CN115731203A

CN115731203A - Cataract image identification method and device, computer equipment and readable storage medium

Info

Publication number: CN115731203A
Application number: CN202211496703.9A
Authority: CN
Inventors: 刘江; 肖尊杰; 东田理沙; 章晓庆; 巫晓
Original assignee: Southern University of Science and Technology
Current assignee: Southern University of Science and Technology
Priority date: 2022-11-25
Filing date: 2022-11-25
Publication date: 2023-03-03

Abstract

The application discloses a cataract image identification method, a cataract image identification device, computer equipment and a readable storage medium, and relates to the field of computer vision. The method includes acquiring a cataract data set, the cataract data set including a plurality of cataract images; carrying out image segmentation processing on each cataract image to obtain a plurality of cataract areas; inputting each cataract area into three preset image feature statistical algorithms respectively to perform feature statistical processing to obtain a first image feature set, a second image feature set and a third image feature set respectively, and merging the three image feature sets to serve as an initial image feature set; fusing a result obtained by inputting the initial image feature set into a preset machine learning model and a result obtained by inputting the initial image feature set into a preset deep neural network model to obtain a target prediction result; and determining the category of the cataract image according to the target prediction result and a preset cataract grade. The embodiment of the application can realize one-stop cataract image identification.

Description

Cataract image identification method and device, computer equipment and readable storage medium

Technical Field

The present application relates to the field of computer vision technologies, and in particular, to a cataract image recognition method, an apparatus, a computer device, and a readable storage medium.

Background

Cataracts are ophthalmic diseases that are primarily blinding and cause visual impairment, and early intervention and cataract surgery can effectively improve the vision and quality of life of patients. Clinical cataract symptoms are manifested as clouding of the lens region, and the cataract can be classified into nuclear cataract, cortical cataract and cystic cataract according to the position of the clouding symptom, wherein the nuclear cataract is manifested as clouding of the lens nuclear region, the cortical cataract is manifested as clouding of the cortical region, and the cystic cataract is manifested as clouding of the cystic region. In order to assist ophthalmologists in cataract diagnosis, researchers develop cataract auxiliary diagnosis and screening systems, however, most of the existing cataract auxiliary diagnosis and screening systems aim at single type cataract image identification, do not consider the actual cataract clinical diagnosis requirement, and cannot identify multiple cataract types at the same time.

Disclosure of Invention

The application aims at solving the problems of the prior art at least to a certain extent, and provides a cataract image identification method, a cataract image identification device, computer equipment and a readable storage medium, which can efficiently and accurately identify various types of cataracts and realize one-stop intelligent identification of cataract images.

The technical scheme of the embodiment of the application is as follows:

in a first aspect, the present application provides a cataract image identification method, including:

acquiring a cataract data set, the cataract data set comprising a plurality of cataract images;

performing image segmentation processing on each cataract image to obtain a plurality of cataract areas;

respectively inputting each cataract area into a preset image intensity algorithm, a preset histogram algorithm and a preset texture feature algorithm for feature statistical processing to obtain a first image feature set corresponding to the image intensity algorithm, a second image feature set corresponding to the histogram algorithm and a third image feature set corresponding to the texture feature algorithm, and merging the first image feature set, the second image feature set and the third image feature set to serve as initial image feature sets;

inputting the initial image feature set into a preset machine learning model for prediction processing to obtain a first initial prediction result, inputting the initial image feature set into a preset deep neural network model for prediction processing to obtain a second initial prediction result, and obtaining a target prediction result according to the first initial prediction result and the second initial prediction result;

and determining the category of the cataract image according to the target prediction result and a preset cataract grade.

According to some embodiments of the application, after said merging the first, second and third image feature sets as an initial image feature set, the method further comprises:

performing feature importance analysis on the initial image feature set to obtain a target image feature set;

inputting the initial image feature set into a preset machine learning model for prediction processing to obtain a first initial prediction result, wherein the method comprises the following steps:

inputting the target image feature set into a preset machine learning model for prediction processing to obtain the first initial prediction result;

inputting the initial image feature set into a preset deep neural network model for prediction processing to obtain a second initial prediction result, wherein the method comprises the following steps:

and inputting the target image feature set into a preset deep neural network model for prediction processing to obtain the second initial prediction result.

According to some embodiments of the present application, obtaining a target prediction result according to the first initial prediction result and the second initial prediction result comprises:

and carrying out averaging processing on the first initial prediction result and the second initial prediction result to obtain the target prediction result.

According to some embodiments of the present application, the inputting the initial image feature set into a preset machine learning model for prediction processing to obtain a first initial prediction result includes:

inputting the initial image feature set into the machine learning model for feature extraction to obtain a first extraction feature set;

visualizing the first extraction feature set to obtain a first feature visualization graph, and displaying the first feature visualization graph;

performing prediction calculation on the features in the first feature visualization graph to obtain a first initial prediction result;

inputting the initial image feature set into the deep neural network model for feature extraction to obtain a second extracted feature set;

visualizing the second extracted feature set to obtain a second feature visualization graph, and displaying the second feature visualization graph;

and performing prediction calculation on the features in the second feature visualization graph to obtain the second initial prediction result.

According to some embodiments of the present application, the performing image segmentation processing on each of the cataract images to obtain a plurality of cataract areas includes:

carrying out crystal segmentation processing on each cataract image to obtain crystal region characteristics;

and performing image processing on the crystal region characteristics to obtain a plurality of cataract regions.

According to some embodiments of the application, the cataract images include a nuclear image, a cortical image, and a cystic image;

the step of performing crystal segmentation processing on each cataract image to obtain crystal region characteristics comprises:

performing crystal segmentation processing on each nuclear image to obtain a plurality of nuclear crystal areas;

performing crystal segmentation processing on each cortical image to obtain a plurality of cortical crystal regions;

and carrying out crystal segmentation processing on each cystic image to obtain a plurality of cystic crystal regions.

