary uncertainty estimation. Background Art

[0002] Cells are the fundamental units of life, and cell segmentation is a crucial problem in many areas of biology. In cell biology, accurately segmenting cell boundaries is a key step in many studies, such as cell division, cell proliferation, and cell morphology. Therefore, developing efficient, accurate, and automated cell boundary segmentation methods is crucial for studying numerous problems in cell biology and medicine.

[0003] Cell segmentation is a crucial problem in many areas of biology. In cell biology, accurately segmenting cell boundaries is a key step in many studies, such as cell division, cell proliferation, and cell morphology. However, existing technologies struggle to accurately segment fuzzy or uncertain cell boundaries in images of cells from different imaging modalities. Summary of the Invention

[0004] Purpose of the invention: The purpose of the present invention is to provide a cell segmentation method based on boundary uncertainty estimation to solve the problems existing in the background technology.

[0005] Technical solution: The cell segmentation method based on boundary uncertainty estimation described in the present invention comprises the following steps:

[0006] S1, performing denoising and enhancement preprocessing on the original cell image to determine multiple pixel positions to be processed;

[0007] S2. Extracting the deep feature code of the preprocessed image and determining whether a cell exists at the pixel position through a classification model;

[0008] S3. Based on the classification results, a regression model is used to generate closed cell contours for the pixels where cells exist;

[0009] S4. Applying a boundary uncertainty estimation regression model to locally refine the cell contour;

[0010] S5. Use non-maximum suppression technology to select the cell bounding box with the highest confidence as the final segmentation result.

[0011] Furthermore, step S2 includes the following steps:

[0012] S21, using U-Net network to extract multi-scale deep feature coding;

[0013] S22, performing binarization determination on each pixel bit through the classification model to generate a cell foreground area mask;

[0014] S23. Construct a cell candidate region based on the full-image pixel determination result.

[0015] Furthermore, step S3 includes the following steps:

[0016] S31. Represent the closed cell contour as a vector of ordered point sets;

[0017] S32, performing Fourier series expansion on the vector to obtain a frequency domain equidistant sampling representation;

[0018] S33, training the regression network to learn the mapping relationship of the contour coordinates;

[0019] S34. Predicting contour coordinates in the inference stage to achieve pixel-level positioning.

[0020] Furthermore, step S4 includes the following steps:

[0021] S41. Construct a bounding box regression loss function based on the intersection-over-union (IoU) ratio, which is expressed as:

[0022] Among them, A is the predicted bounding box and B is the real bounding box;

[0023] S42. A multi-task learning framework is used to jointly optimize the main regression model and the boundary uncertainty estimation sub-network.

[0024] Furthermore, step S5 is specifically implemented as follows:

[0025] S51, establishing a candidate box confidence ranking queue;

[0026] S52, iteratively select the candidate box with the highest confidence, and calculate its overlap with the remaining candidate boxes;

[0027] S53, setting an IoU threshold of 0.6 to remove redundant frames;

[0028] S54, loop execution until all candidate boxes are screened.

[0029] Furthermore, the Fourier series expansion specifically uses the first N harmonic components to represent the contour shape, where the value of N ranges from 8 to 16.

[0030] Furthermore, the boundary uncertainty estimation subnetwork contains four parallel regression heads, corresponding to the uncertainty prediction of the left, right, top, and bottom directions of the cell boundary.

[0031] Furthermore, the confidence calculation integrates the classification probability score and the boundary regression quality assessment index, and the fusion weight is determined by grid search on the validation set.

[0032] An electronic device according to the present invention includes a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of any one of the methods when executing the program.

[0033] The computer-readable storage medium of the present invention is characterized in that it stores a computer program, and when the program is executed by a processor, the steps of any one of the methods are implemented.

[0034] Beneficial effects: Compared with the existing technology, the present invention has the following significant advantages: the present invention combines the advantages of deep neural network region classification and key point detection and positioning optimization, and can effectively use information such as cell boundaries to refine segmentation, thereby achieving accurate instance segmentation of different types of cell images. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 It is a flow chart of the present invention. DETAILED DESCRIPTION

[0036] The technical solution of the present invention will be further described below with reference to the accompanying drawings.

[0037] An embodiment of the present invention provides a cell segmentation method based on boundary uncertainty estimation, comprising the following steps:

[0038] S1, performing denoising and enhancement preprocessing on the original cell image to determine multiple pixel positions;

[0039] S2. Extract the deep feature code of the preprocessed image and use the classification model to determine whether a cell exists at a given pixel position. The specific steps include the following:

[0040] S2.1. Extract multi-scale deep feature encoding of pre-processed images using U-Net network.

[0041] S2.2. Classify multiple pixels using a classification model to determine the pixel locations where cells are present;

[0042] S2.3. According to the result of the U-Net network's judgment on all pixel positions of the image, the cell foreground area and the background area are obtained. The cell foreground area and the background area are obtained according to the result of the U-Net network's judgment on all pixel positions of the image.

