CN107730498A - Novel uterine neck cell core partitioning algorithm - Google Patents

Abstract

A kind of new cervical cell core partitioning algorithm, comprises the following steps：1. using selective search segmentation by a width input picture primary segmentation into multiple ROI；2. calculating the friendship between adjacent R OI and than IOU, final election region is removed；3. utilizing mathematical morphology limit erosion algorithm, final Connected component is obtained, by the judgement to final Connected component quantity, screening obtains including the ROI of nucleus；4. applying CV model algorithms, the ROI obtained after screening is split, obtains the nucleus of fine segmentation.Present invention optimizes the segmentation effect of nucleus cervical cell core, simplify cutting procedure, reduction segmentation complexity, accurate segmentation result can be provided for follow-up cervical cell Quantitative Nuclear Morphometry.

Description

New cervical cell nucleus segmentation algorithm

technical field

The invention belongs to the field of medical image segmentation, in particular to a novel cervical cell nucleus segmentation algorithm.

Background technique

Cervical cancer is the most common gynecological malignancy. The high-incidence age of carcinoma in situ is 30-35 years old, and that of invasive carcinoma is 45-55 years old. In recent years, the onset tends to be younger. In recent decades, the widespread application of cervical cytology screening has enabled early detection and treatment of cervical cancer and precancerous lesions, and the incidence and mortality of cervical cancer have been significantly reduced. How to accurately complete cervical cancer screening has become an important issue. focus of attention.

Among the many cervical cancer screening methods, cervical smear examination is considered to be the most common, common and effective cytological test for early screening of cervical cancer. Quantitative analysis of cell nucleus morphology is an important basis for cervical smear examination. Nuclei segmentation is a basic step in the quantitative analysis of nucleus morphology, and the accuracy of segmentation directly affects the analysis results. Therefore, accurate segmentation of cell nuclei is both a basic task and a core link.

Due to the differences in preparation and staining methods of cervical cell smears, the complexity of the background, the diversity and irregularity of cell shapes, and the overlapping of cells, it is difficult to segment cervical cell images. Because the nucleus of cancer cells contains most of the features, the direct segmentation of the nucleus reduces the difficulty of segmentation and simplifies the segmentation process on the premise of fully retaining effective features.

In summary, a new algorithm for cervical cell nucleus segmentation is proposed, using selective search segmentation to obtain ROI; then, according to the overlapping ROI intersection ratio and the final connected components obtained by using the limit corrosion algorithm to process ROI, the ROI containing the nucleus is screened; finally The CV model algorithm was used to finely segment the cervical cell nucleus.

Contents of the invention

The purpose of the present invention is to overcome the deficiencies in the above-mentioned prior art, and provide a novel cervical cell nucleus segmentation algorithm, which optimizes the segmentation effect of cell nucleus cervical cell nucleus, simplifies the segmentation process, and reduces the complexity of segmentation, which can be used for subsequent cervical cell nuclei. Quantitative analysis of cell nucleus morphology provides accurate segmentation results.

In order to achieve the above object, the solution of the present invention is:

A novel cervical cell nucleus segmentation algorithm is characterized in that: comprising the following steps:

① Use selective search segmentation to initially segment an input image into multiple ROIs;

② Calculate the intersection and union ratio IOU between adjacent ROIs, and remove the selected area;

③Using the mathematical morphology limit corrosion algorithm to obtain the final connected components, and by judging the number of the final connected components, screen the ROI containing the nucleus;

④Use the CV model algorithm to segment the ROI obtained after screening to obtain finely segmented nuclei.