According to some embodiments of the present application, the image processing the characteristics of the crystal region to obtain a plurality of cataract regions includes:

performing region division processing on each nuclear crystal region to obtain an overall nuclear region, an upper semi-nuclear region and a lower semi-nuclear region;

performing region division treatment on each cortical crystal region to obtain an integral cortical region, an upper semi-cortical region and a lower semi-cortical region;

performing area division processing on each cystic crystal area to obtain an integral cystic area, an upper half cystic area and a lower half cystic area;

taking the nuclear region, the superior hemi-nuclear region, the inferior hemi-nuclear region, the global cortical region, the superior hemi-cortical region, the inferior hemi-cortical region, the global cystic region, the superior hemi-cystic region, and the inferior hemi-cystic region as the cataract region.

In a second aspect, the present application provides a cataract image recognition apparatus, the apparatus comprising:

a data acquisition module for acquiring a cataract data set comprising a plurality of cataract images;

the image processing module is used for carrying out image segmentation processing on each cataract image to obtain a plurality of cataract areas;

the feature statistical module is used for inputting each cataract area into a preset image intensity algorithm, a preset histogram algorithm and a preset texture feature algorithm respectively for feature statistical processing to obtain a first image feature set corresponding to the image intensity algorithm, a second image feature set corresponding to the histogram algorithm and a third image feature set corresponding to the texture feature algorithm, and merging the first image feature set, the second image feature set and the third image feature set to serve as initial image feature sets;

the result prediction module is used for inputting the initial image feature set into a preset machine learning model for prediction processing to obtain a first initial prediction result, inputting the initial image feature set into a preset deep neural network model for prediction processing to obtain a second initial prediction result, and obtaining a target prediction result according to the first initial prediction result and the second initial prediction result;

and the suggestion output module is used for determining the category of the cataract image according to the target prediction result and a preset cataract grade.

In a third aspect, the present application provides a computer device comprising a memory and a processor, the memory having stored therein computer-readable instructions which, when executed by one or more of the processors, cause the one or more processors to perform the steps of any one of the methods described above in the first aspect.

In a fourth aspect, the present application also provides a computer-readable storage medium readable by a processor, the storage medium storing computer instructions which, when executed by one or more processors, cause the one or more processors to perform the steps of any of the methods described above in the first aspect.

The technical scheme provided by the embodiment of the application has the following beneficial effects:

the embodiment of the application provides a cataract image identification method, a cataract image identification device, computer equipment and a readable storage medium, wherein the cataract image identification method comprises the steps of firstly obtaining a cataract data set, wherein the cataract data set comprises a plurality of cataract images; performing image segmentation processing on each cataract image to obtain a plurality of cataract areas, and extracting different areas of various cataract images; then, respectively inputting each cataract area into a preset image intensity algorithm, a preset histogram algorithm and a preset texture feature algorithm for feature statistical processing to obtain a first image feature set corresponding to the image intensity algorithm, a second image feature set corresponding to the histogram algorithm and a third image feature set corresponding to the texture feature algorithm, merging the first image feature set, the second image feature set and the third image feature set to obtain an initial image feature set, and obtaining the features corresponding to richer cataract images; inputting the initial image feature set into a preset machine learning model for prediction processing to obtain a first initial prediction result, inputting the initial image feature set into a preset deep neural network model for prediction processing to obtain a second initial prediction result, and obtaining a target prediction result according to the first initial prediction result and the second initial prediction result, so that the accuracy of the prediction result is improved; and determining the category of the cataract image according to the target prediction result and a preset cataract grade. Compared with the prior art, the auxiliary diagnosis system for single type cataract identification has the advantages that the cataract images are respectively input into the image intensity algorithm, the histogram algorithm and the texture feature algorithm for feature statistics, various feature statistical modes are provided, the corresponding features of the abundant cataract images can be obtained, the machine learning model and the deep neural network model are used for result prediction, the prediction accuracy is improved, the type of the cataract images is finally determined, one-stop intelligent identification of the cataract images is realized through image processing, result prediction and type identification, and various types of cataracts can be efficiently and accurately identified.

Additional aspects and advantages of the present application will be set forth in part in the description which follows and, in part, will be obvious from the description, or may be learned by practice of the present application.

Drawings

The above and/or additional aspects and advantages of the present application will become apparent and readily appreciated from the following description of the embodiments, taken in conjunction with the accompanying drawings of which:

fig. 1 is a schematic flow chart of a cataract image identification method according to an embodiment of the present application;

fig. 2 is a schematic flow chart of a cataract image identification method according to another embodiment of the present application;

FIG. 3 is a flow chart illustrating sub-steps of step S400 in FIG. 1;

FIG. 4 is a flow chart illustrating a sub-step of step S200 in FIG. 1;

FIG. 5 is a flow chart illustrating a sub-step of step S210 in FIG. 4;

FIG. 6 is a flow chart illustrating a sub-step of step S220 in FIG. 4;

fig. 7 is a schematic structural diagram of a cataract image recognition device according to an embodiment of the present application;

fig. 8 is a schematic structural diagram of a computer device provided in an embodiment of the present application.

Detailed Description

In order to make the objects, technical solutions and advantages of the present application more clearly understood, the present application is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the present application and are not intended to limit the present application.

It is to be noted that, unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of the present application only and is not intended to be limiting of the application.

In the related art, when one cataract patient is suffering from nuclear cataract, the ophthalmologist should adopt a corresponding treatment plan based on the nuclear cataract, and simultaneously, the ophthalmologist should adopt a corresponding treatment plan for suffering from two cataract types, such as nuclear cataract and cortical cataract. Most of the existing cataract auxiliary diagnosis and screening systems aim at single type cataract identification or screening, for example: a nuclear cataract auxiliary diagnosis system based on slit lamp images and a cataract screening system based on fundus images do not consider the actual clinical diagnosis requirement of cataracts, and can not identify various cataracts.

