CN100433063C - Method for correcting measurement offset caused by tissue section image segmentation - Google Patents

Abstract

The invention discloses biopsy image segmentation measurement error correction method which consists of two parts: 1) target counts result correction: Totally there are two counts in the image analysis. The scope of the first one does not include the analysis image of the boundary line. Get the count C1, and then conduct a count including the boundary line of image to get count C2, calculate the final value C; 2) the measuring result correction: the image acquisition resolution separate line width can induce measurement loss, firstly do preliminary image analysis to get perimeter, long axis and short axis and area four parameters, and then do losses correction according to the image Acquisition resolution and segmentation line width and their geometric relationships. The invention uses existing technology and mathematical methods to derive correction method, the count values, and the impact of the corresponding parameters of objectives by the division line and acquisition resolution is significantly lowered without statistically significant difference.

Description

The bearing calibration of measurement offset caused by tissue section image segmentation

Technical field

The present invention relates to the micro image analysis field in the signal Processing, relate in particular to a kind of bearing calibration of measurement offset caused by tissue section image segmentation.

Background technology

Image analysis technology is widely used in medical science and field of biology, has great importance for animal morphology research and veterinary clinical pathology research.The collection of sectioning image and analysis are key areas of medical science and biological study, this technology has been widely used in the analysis that enzyme immunohistochemistry, immunohistochemistry and conventional organization are learned section at present.Gather and all will pass through image segmentation before the digital picture that obtains is analyzed, evaluating objects is extracted from image background or identifies, this is to carry out one of critical step in the image analysis process, also is the emphasis and the difficult point of histology and pathology graphical analysis.

Image segmentation mainly can be undertaken by cutting apart automatically and manually cutting apart dual mode, the former mainly is based on the uncontinuity and the similarity of image adjacent pixel values, for some have relatively simple for structure, evaluating objects and the image background contrast bigger, the image of feature such as at regular intervals has segmentation ability preferably between target.The latter is that the cut-off rule that manually utilizes various edit tools to add certain width between adjacent analysis of cells is finished.Yet histology and pathological sectioning image are because the singularity of this body structure, the reason of aspects such as the diversity of dyeing property and contrast, utilize the automatic dividing function of existing business software to be difficult to effectively to carry out target and cut apart, therefore manually cut apart more and have superiority.In manually cutting apart, the acquisition resolution of sectioning image and cut-off rule width are two principal elements of impact analysis result.

Manually cut apart and also can adopt several different methods to finish, can utilize the erasing rubber (erase) in the Computer Image Processing software, path (path) and handwritten instruments (freehand) are carried out, its fundamental purpose is to form a septal line between evaluating objects, theoretically, the raw information of narrow more favourable more preservation of this line and reproduced image, but in fact, it is subjected to software inhouse setting, acquisition resolution, digitized process, all multifactor influences such as image analysis processing, cut-off rule less than 3 pixels can't carry out effective image segmentation in existing business software, certainly also just can't carry out image analysis processing, promptly enable to analyze, error rate is also very high.Present most business software all passes through line ball or the not count results of the whole evaluating objects of line ball of counting in the picture, this can cause bigger error to a certain extent, the author thinks should be with reference to the method for counting of cell counting count board, the evaluating objects of line ball is only counted two adjacent edges, and the line ball target of omitting two other edge, half of the difference of the inclusive count value of line ball target and the count value that do not count added the count value that the line ball target does not count, and the summation that draws at last is as the final count results of this picture.Through multiple authentication, adopt the method for counting among the present invention to have more rationality and reliability.

Acquisition resolution for image, yes is the bigger the better, yet for various reasons, often acquisition resolution is lower for the hardware of the existing image analysis system of most research institutions, the image of low resolution in can only obtaining, and along with the raising of resolution, image analysis software increases substantially the requirement of computer process ability, common computing machine can't batch processing, and this has just limited promoting the use of of software.

The existing more research of the graphical analysis of at present relevant biological specimen, be widely used in pig, the myofibrillar measurement of chicken animal at animal and veterinary field image analysis technology, but the measured data of different researchers have very big-difference, the author has got rid of reasons such as biological diversity own and species variation, the method, particularly dividing method that depend on the graphical analysis pre-treatment on the widely different degree of measurement data.Given this,, obtain more accurate result for the image of low resolution collection in making full use of and common computing machine,

Summary of the invention

The purpose of this invention is to provide a kind of bearing calibration of measurement offset caused by tissue section image segmentation.