The specific method of the above step ① is:

① Segment the original image into multiple small regions r based on the graph theory Felzenszwalb algorithm;

② Calculate the similarity between each area: color similarity S _color (r _i , r _j ), texture similarity S _texture (r _i , r _j ), size similarity S _size (r _i , r _j ) and matching degree S _fill (r _i ,r _j ), and then calculate the total similarity between regions S(r _i ,r _j )＝a ₁ S _color (r _i ,r _j )+a ₂ S _texture (r _i ,r _j )+ a ₃ S _color (r _i ,r _j )+a ₄ S _fill (r _i ,r _j ), where r _i and r _j are the i-th and j-th small regions respectively,

Each similarity is calculated by the following formula:

in with are the k-th dimension in the 75-dimensional color vectors of the i-th and j-th small regions, respectively,

in with are the k-th dimension in the 240-dimensional texture vector of the i-th and j-th small regions,

where size(r _i ), size(r _j ) and size(im) are the sizes of the i-th and j-th small regions and the initial image respectively,

Where size(BB _ij ) refers to the smallest outsourcing area including the i-th and j-th small areas,

③ Perform regional fusion according to the total similarity to obtain the primary ROI.

The specific calculation method of the above step ② is:

The specific calculation method of the above step ④ is based on the following formula:

E(C)＝μL(C)+υ*Area(inside(C))+λ ₁ E _in (C)+λ ₂ E _out (C)

＝μL(C)+υ*Area(inside(C))+λ ₁ ∫ _inside(C) |I(x,y)-c _O | ² dxdy+λ ₂ ∫ _outside(C) |I(x,y )-c _b | ² dxdy

Among them, L(C) represents the length of the curve C; μ represents the length coefficient, and the value depends on the size of the detected target; Area(inside(C)) represents the area of the inner area surrounded by the curve C; υ represents the area parameter; E _in (C) represents the internal energy of the closed curve C; λ ₁ represents the internal energy coefficient; E _out (C) represents the external energy of the closed curve C; λ ₂ represents the internal energy coefficient; The gray level of a pixel; c _O means the average gray level of the inner area; c _b means the average gray level of the outer area.

The present invention has following advantage and positive effect:

1. The present invention preliminarily segments the original image with a complex background into ROIs that may contain nuclei through selective search segmentation, and simplifies the complex multi-objective segmentation task into a single-objective segmentation task, which greatly reduces the complexity of segmentation.

2. The present invention removes the double-selected region by calculating the intersection ratio between adjacent ROIs; identifies the final connected region obtained by limit corrosion, and screens out the ROI that actually contains the nucleus, which reduces the segmentation task to a certain extent and improves the segmentation efficiency .

3. The present invention uses the CV model to finely segment the nucleus, and the segmentation effect is better than that of the traditional cell nucleus segmentation method, and the segmentation process is simple, thereby providing accurate and available data for the subsequent quantitative analysis of the nucleus morphology.

Description of drawings:

Fig. 1 is the image diagram of selecting cervical cells in the database;

Fig. 2 is a schematic diagram of preliminary segmentation ROI;

Figure 3 is an image of the primary ROI;

Figure 4 is a screening ROI image;

Figure 5 is the nucleus and its final connected components;

Figure 6 is the non-nucleus and its final connected components;

Fig. 7 is secondary screening ROI image;

Figure 8 shows some ROIs and their precise segmentation results.

detailed description:

A new cervical cell nucleus segmentation algorithm uses selective search segmentation to initially segment an input image into multiple ROIs that may contain nuclei, which simplifies the multi-target segmentation problem with a complex background into a single-target segmentation problem with a relatively simple background, reducing the The complexity of the segmentation, and then calculate the intersection ratio of overlapping ROIs, and screen out the double-selected areas; then, by judging the number of final connected components obtained by limit corrosion, screen the ROIs containing the nucleus, and finally use the traditional CV model to finely segment the nucleus. Specific steps are as follows:

1. Input the original image. Randomly select cervical cell smear images in the database, the size is 2592*1944, the image format is BMP, and the color mode is RGB, as shown in Figure 1.

2. Preliminary segmentation of ROI. Use selective search segmentation to segment the original image into multiple primary ROIs. The task of Selective search segmentation is to segment the area that may contain the nucleus, that is, to segment a large image with a complex background into a small ROI with a simple background, which greatly reduces the difficulty of segmentation, as shown in Figure 2:

① Segment the original image into multiple small regions r based on the graph theory Felzenszwalb algorithm.