Based on this, the embodiment of the present application provides a cataract image identification method, an apparatus, a computer device and a readable storage medium, where the cataract image identification method first obtains a cataract data set, where the cataract data set includes a plurality of cataract images; performing image segmentation processing on each cataract image to obtain a plurality of cataract areas, and extracting different areas of various cataract images; then, inputting each cataract area into a preset image intensity algorithm, a preset histogram algorithm and a preset texture feature algorithm respectively to perform feature statistical processing to obtain a first image feature set corresponding to the image intensity algorithm, a second image feature set corresponding to the histogram algorithm and a third image feature set corresponding to the texture feature algorithm, and combining the first image feature set, the second image feature set and the third image feature set as initial image feature sets to obtain the features corresponding to richer cataract images; inputting the initial image feature set into a preset machine learning model for prediction processing to obtain a first initial prediction result, inputting the initial image feature set into a preset deep neural network model for prediction processing to obtain a second initial prediction result, and obtaining a target prediction result according to the first initial prediction result and the second initial prediction result, so that the accuracy of the prediction result is improved; and determining the category of the cataract image according to the target prediction result and a preset cataract grade. Compared with the prior art, the auxiliary diagnosis system for single type cataract identification has the advantages that the cataract images are respectively input into the image intensity algorithm, the histogram algorithm and the texture feature algorithm for feature statistics, various feature statistical modes are provided, the corresponding features of the abundant cataract images can be obtained, the machine learning model and the deep neural network model are used for result prediction, the prediction accuracy is improved, the type of the cataract images is finally determined, one-stop intelligent identification of the cataract images is realized through image processing, result prediction and type identification, and various types of cataracts can be efficiently and accurately identified.

In one embodiment, the cataract image identification method can be applied to identification and detection of cataract images, can also be applied to other medical images, can be used for two-dimensional ophthalmic images and three-dimensional ophthalmic images, and has wider application. The cataract image identification method is not limited to be deployed in anterior segment OCT imaging equipment, and can also be deployed in ophthalmological equipment, and can also be deployed in servers, workstations, super computing centers, cloud computing centers, personal computers, mobile phones, edge equipment, medical equipment and the like.

The cataract image identification method, device, computer equipment and readable storage medium provided by the embodiments of the present application are described below with reference to the accompanying drawings.

Referring to fig. 1, fig. 1 shows a flowchart of a cataract image identification method provided in an embodiment of the present application. The cataract image identification method includes, but is not limited to, step S100, step S200, step S300, step S400 and step S500.

Step S100, a cataract data set is acquired, where the cataract data set includes a plurality of cataract images.

In one embodiment, the anterior segment optical coherence tomography (AS-OCT) is a novel ophthalmic imaging technique, which has the characteristics of high sensitivity, high resolution, fast imaging speed, objective quantitative measurement, and the like, and has been gradually used for clinical cataract screening and pre-and post-cataract surgery examination. The cataract data set is obtained through the anterior segment optical coherence tomography technology, and comprises a plurality of cataract images, wherein the cataract images have clearer crystalline lens structures, and effective data support is provided for identifying the type of cataract.

In one embodiment, the acquired cataract image may be subjected to a preprocessing operation, which may be an image format conversion, an image denoising operation, an image size stretching operation, or the like, and the acquired cataract image may be subjected to a basic data processing operation. Illustratively, the acquired cataract image is subjected to image denoising processing, so that the image is clearer and the prediction accuracy is favorably improved.

Step S200, image segmentation processing is carried out on each cataract image to obtain a plurality of cataract areas.

As shown in fig. 4, the image segmentation process is performed on each cataract image to obtain a plurality of cataract regions, including but not limited to the following steps:

step S210, carrying out crystal segmentation processing on each cataract image to obtain crystal region characteristics.

In an embodiment, each cataract image can be subjected to crystal segmentation processing by using a trained deep segmentation network model, and a crystal structure is automatically segmented to obtain crystal region characteristics, so that the crystal region characteristics can be conveniently segmented again subsequently to obtain abundant cataract image characteristics, and thus one-stop intelligent identification of the cataract images is realized. The trained deep segmentation network model may be a pnet model or a variation of the pnet model, which is not described herein again.

In one embodiment, the cataract image includes a nuclear image, a cortical image, and a cystic image; as shown in fig. 5, a crystal segmentation process is performed on each cataract image to obtain the crystal region characteristics, including but not limited to the following steps:

in step S211, crystal segmentation processing is performed on each of the nuclear images to obtain a plurality of nuclear crystal regions.

In step S212, a crystal segmentation process is performed on each cortical image to obtain a plurality of cortical crystal regions.

In step S213, a crystal segmentation process is performed on each cystic image to obtain a plurality of cystic crystal regions.

In one embodiment, the cataract may be classified into nuclear cataract, cortical cataract and cystic cataract according to the location of the opacity symptom because the clinical symptom of the cataract is manifested as opacity in the lens region according to the obtained cataract image. According to the position of the crystalline region, carrying out crystal segmentation processing on the cataract image including a nuclear image, a cortical image and a cystic image, and specifically comprising the following steps: performing crystal segmentation processing on each nuclear image to obtain a plurality of nuclear crystal regions, and performing image processing on the nuclear crystal regions again in the follow-up process by obtaining the nuclear crystal regions so as to obtain cataract regions; performing crystal segmentation processing on each cortical image to obtain a plurality of cortical crystal regions, wherein the obtained cortical crystal regions are beneficial to performing image processing on the cortical crystal regions again in the follow-up process, so that a cataract region is obtained; and performing crystal segmentation processing on each cystic image to obtain a plurality of cystic crystal regions, and performing image processing on the cystic crystal regions again in the follow-up process by obtaining the cystic crystal regions to obtain the cataract region.

Step S220, image processing is performed on the crystal region features to obtain a plurality of cataract regions.

In an embodiment, a preset image processing method can be adopted to perform image processing on the features of the crystal region to obtain a plurality of cataract regions, which is beneficial to performing feature statistical calculation according to the cataract regions. The preset image processing method may be a symmetric processing method, and because of the symmetry of the eye tissue structure, the central axis is directly searched for segmentation, and the processing may also be according to other proportions, which is not described herein again.

As shown in fig. 6, image processing is performed on the features of the lens region to obtain a plurality of cataract regions, including but not limited to the following steps:

step S221, performing region division processing on each of the nucleated crystal regions to obtain an entire nucleated region, an upper semi-nucleated region, and a lower semi-nucleated region.

In one embodiment, each nuclear crystal region is subjected to region division processing by using a symmetry processing method, and is divided into an entire region when not divided, and then the nuclear crystal region is divided into two regions, namely an upper half region and a lower half region, so that the entire nuclear region, the upper half region and the lower half region are obtained. By carrying out multiple division, different areas of various cataract images can be extracted, which is beneficial to the follow-up statistics of the characteristics of different states.