It comprises following two parts:

1) object count result's correction:

In image analysis process, carry out twice count measurement altogether, the scope of counting does not comprise the evaluating objects on the image boundary line for the first time, obtain objective count value (C1), and then once comprise the counting of the evaluating objects on the image boundary line, obtain objective count value (C2), proofread and correct, obtain final count results (C) with following formula:

$C=\mathrm{<figure-callout\; id="1"\; label="C"\; filenames="C20061015559500101.png"\; state="\{\{state\}\}">C</figure-callout>}1+\left(\frac{C2-\mathrm{<figure-callout\; id="1"\; label="C"\; filenames="C20061015559500101.png"\; state="\{\{state\}\}">C</figure-callout>}1}{2}\right)---\left(1\right)$

2) morphometry result's correction:

In image analysis process, morphological index and measurement thereof are mainly by girth, major axis, minor axis, 4 parameters of area are represented, because the height of image acquisition resolution (R) and the size of cut-off rule width (W) can cause morphometry loss in various degree, carry out preliminary graphical analysis earlier, obtain girth (P1), major axis (Ma1), minor axis (Mi1) and 4 parameters of area (S1), and then measure loss according to the geometric relationship of the width of image acquisition resolution and cut-off rule and 4 measured value of parameters and proofread and correct, this measurement loss is with following formula (2), (3), (4), (5) proofread and correct:

Girth (P) is measured loss and can be adopted following formula to proofread and correct:

$P=\mathrm{<figure-callout\; id="1"\; label="P"\; filenames="C20061015559500101.png"\; state="\{\{state\}\}">P</figure-callout>}1+\frac{W\&times;\mathrm{\&pi;}}{R}---\left(2\right)$

Major axis (Ma) is measured loss and is adopted following formula to proofread and correct:

$\mathrm{Ma}=\mathrm{<figure-callout\; id="1"\; label="Ma"\; filenames="C20061015559500101.png"\; state="\{\{state\}\}">Ma</figure-callout>}1+\frac{W}{R}---\left(3\right)$

Minor axis (Mi) is measured loss and is adopted following formula to proofread and correct:

$\mathrm{Mi}=\mathrm{<figure-callout\; id="1"\; label="Mi"\; filenames="C20061015559500101.png"\; state="\{\{state\}\}">Mi</figure-callout>}1+\frac{W}{R}---\left(4\right)$

Area (S) is measured loss and is adopted following formula to proofread and correct:

$S=\mathrm{<figure-callout\; id="1"\; label="S"\; filenames="C20061015559500101.png"\; state="\{\{state\}\}">S</figure-callout>}1+\frac{1}{2}\&times;(\frac{P\&times;W}{R}+\mathrm{\&pi;}\&times;{\left(\frac{W}{R}\right)}^2)---\left(5\right)$

The bearing calibration that the present invention utilizes prior art means and algorithmic approach to derive, parameters such as the count value after the correction, the area of evaluating objects, girth, major axis, minor axis are obviously reduced by the influence of cut-off rule and acquisition resolution, no statistically-significant difference.The The above results of proofreading and correct is applied to circularity, form factor, compactness, the isoparametric calculating of axial ratio, correspondingly can improves precision of analysis effectively.Be applied to the software development in morphology micro-image and biology correlation analysis field, improved precision of analysis effectively.Be applied to pathology micro image analysis field, can effectively improve the reliability of analysis result, reduce misdiagnosis rate.Can be applied to the abiology art of image analysis, assist remedial frames to cut apart caused measurement loss and deviation.

Description of drawings

Fig. 1 is area, girth and diameter measurement loss and the calibration result figure thereof when showing the 1200M sampling;

Fig. 2 shows the area measurement value of different sampling resolutions and the distribution design sketch after the correction thereof.

Embodiment

As everyone knows, when the cut-off rule width was x, according to the girth computing formula of circle, the girth of the object construction after cutting apart (Po) had reduced

Individual unit, historical facts or anecdotes border girth can be with recording girth

Estimation.Also be formula (3), space wastage can be the target girth with length, and width is

Rectangular area estimate that real area equals $\mathrm{Sm}+\frac{x}{2}\&times;(\mathrm{Pm}+2\&times;\mathrm{\&pi;}\&times;\left(\frac{x}{2}\right)),$ Also be formula (4).Utilize the area after proofreading and correct to obtain form factor, the just more approaching reality of the result who obtains like this.The results of analysis of variance shows that after the correction, parameters such as area, girth and diameter are obviously reduced by the influence of cut-off rule and acquisition resolution, and at present common software does not all have this correction, so measurement data and actual result have certain deviation.Statistical results show is utilized updating formula (3) that this paper proposes and (4) to eliminate or to reduce by cut-off rule and resolution and is made deviation.

Microscope is calibrated through micrometer, the automatic exposure images acquired, be scaled DPU (pixels per micrometer) according to acquisition resolution (R), the gained photo stores with the Tiff form, and utilize the corresponding tool of image processing software, background is set for white, utilizes different in width (W) line, object construction is carried out line respectively cut apart.