② Calculate the similarity between each area: color similarity S _color (r _i , r _j ), texture similarity S _texture (r _i , r _j ), size similarity S _size (r _i , r _j ) and matching degree S _fill (r _i ,r _j ), and then calculate the total similarity between regions S(r _i ,r _j )＝a ₁ S _color (r _i ,r _j )+a ₂ S _texture (r _i ,r _j )+ a ₃ S _color (r _i ,r _j )+a ₄ S _fill (r _i ,r _j ), where r _i and r _j are the i-th and j-th small regions, respectively.

Each similarity is calculated by the following formula:

in with are the k-th dimension in the 75-dimensional color vectors of the i-th and j-th small regions, respectively.

in with are the k-th dimension in the 240-dimensional texture vectors of the i-th and j-th small regions, respectively.

Among them, size(r _i ), size(r _j ) and size(im) are the sizes of the i-th and j-th small regions and the initial image, respectively.

Among them, size(BB _ij ) refers to the smallest outsourcing area including the i-th and j-th small areas.

③ Perform regional fusion according to the total similarity to obtain the primary ROI.

3. Screen ROI:

①As shown in Figures 3 and 4: According to the size of the ROI and the intersection and union ratio of adjacent areas, remove the selected area. The specific calculation method is:

②As shown in Figures 5, 6, and 7: use the number of final connected components obtained by the mathematical morphology limit corrosion algorithm to identify ROIs, screen out ROIs containing nuclei, reduce the number of ROIs to be segmented, and improve segmentation efficiency.

4. As shown in Figure 8: Apply the CV model algorithm to segment the ROI obtained after screening to obtain fine cell nucleus segmentation results. The specific method is:

E(C)＝μL(C)+υ*Area(inside(C))+λ ₁ E _in (C)+λ ₂ E _out (C)

＝μL(C)+υ*Area(inside(C))+λ ₁ ∫ _inside(C) |I(x,y)-c _O | ² dxdy+λ ₂ ∫ _outside(C) |I(x,y )-c _b | ² dxdy

Among them, L(C) represents the length of curve C; μ represents the length coefficient, and its value depends on the size of the detected object; Area(inside(C)) represents the area of the inner area surrounded by curve C; υ represents the area parameter. E _in (C) represents the internal energy of the closed curve C; λ ₁ represents the internal energy coefficient; E _out (C) represents the external energy of the closed curve C; λ ₂ represents the internal energy coefficient; The gray level of a pixel; c _O means the average gray level of the inner area; c _b means the average gray level of the outer area.

It should be noted that the above descriptions are only preferred embodiments of the present invention, and are only for explaining the present invention, and are not intended to limit the patent scope of the present invention. Modifications that are only obvious and belong to the technical concept of the present invention are also within the protection scope of the present invention.

Claims (4)