[0043] S3. Based on the classification, according to the feature coding, a regression model is used to obtain complete and closed cell contours for the pixels where the classification model believes that cells exist. The specific steps include the following:

[0044] S3.1. Assume that each cell has a closed contour and represent each cell contour as a vector.

[0045] S3.2. Apply Fourier sine and cosine transforms to the two-dimensional coordinate set of the vectors to obtain equally spaced cell outline representations;

[0046] S3.3. Use the regression model to fit the two-dimensional coordinate data of the cell outlines in the training set, and store the learned coordinate representation information in the network parameters;

[0047] S3.4. In the segmentation inference stage, the trained regression model is used to predict the cell contour coordinates and the cell contour is located based on the pixel position.

[0048] S4, using an additional cell surrounding boundary uncertainty estimation regression model to locally refine the obtained cell contour; specifically comprising the following steps:

[0049] S4.1. Define the cell rectangle using the two-dimensional coordinates of the upper left corner and the lower right corner to obtain the pixel set A estimated by the current regression model;

[0050] S4.2. Calculate the intersection-and-union ratio of the previous step set A and the true cell rectangular bounding box set B as a loss term of the model. The calculation formula for the intersection-and-union ratio is:

[0051] S4.3. Use another additional regression model to locally refine the cell contour obtained in step S3.

[0052] S5. Use non-maximum suppression technology to filter out the cell box with the highest confidence and output it to complete cell segmentation. The specific steps include the following:

[0053] S5.1. Sort all locally refined cell boxes according to their confidence levels.

[0054] S5.2. Select the cell box with the highest current confidence and use it as one of the final output results;

[0055] S5.3. Calculate the Intersection over Union (IoU) overlap between the remaining cell boxes and the currently selected cell box. If the overlap is greater than a threshold of 0.6, delete the cell box being compared.

[0056] S5.4. Repeat steps S5.2 and S5.3 until all cell boxes are checked, output the cell box with the highest confidence, and complete the cell segmentation.

Claims

1. A cell segmentation method based on boundary uncertainty estimation, characterized in that: The following steps are involved: S1, performing denoising and enhancement preprocessing on the original cell image to determine multiple pixel positions to be processed; S2. Extracting the deep feature code of the preprocessed image and determining whether a cell exists at the pixel position through a classification model; S3. Based on the classification results, a regression model is used to generate closed cell contours for the pixels where cells exist; S4. Applying a boundary uncertainty estimation regression model to locally refine the cell contour; S5. Use non-maximum suppression technology to select the cell bounding box with the highest confidence as the final segmentation result.

2. A cell segmentation method based on boundary uncertainty estimation according to claim 1, characterized in that: Step S2 includes the following steps: S21, using U-Net network to extract multi-scale deep feature coding; S22, performing binarization determination on each pixel bit through the classification model to generate a cell foreground area mask; S23. Construct a cell candidate region based on the full-image pixel determination result.

3. A cell segmentation method based on boundary uncertainty estimation according to claim 1, characterized in that: Step S3 includes the following steps: S31. Represent the closed cell contour as a vector of ordered point sets; S32, performing Fourier series expansion on the vector to obtain a frequency domain equidistant sampling representation; S33, training the regression network to learn the mapping relationship of the contour coordinates; S34. Predicting contour coordinates in the inference stage to achieve pixel-level positioning.

4. The cell segmentation method based on boundary uncertainty estimation according to claim 1, characterized in that: Step S4: S41. Construct a bounding box regression loss function based on the intersection-over-union (IoU) ratio, which is expressed as: Among them, A is the predicted bounding box and B is the real bounding box; S42. A multi-task learning framework is used to jointly optimize the main regression model and the boundary uncertainty estimation sub-network.

5. The cell segmentation method based on boundary uncertainty estimation according to claim 1, characterized in that: Step S5 is specifically implemented as follows: S51, establishing a candidate box confidence ranking queue; S52, iteratively select the candidate box with the highest confidence, and calculate its overlap with the remaining candidate boxes; S53, setting an IoU threshold of 0.6 to remove redundant frames; S54, loop execution until all candidate boxes are screened.

6. A cell segmentation method based on boundary uncertainty estimation according to claim 3, characterized in that: The Fourier series expansion specifically uses the first N harmonic components to represent the contour shape, where the value of N ranges from 8 to 16.

7. A cell segmentation method based on boundary uncertainty estimation according to claim 4, characterized in that: The boundary uncertainty estimation subnetwork contains four parallel regression heads, corresponding to the uncertainty prediction of the left, right, top, and bottom directions of the cell boundary.

8. The cell segmentation method based on boundary uncertainty estimation according to claim 5, characterized in that: The confidence calculation combines the classification probability score and the boundary regression quality assessment index, and the fusion weight is determined by grid search on the validation set.

9. An electronic device, characterized in that: The method comprises a memory and a processor, wherein the memory stores a computer program, and when the processor executes the program, the steps of the method according to any one of claims 1 to 8 are implemented.

10. A computer-readable storage medium, characterized in that A computer program is stored, and when the program is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

Images