In step S222, a region division process is performed on each cortical crystal region to obtain an overall cortical region, a top-half cortical region, and a bottom-half cortical region.

In one embodiment, each cortical crystal region is subjected to a region division process by using a symmetry processing method, and is divided into an overall region when not divided, and then the cortical crystal region is divided into two parts, namely an upper half region and a lower half region, so that an overall cortical region, an upper half cortical region and a lower half cortical region are obtained. By carrying out multiple division, different areas of various cataract images can be extracted, which is beneficial to the follow-up statistics of the characteristics of different states.

Step S223, performing area division processing on each cystic crystal area to obtain an overall cystic area, an upper semi-cystic area, and a lower semi-cystic area.

In one embodiment, each cystic crystal region is subjected to region division processing by using a symmetry processing method, and is divided into an overall region when not divided, and then the cystic crystal region is divided into two parts, namely an upper half region and a lower half region, so that the overall cystic region, the upper half cystic region and the lower half cystic region are obtained. By carrying out multiple division, different areas of various cataract images can be extracted, which is beneficial to the follow-up statistics of the characteristics of different states.

In step S224, the nuclear region, the supranuclear region, the subnuclear region, the whole cortical region, the suprasemicortical region, the subsemicortical region, the whole cystic region, the suprasemicystic region, and the subsemicystic region are set as the cataract region.

In one embodiment, the nuclear region, the upper nuclear region, the lower nuclear region, the whole cortical region, the upper semi-cortical region, the lower semi-cortical region, the whole cystic region, the upper semi-cystic region and the lower semi-cystic region are obtained from steps S221 to S224 as the cataract region, so as to provide data support for the subsequent feature statistics.

Step S300, respectively inputting each cataract area into a preset image intensity algorithm, a preset histogram algorithm and a preset texture feature algorithm for feature statistical processing to obtain a first image feature set corresponding to the image intensity algorithm, a second image feature set corresponding to the histogram algorithm and a third image feature set corresponding to the texture feature algorithm, and merging the first image feature set, the second image feature set and the third image feature set to serve as initial image feature sets.

In an embodiment, the preset image intensity algorithm may be a feature statistical method based on image intensity, and according to step S210 and step S220, feature statistics is performed on each type of cataract image by using the feature statistical method based on image intensity, so that relatively rich features can be obtained. For example, taking the identification of the nuclear cataract as an example, feature statistics is performed on the nuclear region, the upper nuclear region and the lower nuclear region by using a feature statistical method based on image intensity, and statistical features of 15 image intensities are extracted from each region, and the 15 feature statistics methods are respectively as follows: (1) Mean (Mean): mean of all pixels of the nucleated region. The distribution of the turbidity of the nuclear cataract is relatively uniform, and the method is used as an important reference standard for judging the severity of the nuclear cataract; (2) Variance (Variance): the average of the squared values of the difference between each pixel value and the pixel mean can be used to measure the degree of dispersion (uniformity) of the image pixels; (3) Maximum (Maximum): a maximum pixel value of the nucleated region; (4) Minimum (Minimum): a minimum pixel value of the nucleated region; (5) pixel Range (Range): a difference between a maximum pixel value and a minimum pixel value of the nucleated region; (6) Standard deviation (Std): the square root of the variance of the pixel values of the nucleated region; (7) Median (Median): a value at an intermediate magnitude describing a pixel gray value; (8) Energy (Energy): the mean of the sum of squares of all pixel values of the nucleated region; (9) Root Mean Square (RMS): the square root of the energy characteristic of the pixel; (10) Pixel grayscale at 10% (10 th pixel percentile, P10): all pixel gray values of the nuclear region are sorted from small to large, and the pixel values positioned in the first 10 percent can replace the minimum pixel gray value; (11) Pixel gray value at 90% (90th pixel percentile, p90): all pixel gray values of the nuclear region are sorted from small to large, and the pixel gray value with the ranking of 90% can replace the maximum pixel gray value; (12) Interquartile range of pixels: sorting all pixel gray values of the nuclear region from small to large, wherein the ranking is the difference between the pixel gray value with the ranking of 75% and the pixel gray value with the ranking of 25%; (13) Mean absolute deviation (Mean absolute deviation) of pixels: and the average value of the sum of the absolute values of the differences between all the pixel gray values and the pixel average value in the nuclear region is used for measuring the deviation degree between all the pixel gray values and the pixel average value. The larger the average absolute deviation value is, the pixel gray value distribution is discrete; otherwise, the distribution of the pixel gray values is concentrated; (14) Skewness of pixel (Skewness): measuring the deflection direction and degree of the gray value distribution of all pixels in the nuclear region, wherein the deflection direction and degree are digital characteristics used for counting the asymmetric degree of the pixel distribution; (15) Kurtosis (Kurtosis): measuring the peak value (flatness) of the pixel gray value distribution; the higher the kurtosis, the more the pixel distribution is centered on the tail rather than the mean; lower kurtosis means the opposite.

In another embodiment, the preset histogram algorithm may be a histogram-based statistical feature method, and according to step S210 and step S220, feature statistics is performed on each type of cataract image by using a feature statistical method based on image intensity, so that more abundant features can be obtained. Exemplarily, taking a nuclear cataract as an example, feature statistics is performed on a nuclear region, a top nuclear region and a bottom nuclear region by using a histogram-based statistical feature method, and 23 histogram statistical features are extracted from each region, where a specific statistical feature mode is as follows:

(1) Discretized mean (mean discrete intensity): the mean value of the pixel distribution used for measuring the histogram can be used as a clinical reference index for evaluating the cataract severity level, and the formula is as follows:

(2) Discretized variance (discretized intensity variance): the dispersion (uniformity) of the histogram distribution of an image can be measured) The extent, expressed as:

(3) Discretized intensity (discretized intensity) bias: used to measure the asymmetry of the histogram distribution, is expressed as:

(4) Discretized intensity kurtosis (discrete intensity kurtosis): the kurtosis of the histogram distribution can be measured as:

(5) Discretized median (median discretized intensity): the middle position of the histogram distribution can be described, and the distribution trend of the pixels is reflected;

(6) Absolute deviation of discretized mean (intensity histogram mean deviation): the difference between the histogram pixel values and the discretized mean can be measured as:

(7) Absolute robust deviation (intensity histogram mean absolute deviation) of the discretized mean: the absolute deviation of the histogram pixel values from the discretized mean, which is a complement of the absolute deviation feature of the discretized mean, is between 10% and 90%, and is represented by

(8) Absolute deviation of discretized median (intensity histogram mean absolute deviation): can be used to measure the difference between histogram and discretized median, and is expressed as

(9) Discretized coefficient of variation (intensity histogram coeffecification of variation): histograms can be scaledExpressed as:

(10) Discrete entropy (discrete entropy): the uncertainty used to measure the histogram is expressed as:

(11) Discretized intensity uniformity (discrete intensity uniformity): used to measure the degree of deviation or randomness of the histogram from the average, is expressed as:

(12) 10th discretized intensity value interval (10 th discrete intensity property): the sequence number of the interval in which 10% of pixels are located in the histogram;

(13) 90th discretized intensity value interval (90 th discrete intensity property): the sequence number of the interval where 90% of pixels are located in the histogram;

(14) Discretized mode (intensity histogram mode): refers to the most numerous pixel values in a histogram;

(15) Discretized minimum (minimum discrete intensity): represents the left boundary value of the smallest non-zero interval of pixels in the histogram, expressed as: l is _min ＝min(L _i )；

(16) Discretized maximum (maximum discretized intensity): represents the largest non-zero pixel bin left boundary in the histogram, expressed as: l is _max ＝max(L _i )；

(17) Discretized interquartile range (discrete intensity range): the approximate distribution trend of the histogram can be measured and expressed as: l is ₇₅ th-L ₂₅ th；

(18) Discretized intensity range (discretized intensity range): the overall bin distribution of pixel values for a metric histogram is expressed as: l is ₁₀₀ th-L ₁ th；

(19) Discretized quartile dispersionCoefficient (intensity histogram coefficient of dispersion): the distribution of the pixel values of the histogram is measured and expressed as: (L) ₇₅ th-L ₂₅ th)/(L ₇₅ th+L ₂₅ th)；

(20) Maximum histogram gradient (maximum histogram gradient): the maximum value of the gradient direction in the histogram;

(21) Maximum histogram gradient intensity value (maximum histogram gradient intensity): the sequence number of the interval where the maximum value of the gradient direction in the histogram is located;

(22) Minimum Histogram Gradient (Minimum Histogram Gradient): the minimum value of the gradient direction in the histogram;

(23) Minimum Histogram Gradient Intensity value (Minimum Histogram Gradient Intensity): and the sequence number of the interval where the minimum value of the gradient direction in the histogram is located.

Wherein, in the above expression, μ represents a discretized mean, i represents an interval index, N represents a total number of intervals, X _i The number of pixels in the interval i is the ratio of the total number, m is the discretized median, σ is the discretized standard deviation, and L is _i Indicating the left boundary value, L, of the finger interval i _min Indicates the minimum interval left boundary value, L _max Indicates the maximum interval left boundary value, L ₂₅ th and L ₇₅ th represents the left boundary value of the interval in which the lower quartile (25%) of pixels is located and the left boundary value of the interval in which the upper quartile (75%) of pixels is located, L ₁₀₀ th and L ₁ th denotes the left boundary value of the bin where the histogram is largest and the left boundary value of the bin where the histogram is smallest, respectively.

In another embodiment, the preset texture feature algorithm may be a gray level co-occurrence matrix, a gray level area size matrix, or a field gray level difference matrix method, and according to step S210 and step S220, feature statistics is performed on each type of cataract image by using the gray level co-occurrence matrix, the gray level area size matrix, or the field gray level difference matrix method, respectively, so that relatively rich features can be obtained. Exemplarily, taking a nuclear cataract as an example, respectively performing feature statistics on a nuclear region, an upper semi-nuclear region and a lower semi-nuclear region by using a gray level co-occurrence matrix, a gray level region size matrix or a domain gray level difference matrix method, and extracting 22 texture statistical features from each region, wherein the specific statistical feature mode is as follows:

(1) The gray level co-occurrence matrix is usually used to describe the spatial correlation characteristics of gray levels, and also reflects the position distribution rule of pixels with similar brightness, and can be represented by a size N _g ×N _g Is represented by a matrix of N _g The number of levels representing the maximum gray scale value of the image may be selected as N _g =16. The gray level co-occurrence matrix cannot be directly used as a feature, and 6 common texture features are extracted in the embodiment: energy, entropy, contrast, homogeneity, correlation, dissimilarity, and reciprocal distance, which can be expressed using the following calculation formulas:

energy (energy): the texture thickness degree and the gray distribution uniformity degree of the AS-OCT image can be described, and are expressed AS follows:

wherein P (i, j) represents a normalized gray level co-occurrence matrix;

entropy (entropy): describing the disorder degree of the texture, the larger the entropy value is, the more complicated the gray level distribution of the AS-OCT image is, and the expression is:

homogeneity (homogeneity): the degree of regularity in describing the texture is expressed as:

correlation (correlation): describing the similarity degree of the gray level co-occurrence matrix elements in the row or column direction, the local correlation of the image can be reflected, and the larger the value is, the larger the correlation is, which is expressed as:

wherein,

dissimilarity (dissimilarity): the clarity of the texture is described as:

inverse Difference (IDM): describing the degree of regularity of the texture, measuring the local variation of the texture, and expressing as:

(2) The gray scale area size matrix reflects the distribution characteristics, such as size and number, of pixels with the same gray scale level in the image. The embodiment considers eight connectivity, that is, each pixel has 8 adjacent pixels, and p (i, j) is used to represent the number of areas with the image gray value i and the area of the connected region j. 11 features are extracted from the gray scale area size matrix, and the details are as follows:

small zone preference (SZE): the fineness of the texture is described, and the larger the numerical value is, the more small areas are indicated, the more fine the texture of the image is, and the expression is:

wherein, N _z Denotes the area size of the maximum connected region, p _z The number of areas of the connected region is represented as j, and the calculation formula is as follows:

large zone preferences (LZE): the roughness of the texture is described, and the larger the value, the more the large area is, the coarser the image texture is, expressed as:

zone Percentage (ZP): the proportion of the number of connected region pixels to the total number of pixels in the image is expressed as:

wherein N is _s Indicating the number of connected regions

N _p Representing the number of pixels in the image;

low gray level zone preference (LGZE): the proportion of the low-gray-scale area of the image in the image is described, and the higher the value of the low-gray-scale area is, the smaller the proportion of the low-gray-scale area in the image is shown as follows:

wherein p is _g Representing the number of areas with gray value j of connected region