Graphical analysis; The imagery exploitation ImageJ (NIH shareware) of storage carries out graphical analysis, analytic process is roughly as follows: read in software, software is converted into gray scale (8) image, setting threshold, calibration is carried out graphical analysis and is obtained parameters such as area, girth, major axis, minor axis, circularity.

Here the loss of cutting apart of indication specially refers under different acquisition resolution, because of area, the girth that adopts artificial cut-off rule to cut the evaluating objects that produces between two adjacent evaluating objects reduces.

The correction of the method for object count: also promptly carry out when shape measure the line ball target and do not add up, record object count (C1), each evaluating objects area (S1), girth (P1), major axis (Ma1) and minor axis parameters such as (Mi1), after finishing shape measure, once comprise the measurement of line ball target again, record object count C2 as a result.The available following formula of final counting calculates to be proofreaied and correct to obtain more accurate reasonably final counting C:

$C=\mathrm{<figure-callout\; id="1"\; label="C"\; filenames="C20061015559500101.png"\; state="\{\{state\}\}">C</figure-callout>}1+\left(\frac{C2-\mathrm{<figure-callout\; id="1"\; label="C"\; filenames="C20061015559500101.png"\; state="\{\{state\}\}">C</figure-callout>}1}{2}\right)---\left(1\right)$

Girth (P) loss adopts following formula to proofread and correct:

$P=\mathrm{<figure-callout\; id="1"\; label="P"\; filenames="C20061015559500101.png"\; state="\{\{state\}\}">P</figure-callout>}1+\frac{W\&times;\mathrm{\&pi;}}{R}---\left(2\right)$

Major axis (Ma) loss adopts following formula to proofread and correct:

$\mathrm{Ma}=\mathrm{<figure-callout\; id="1"\; label="Ma"\; filenames="C20061015559500101.png"\; state="\{\{state\}\}">Ma</figure-callout>}1+\frac{W}{R}---\left(3\right)$

Minor axis (Mi) loss adopts following formula to proofread and correct:

$\mathrm{Mi}=\mathrm{Mi}1+\frac{W}{R}---\left(4\right)$

Area (S) loss adopts following formula to proofread and correct:

$S=\mathrm{<figure-callout\; id="1"\; label="S"\; filenames="C20061015559500101.png"\; state="\{\{state\}\}">S</figure-callout>}1+\frac{1}{2}\&times;(\frac{P\&times;W}{R}+\mathrm{\&pi;}\&times;{\left(\frac{W}{R}\right)}^2)---\left(5\right)$

And utilize area, girth, major axis after proofreading and correct to calculate equivalent diameter (ED), form factor (FI), circularity (CI), compactness and axial ratio parameters such as (AP) again.

$\mathrm{ED}=2\&times;\sqrt{\frac{S}{\mathrm{\&pi;}}}---\left(6\right)$

$\mathrm{FI}=\frac{4\mathrm{\&pi;}\&times;S}{p^2}---\left(7\right)$

$\mathrm{CI}=\frac{4\&times;S}{\mathrm{\&pi;}\&times;\mathrm{Ma}}---\left(8\right)$

$\mathrm{CP}=\sqrt{\frac{4\&times;S}{\mathrm{\&pi;}\&times;{\mathrm{Ma}}^2}}---\left(9\right)$

Embodiment:

The correction of the myofibrillar micro image analysis outcome measurement loss of pig

1 materials and methods

1.1 draw materials and the preparation (summary) of cutting into slices

1.2 image acquisition and processing

Observe under 20 times of object lens with the Olympus microscope, light source is through blue color filter filtering, automatic exposure, to the same visual field with 1,000,000 pixels (3.75DPU (dot per micrometer, DPU; 1020 * 768) (be called low resolution), 3,000,000 pixel (7.5DPU, 2040 * 1536) (be called intermediate-resolution) and 1,200 ten thousand pixel (15.0DPU, 4080 * 4072) (being called high resolving power) takes pictures respectively, the gained photo stores with the Tiff form, and select the Eraser Tool of PHOTOSHOP image processing software for use, background is set is white, hardness is 100%, width is respectively 3,4,5,6,7,8,9,10,15,20 pixels, muscle fibre is carried out line respectively cut apart, the image after cutting apart is deposited separately.

1.3 graphical analysis

The imagery exploitation ImageJ (NIH shareware) of storage carries out graphical analysis, analytic process is roughly as follows: read in software, software is converted into gray scale (8) image, setting threshold, calibration is carried out graphical analysis and is obtained parameters such as area, girth, major axis, minor axis, circularity.The fiber counting is with reference to the method for blood count, the line ball fiber is not added up when promptly carrying out above-mentioned shape measure yet, after finishing shape measure, again once comprise the measurement of line ball fiber, final fiber number with the difference of twice counting half with do not comprise that the line ball fiber records the fiber number and be as the criterion.For verifying the actual fibers number of each photo, also carry out artificial counting, the line ball cell is only counted the top and the left side.