1. a novel cervical cell nucleus segmentation algorithm, is characterized in that: comprise the steps: ① Use selective search segmentation to initially segment an input image into multiple ROIs; ② Calculate the intersection and union ratio IOU between adjacent ROIs, and remove the selected area; ③Using the mathematical morphology limit corrosion algorithm to obtain the final connected components, and by judging the number of the final connected components, screen the ROI containing the nucleus; ④Use the CV model algorithm to segment the ROI obtained after screening to obtain finely segmented nuclei. 2. a kind of novel cervical cell nucleus segmentation algorithm according to claim 1 is characterized in that: the concrete method of above-mentioned step 1. is: ① Segment the original image into multiple small regions r based on the graph theory Felzenszwalb algorithm; ② Calculate the similarity between each area: color similarity S _color (r _i , r _j ), texture similarity S _texture (r _i , r _j ), size similarity S _size (r _i , r _j ) and matching degree S _fill (r _i ,r _j ), and then calculate the total similarity between regions S(r _i ,r _j )＝a ₁ S _color (r _i ,r _j )+a ₂ S _texture (r _i ,r _j )+ a ₃ S _color (r _i ,r _j )+a ₄ S _fill (r _i ,r _j ), where r _i and r _j are the i-th and j-th small regions respectively, Each similarity is calculated by the following formula: <mrow><msub><mi>S</mi><mrow><mi>c</mi><mi>o</mi><mi>l</mi><mi>o</mi><mi>r</mi></mrow></msub><mrow><mo>(</mo><msub><mi>r</mi><mi>i</mi></msub><mo>,</mo><msub><mi>r</mi><mi>j</mi></msub><mo>)</mo></mrow><mo>=</mo><munderover><mo>&amp;Sigma;</mo><mrow><mi>k</mi><mo>=</mo><mn>1</mn></mrow><mi>n</mi></munderover><mi>m</mi><mi>i</mi><mi>n</mi><mrow><mo>(</mo><msubsup><mi>c</mi><mi>i</mi><mi>k</mi></msubsup><mo>,</mo><msubsup><mi>c</mi><mi>j</mi><mi>k</mi></msubsup><mo>)</mo></mrow></mrow> in with are the k-th dimension in the 75-dimensional color vectors of the i-th and j-th small regions, respectively, <mrow><msub><mi>S</mi><mrow><mi>t</mi><mi>e</mi><mi>x</mi><mi>t</mi><mi>u</mi><mi>r</mi><mi>e</mi></mrow></msub><mrow><mo>(</mo><msub><mi>r</mi><mi>i</mi></msub><mo>,</mo><msub><mi>r</mi><mi>j</mi></msub><mo>)</mo></mrow><mo>=</mo><munderover><mo>&amp;Sigma;</mo><mrow><mi>k</mi><mo>=</mo><mn>1</mn></mrow><mi>n</mi></munderover><mi>m</mi><mi>i</mi><mi>n</mi><mrow><mo>(</mo><msubsup><mi>t</mi><mi>i</mi><mi>k</mi></msubsup><mo>,</mo><msubsup><mi>t</mi><mi>j</mi><mi>k</mi></msubsup><mo>)</mo></mrow></mrow> in with are the k-th dimension in the 240-dimensional texture vector of the i-th and j-th small regions, <mrow><msub><mi>S</mi><mrow><mi>s</mi><mi>i</mi><mi>z</mi><mi>e</mi></mrow></msub><mrow><mo>(</mo><msub><mi>r</mi><mi>i</mi></msub><mo>,</mo><msub><mi>r</mi><mi>j</mi></msub><mo>)</mo></mrow><mo>=</mo><mn>1</mn><mo>-</mo><mfrac><mrow><mi>s</mi><mi>i</mi><mi>z</mi><mi>e</mi><mrow><mo>(</mo><msub><mi>r</mi><mi>i</mi></msub><mo>)</mo></mrow><mo>+</mo><mi>s</mi><mi>i</mi><mi>z</mi><mi>e</mi><mrow><mo>(</mo><msub><mi>r</mi><mi>j</mi></msub><mo>)</mo></mrow></mrow><mrow><mi>s</mi><mi>i</mi><mi>z</mi><mi>e</mi><mrow><mo>(</mo><mi>i</mi><mi>m</mi><mo>)</mo>mo></mrow></mrow></mfrac></mrow> where