High gray level zone preference (HGZE): the proportion of the high-gray-scale area of the image in the image is described, and the higher the value, the larger the proportion of the high-gray-scale area in the image is shown as follows:

small area low gray level preference (SZLGE): the proportion of small area regions of low gray values in the image is described, and is expressed as:

small area high gray level preference (SZHGE): the proportion of small area regions of high grey value in the GLSZM image is described and expressed as:

large area low gray level preference (LZLGE): the proportion of a large area with a low gray scale value in the image is described and expressed as:

large area high gray level preference (LZHGE): the proportion of a large area of high grey value in the image is described and expressed as:

gray Level Variance (GLV): the regional variance of the gray level calculation is described as:

wherein, mu _g Mean value of gray-scale values representing regions

Zone-size variance (ZSV): the gray variance of regions of different area sizes is described and expressed as:

wherein, mu _z Mean value of gray-scale values representing regions

(3) The neighborhood gray difference matrix is used for describing the correlation between the gray value of a pixel and the mean value of the neighborhood gray, s (i) is used for representing the sum of the pixel with the gray level i of an image and the mean value of the gray level of the pixel with the distance d, and d is set to be 1, wherein s (i) is expressed as:

wherein A is _i Representing the gray values of the pixels adjacent to the pixel.

The gray scale area size matrix extracts 5 features, specifically as follows:

roughness (roughness): the average difference between the intermediate pixel and the neighbors is described and expressed as:

wherein N is _g Represents the highest gray level in the image, ε is a minimum value that prevents courense from being infinite;

contrast (contrast, ct): the dynamic range, which represents the image intensity variation and overall gray level, is expressed as:

wherein N is _p Representing the number of grey levels in the image, p (i) representing the probability that the grey level is i

n _i Representing a pixel with a gray value i; n is the total number of pixels ∑ N _i ；

Complexity (busyness): this indicates the rate of change from one pixel to the surrounding pixels (p (i) ≠ 0, p (j) ≠ 0), as:

complexity (cp): the intensity variation (p (i) ≠ 0, p (j) ≠ 0) indicative of the uniformity of the pixels in the NGTDM matrix is expressed as:

strength (strength): the rate of change and the coarseness (p (i) ≠ 0, p (j) ≠ 0) used to describe the intensity of the gray scale of the image, expressed as:

in one embodiment, a first image feature set is obtained by performing feature statistics on different regions of each type of cataract by using a feature statistical method based on image intensity, a second image feature set is obtained by performing feature statistics on different regions of each type of cataract by using a feature statistical method based on a histogram, and a third image feature set is obtained by performing feature statistics on different regions of each type of cataract by using a feature statistical method based on image texture, and the first image feature set, the second image feature set and the third image feature set obtained by the statistics are collected and computed to be used as an initial image feature set which has rich features, so that the prediction accuracy is high. With the above embodiment, each type of region has 60 statistics, so that for each cataract type, the statistics of 180 statistics can be set as required, which is not described herein. The first image feature set is a set obtained by feature statistical method statistical features based on image intensity, the second image feature set is a set obtained by feature statistical method statistical features based on a histogram, and the third image feature set is a set obtained by feature statistical method statistical features based on image texture.

As shown in fig. 2, after the first image feature set, the second image feature set and the third image feature set are combined as the initial image feature set, the cataract image identification method further includes, but is not limited to, the following steps:

and step S600, performing feature importance analysis on the initial image feature set to obtain a target image feature set.

Step S410, inputting the target image feature set into a preset machine learning model for prediction processing, so as to obtain a first initial prediction result.

Step S420, inputting the target image feature set into a preset deep neural network model for prediction processing, and obtaining a second initial prediction result.

In an embodiment, the cataract image identification method can perform feature importance analysis on the initial image feature set by using a feature iterative deletion method or a logistic regression method which is subjected to dimension reduction processing and trained to obtain a target image feature set, and can also obtain the target image feature set by deleting redundant features, namely the target image feature set is used for cataract severity level prediction features. The method can be suitable for analyzing important characteristics of different types of cataract images so as to improve the prediction accuracy. For example, a Principal Component Analysis (PCA) may be used to perform a dimensionality reduction process on the initial image feature set, and a target image feature set of the cataract image may be determined according to a result of the dimensionality reduction. The initial image feature set can be analyzed based on the importance analysis model to determine the score of the initial image feature; and taking the initial image features with the scores larger than the set values as target image features. Wherein, the importance analysis model is determined by a person skilled in the art based on a logistic regression method and a characteristic iterative deletion method; the set value is set by a person skilled in the art according to actual conditions. Inputting the initial image feature set into an importance analysis model to obtain the score of each initial image feature, and if the score is greater than or equal to a set value, retaining the initial image feature; and if the score is smaller than a set value, deleting the initial image characteristics.

In an embodiment, after the target image feature set is obtained, the target image feature set is input into a preset machine learning model for prediction processing to obtain a first initial prediction result, and the target image feature set is input into a preset deep neural network model for prediction processing to obtain a second initial prediction result, which is beneficial to improving the accuracy of subsequent target prediction results.

And S400, inputting the initial image feature set into a preset machine learning model for prediction processing to obtain a first initial prediction result, inputting the initial image feature set into a preset deep neural network model for prediction processing to obtain a second initial prediction result, and obtaining a target prediction result according to the first initial prediction result and the second initial prediction result.

In an embodiment, the preset machine learning model may be a logistic regression algorithm, and the target image feature set is input into the logistic regression algorithm to perform feature extraction, and then result prediction is performed to obtain a first initial prediction result, which is beneficial to subsequently calculating a target prediction result by using the first initial prediction result. The first initial prediction result is the result of predicting the cataract image by the machine learning model.

In an embodiment, the preset deep neural network model may be an attention-based neural network model or a variant of the attention-based neural network model, the target image feature set is input into the attention-based neural network model to perform feature extraction, and then result prediction is performed to obtain a second initial prediction result, which is beneficial to subsequently calculating a target prediction result by using the second initial prediction result. And the second initial prediction result is a result of predicting the cataract image by the deep neural network model.