1.4 cut apart the correction of loss

As referred to hereinly cut apart loss and specially refer to because of adopting artificial cut-off rule between two adjacent evaluating objects, to cut the area minimizing of the evaluating objects that produces.Adopt following formula to proofread and correct because of cutting apart the girth and the space wastage that produce in the split image analytic process:

Utilize girth and the area proofreaied and correct to obtain parameters such as corresponding equivalent diameter, form factor and circularity.

1.5 data statistics and analysis

To gather pixel and cut-off rule width is influence factor, the gained data have been carried out the evaluation that influences of measurement parameter respectively with the variance analysis module of SAS software, measurement result is expressed as average ± standard deviation, and utilize the multiple range method of Deng Kenshi to carry out multiple ratio, P＜0.05 expression significant difference.

2 results

2.1 image acquisition resolution and the manual influence of cutting apart count results

Method of counting with reference to cell counting count board, the cell of line ball is only counted two adjacent edges, and omit the line ball cell at two other edge, with half and the count value sum that do not count final count results as this picture of the inclusive count value of line ball cell with the difference of the count value that does not count.Empirical tests, this method calculate the fiber number be 270 (concrete numerical value see Table), similar to 273 of the numerical value that artificial counting obtains, and utilize the computing machine counting time line ball fiber to be counted is 321 (averages), do not count is 219 (averages), bigger discrepancy is all arranged, this shows, the method for counting that adopts this paper to propose has more rationality and reliability.

2.2 the cut-off rule width is cut apart the impact analysis of the measurement result accuracy after the loss to correction

The analysis result of the image of different resolution compared the analysis showed that low-resolution image has bigger measurement loss, and measure loss along with the increase of link width and increase (Fig. 1) gradually.After utilizing formula provided by the invention that area is proofreaied and correct, recomputate parameters such as trying to achieve fibre diameter and form factor, utilize that SAS is long-pending to fiber cross section with regard to the cut-off rule width, the influence of parameter analyses such as equivalent diameter and form factor lists in table 1, after overcorrect, cut-off rule width and image acquisition resolution obviously reduce (Fig. 2) to the influence of parameter analysis results such as the area of target fibers, equivalent diameter, all do not have significant difference between processing.

Table 1 cut-off rule width is cut apart the isoparametric influence of fiber area, equivalent diameter and form factor after the loss to correction

^*The result is expressed as average ± standard deviation, shoulder mark expression Deng Kenshi multipole difference The result of multiple comparisons, and different alphabetical a represent significant difference between b and the c, same letter represents that difference is not remarkable.

Claims (1)

1. the bearing calibration of a measurement offset caused by tissue section image segmentation is characterized in that comprising following two parts:

1) object count result's correction:

In image analysis process, carry out twice count measurement altogether, the scope of counting does not comprise the evaluating objects on the image boundary line for the first time, obtain objective count value C1, and then once comprise the counting of the evaluating objects on the image boundary line, obtain objective count value C2, proofread and correct, obtain final count results C with following formula:

$C=C1+\left(\frac{C2-C1}{2}\right)---\left(1\right)$

2) morphometry result's correction:

In image analysis process, morphological index and measurement thereof comprise girth, major axis, minor axis and 4 parameters of area, because the height of image acquisition resolution R and the size of cut-off rule width W can cause morphometry loss in various degree, carry out preliminary graphical analysis earlier, obtain girth P1, major axis Mal, minor axis Mil and 4 parameters of area S1, and then measure loss according to the geometric relationship of the width of image acquisition resolution and cut-off rule and 4 measured value of parameters and proofread and correct, this measurement loss is proofreaied and correct with following formula (2), (3), (4), (5):

Girth P after the correction measures loss and adopts following formula to proofread and correct:

$P=P1+\frac{W\&times;\mathrm{\&pi;}}{R}---\left(2\right)$

Major axis Ma after the correction measures loss and adopts following formula to proofread and correct:

$\mathrm{Ma}=\mathrm{Ma}1+\frac{W}{R}---\left(3\right)$

Minor axis Mi after the correction measures loss and adopts following formula to proofread and correct:

$\mathrm{Mi}=\mathrm{Mi}1+\frac{W}{R}---\left(4\right)$

Area S after the correction measures loss and adopts following formula to proofread and correct:

$S=S1+\frac{1}{2}\&times;(\frac{P\&times;W}{R}+\mathrm{\&pi;}\&times;{\left(\frac{W}{R}\right)}^2)---\left(5\right).$

Images