size(r _i ), size(r _j ) and size(im) are the sizes of the i-th and j-th small regions and the initial image respectively, <mrow><msub><mi>S</mi><mrow><mi>f</mi><mi>i</mi><mi>l</mi><mi>l</mi></mrow></msub><mrow><mo>(</mo><msub><mi>r</mi><mi>i</mi></msub><mo>,</mo><msub><mi>r</mi><mi>j</mi></msub><mo>)</mo></mrow><mo>=</mo><mn>1</mn><mo>-</mo><mfrac><mrow><mi>s</mi><mi>i</mi><mi>z</mi><mi>e</mi><mrow><mo>(</mo><msub><mi>BB</mi><mrow><mi>i</mi><mi>j</mi></mrow></msub><mo>)</mo></mrow><mo>-</mo><mi>s</mi><mi>i</mi><mi>z</mi><mi>e</mi><mrow><mo>(</mo><msub><mi>r</mi><mi>i</mi></msub><mo>)</mo></mrow><mo>-</mo><mi>s</mi><mi>i</mi><mi>z</mi><mi>e</mi><mrow><mo>(</mo><msub><mi>r</mi><mi>j</mi></msub><mo>)</mo></mrow></mrow><mrow><mi>s</mi><mi>i</mi><mi>z</mi><mi>e</mi><mrow><mo>(</mo><mi>i</mi><mi>m</mi><mo>)</mo></mrow></mrow></mfrac></mrow> Where size(BB _ij ) refers to the smallest outsourcing area including the i-th and j-th small areas, ③ Perform regional fusion according to the total similarity to obtain the primary ROI. 3. a kind of novel cervical cell nucleus segmentation algorithm according to claim 1, is characterized in that: the concrete computing method of above-mentioned step 2. is: <mrow><mi>I</mi><mi>O</mi><mi>U</mi><mo>=</mo><mfrac><mrow><mi>a</mi><mi>r</mi><mi>e</mi><mi>a</mi><mrow><mo>(</mo><msub><mi>r</mi><mi>i</mi></msub><mo>&amp;cap;</mo><msub><mi>r</mi><mi>j</mi></msub><mo>)</mo></mrow></mrow><mrow><mi>a</mi><mi>r</mi><mi>e</mi><mi>a</mi><mrow><mo>(</mo><msub><mi>r</mi><mi>i</mi></msub><mo>&amp;cup;</mo><msub><mi>r</mi><mi>j</mi></msub><mo>)</mo></mrow></mrow></mfrac><mo>=</mo><mfrac><mrow><mi>a</mi><mi>r</mi><mi>e</mi><mi>a</mi><mrow><mo>(</mo><msub><mi>r</mi><mi>i</mi></msub><mo>&amp;cap;</mo><msub><mi>r</mi><mi>j</mi></msub><mo>)</mo></mrow></mrow><mrow><mi>a</mi><mi>r</mi><mi>e</mi><mi>a</mi><mrow><mo>(</mo><msub><mi>r</mi><mi>i</mi></msub><mo>)</mo></mrow><mo>+</mo><mi>a</mi><mi>r</mi><mi>e</mi><mi>a</mi><mrow><mo>(</mo><msub><mi>r</mi><mi>j</mi></msub><mo>)</mo></mrow><mo>-</mo><mi>a</mi><mi>r</mi><mi>e</mi><mi>a</mi><mrow><mo>(</mo><msub><mi>r</mi><mi>i</mi></msub><mo>&amp;cap;</mo><msub><mi>r</mi><mi>j</mi></msub><mo>)</mo></mrow></mrow></mfrac><mo>.</mo></mrow> 4. a kind of novel cervical cell nucleus segmentation algorithm according to claim 1, is characterized in that: the concrete computing method of above-mentioned step 4. is based on following formula: E(C)＝μL(C)+υ*Area(inside(C))+λ ₁ E _in (C)+λ ₂ E _out (C) ＝μL(C)+υ*Area(inside(C))+λ ₁ ∫ _inside(C) |I(x,y)-c _O | ² dxdy+λ ₂ ∫ _outside(C) |I(x,y )-c _b | ² dxdy Among them, L(C) represents the length of the curve C; μ represents the length coefficient, and the value depends on the size of the detected target; Area(inside(C)) represents the area of the inner area surrounded by the curve C; υ represents the area parameter; E _in (C) represents the internal energy of the closed curve C; λ ₁ represents the internal energy coefficient; E _out (C) represents the external energy of the closed curve C; λ ₂ represents the internal energy coefficient; The gray level of a pixel; c _O means the average gray level of the inner area; c _b means the average gray level of the outer area.