In an embodiment, the first initial prediction result and the second initial prediction result are subjected to fusion processing by using a fusion algorithm of weighted summation and averaging to obtain a target prediction result, or the first initial prediction result and the second initial prediction result are subjected to fusion processing by using a fusion processing mode selected for an optimal result to obtain the target prediction result. The accuracy of the prediction result can be improved by obtaining the target prediction result by fusing the prediction results. And the target prediction result is expressed as a prediction result of a fusion machine learning model and a deep neural network model. Illustratively, the first initial prediction result and the second initial prediction result are fused through optimal result selection, specifically, the sizes of the first initial prediction result and the second initial prediction result are compared, and the target prediction result with a larger prediction result is selected. The fusion mode of averaging is arithmetic mode of two-number averaging calculation, which is not described herein.

As shown in fig. 3, inputting the initial image feature set into a preset machine learning model for prediction processing to obtain a first initial prediction result, including but not limited to the following steps:

step S430, inputting the initial image feature set into a machine learning model for feature extraction to obtain a first extracted feature set.

Step S440, visualizing the first extracted feature set to obtain a first feature visualization map, and displaying the first feature visualization map.

Step S450, performing prediction calculation on the features in the first feature visualization graph to obtain a first initial prediction result.

In one embodiment, in the process of inputting the initial image feature set into a preset machine learning model for prediction processing, firstly, the machine learning model performs feature extraction on the initial image feature set, and visualizes the first extracted feature set by using a visualization tool to obtain a first feature visualization graph, wherein the first feature visualization graph improves interpretability and credibility of a prediction result. The first characteristic visualization graph is a visualization graph of extracted characteristics of the machine learning model; the first extraction feature set is a set obtained by extracting features by using a machine learning model. After feature extraction is carried out on the features, result prediction processing is carried out by using the features in the first feature visualization graph, and a first initial prediction result is obtained.

Inputting the initial image feature set into a preset deep neural network model for prediction processing to obtain a second initial prediction result, wherein the method comprises the following steps of:

step S460, inputting the initial image feature set into a deep neural network model for feature extraction to obtain a second extracted feature set;

step S470, visualizing the second extracted feature set to obtain a second feature visualization map, and displaying the second feature visualization map;

and step S480, performing prediction calculation on the features in the second feature visualization graph to obtain a second initial prediction result.

In one embodiment, in the process of inputting the initial image feature set into the deep neural network model for prediction processing, firstly, the deep neural network model performs feature extraction on the initial image feature set, and visualizes the second extracted feature set by using a visualization tool to obtain a second feature visualization graph, wherein the interpretability and the credibility of a prediction result are improved by the second feature visualization graph. The second characteristic visualization graph is a visualization graph of extracted characteristics of the deep neural network model; the second extraction characteristic set is a set obtained by extracting characteristics by using a deep neural network model. And after the features are extracted, performing result prediction processing by using the features in the second feature visualization graph to obtain a second initial prediction result.

In one embodiment, the first characteristic visualization graph and the second characteristic visualization graph are stored in a database, sent to a terminal and displayed to medical staff for watching through the terminal, so that the medical staff can fully know the medical image.

And step S500, determining the category of the cataract image according to the target prediction result and the preset cataract grade.

In an embodiment, according to the target prediction result obtained in step S400, the category of the cataract image is determined according to the preset cataract category and the preset cataract level, the identified image category is stored in the database, and is sent to the terminal, and is displayed to the medical staff through the terminal for viewing, wherein the preset cataract level may be divided into five levels, and the preset cataract level can be set according to the scene corresponding to the severity of the three cataract pathologies, which is not described herein.

As shown in fig. 7, the cataract image identification device 100 according to the embodiment of the present application first acquires a cataract data set by using the data acquisition module 110, where the cataract data set includes a plurality of cataract images; then, the image processing module 120 performs image segmentation processing on each cataract image to obtain a plurality of cataract areas; then, the feature statistical module 130 is adopted to input each cataract area into a preset image intensity algorithm, a preset histogram algorithm and a preset texture feature algorithm respectively for feature statistical processing, so as to obtain a first image feature set corresponding to the image intensity algorithm, a second image feature set corresponding to the histogram algorithm and a third image feature set corresponding to the texture feature algorithm, and the first image feature set, the second image feature set and the third image feature set are combined to be used as an initial image feature set, so that the features corresponding to the abundant cataract images can be obtained through the feature statistics; the result prediction module 140 is used for inputting the initial image feature set into a preset machine learning model for prediction processing to obtain a first initial prediction result, inputting the initial image feature set into a preset deep neural network model for prediction processing to obtain a second initial prediction result, and according to the first initial prediction result and the second initial prediction result, a target prediction result is obtained, so that the accuracy of the prediction result is improved; and finally, determining the category of the cataract image according to the target prediction result and the preset cataract grade by utilizing the suggestion output module 150.

It should be noted that the data obtaining module 110 is connected to the image processing module 120, the image processing module 120 is connected to the feature counting module 130, the feature counting module 130 is connected to the result predicting module 140, and the result predicting module 140 is connected to the suggestion outputting module 150. The cataract image identification method is applied to the cataract image identification device 100, the cataract image identification device 100 in the embodiment of the application provides a plurality of characteristic statistical modes by inputting the cataract image into the image intensity algorithm, the histogram algorithm and the texture characteristic algorithm respectively for characteristic statistics, can obtain abundant characteristics corresponding to the cataract image, predicts the result by using the machine learning model and the deep neural network model, improves the prediction accuracy, finally outputs the type of the cataract image, realizes the one-stop intelligent identification of the cataract image through image processing, result prediction and type identification, and can efficiently and accurately identify various cataracts.

The embodiments described in the embodiments of the present application are for more clearly illustrating the technical solutions of the embodiments of the present application, and do not constitute a limitation to the technical solutions provided in the embodiments of the present application, and it is obvious to those skilled in the art that the technical solutions provided in the embodiments of the present application are also applicable to similar technical problems with the evolution of technology and the emergence of new application scenarios.

Fig. 8 illustrates a computer device 500 provided by an embodiment of the present application. The computer device 500 may be a base station or a terminal, and the internal structure of the computer device 500 includes but is not limited to:

a memory 510 for storing programs;

and a processor 520 configured to execute the program stored in the memory 510, wherein when the processor 520 executes the program stored in the memory 510, the processor 520 is configured to perform the cataract image identification method described above.

The processor 520 and the memory 510 may be connected by a bus or other means.

The memory 510 is a non-transitory computer readable storage medium, and can be used to store a non-transitory software program and a non-transitory computer executable program, such as the cataract image recognition method described in any embodiment of the present application. The processor 520 implements the cataract image identification method described above by executing non-transitory software programs and instructions stored in the memory 510.

The memory 510 may include a storage program area and a storage data area, wherein the storage program area may store an operating system, an application program required for at least one function; the storage data area may store data for performing the cataract image recognition method described above. Further, memory 510 may include high speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid state storage device. In some embodiments, the memory 510 may optionally include memory located remotely from the processor 520, which may be connected to the processor 520 via a network. Examples of such networks include, but are not limited to, the internet, intranets, local area networks, mobile communication networks, and combinations thereof.

The non-transitory software programs and instructions necessary to implement the cataract image identification methods described above are stored in the memory 510 and when executed by the one or more processors 520 perform the cataract image identification methods provided by any of the embodiments of the present application.

The embodiment of the application also provides a computer-readable storage medium, which stores computer-executable instructions for executing the cataract image identification method.

In one embodiment, the storage medium stores computer-executable instructions that, when executed by one or more control processors 520, for example, by one of the processors 520 of the computer device 500, cause the one or more processors 520 to perform the cataract image recognition method provided in any of the embodiments of the present application.

The embodiments described above are merely illustrative, where elements described as separate components may or may not be physically separate, may be located in one place, or may be distributed over a plurality of network elements. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of the present embodiment.

The terms "first," "second," "third," and the like (if any) in the description of the present application and in the above-described figures are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used is interchangeable under appropriate circumstances such that the embodiments of the application described herein are capable of operation in sequences other than those illustrated or described herein. Furthermore, the terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed, but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus.

It should be understood that, in this application, "at least one" means one or more, "a plurality" means two or more. "and/or" for describing an association relationship of associated objects, indicating that there may be three relationships, e.g., "a and/or B" may indicate: only A, only B and both A and B are present, wherein A and B may be singular or plural. The character "/" generally indicates that the former and latter associated objects are in an "or" relationship. "at least one of the following" or similar expressions refer to any combination of these items, including any combination of the singular or plural items. For example, at least one (one) of a, b, or c, may represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", wherein a, b, c may be single or plural.

It will be understood by those of ordinary skill in the art that all or some of the steps, systems, and methods disclosed above may be implemented as software, firmware, hardware, or suitable combinations thereof. Some or all of the physical components may be implemented as software executed by a processor, such as a central processing unit, digital signal processor, or microprocessor, or as hardware, or as an integrated circuit, such as an application specific integrated circuit. Such software may be distributed on computer readable media, which may include computer storage media (or non-transitory media) and communication media (or transitory media). The term computer storage media includes volatile and nonvolatile, removable and non-removable media implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data, as is well known to those of ordinary skill in the art. Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital Versatile Disks (DVD) or other optical disk storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other medium which can be used to store the desired information and which can accessed by a computer. In addition, communication media typically embodies computer readable instructions, data structures, program modules or other data in a modulated data signal such as a carrier wave or other transport mechanism and includes any information delivery media as is well known to those skilled in the art.

While the preferred embodiments of the present application have been described in detail, the present application is not limited to the above embodiments, and those skilled in the art will appreciate that the present application is not limited thereto. Under the shared conditions, various equivalent modifications or substitutions can be made, and the equivalent modifications or substitutions are included in the scope defined by the claims of the present application.

Claims

1. A cataract image identification method, characterized in that the method comprises:

respectively inputting each cataract area into a preset image intensity algorithm, a preset histogram algorithm and a preset texture feature algorithm for feature statistical processing to obtain a first image feature set corresponding to the image intensity algorithm, a second image feature set corresponding to the histogram algorithm and a third image feature set corresponding to the texture feature algorithm, and merging the first image feature set, the second image feature set and the third image feature set to obtain an initial image feature set;

2. The cataract image identification method of claim 1, wherein after the merging of the first, second, and third image feature sets as an initial image feature set, the method further comprises:

3. The cataract image identification method according to claim 1 or 2, wherein the obtaining the target prediction result according to the first initial prediction result and the second initial prediction result comprises:

4. The cataract image identification method according to claim 1, wherein the step of inputting the initial image feature set into a preset machine learning model for prediction processing to obtain a first initial prediction result comprises:

inputting the initial image feature set into the machine learning model for feature extraction to obtain a first extracted feature set;

inputting the initial image feature set into the deep neural network model for feature extraction to obtain a second extraction feature set;

and performing prediction calculation on the features in the second feature visualization graph to obtain a second initial prediction result.

5. The cataract image recognition method of claim 1, wherein the image segmentation processing on each of the cataract images to obtain a plurality of cataract regions comprises:

and carrying out image processing on the crystal region characteristics to obtain a plurality of cataract regions.

6. The cataract image identification method according to claim 5, wherein the cataract image includes a nuclear image, a cortical image, and a cystic image;

7. The cataract image identification method according to claim 6, wherein the image processing the crystal region feature to obtain a plurality of cataract regions comprises:

performing region division processing on each nuclear crystal region to obtain an integral nuclear region, an upper semi-nuclear region and a lower semi-nuclear region;

carrying out region division treatment on each cortical crystal region to obtain an integral cortical region, an upper half cortical region and a lower half cortical region;

8. An image identification device for cataract, the device comprising:

a data acquisition module to acquire a cataract data set, the cataract data set comprising a plurality of cataract images;

9. A computer device comprising a memory and a processor, the memory having stored therein computer-readable instructions which, when executed by one or more of the processors, cause the one or more processors to perform the steps of the method of any one of claims 1 to 7.

10. A computer-readable storage medium readable by a processor, the storage medium readable by the processor and storing computer instructions which, when executed by one or more processors, cause the one or more processors to perform the steps of the method of any one of claims 1 to 7.