ES2614043T3 - Evaluation of lesions in an image - Google Patents

Abstract

A method for assessing the presence or absence of injury or lesions in an image, comprising a) the establishment of at least one representative starting point of a lesion in the image, b) the selection of at least one starting point, estimating a subset of the image corresponding to a candidate lesion by means of isocurves that grow around the starting point, with an integer of at least 1, until the periphery of the area of the candidate lesion is established, the isocurves extending from a vector of one or more visibility characteristics, c) the determination of a value of the visibility characteristic for the visibility of the area of candidate injury, the value of the visibility characteristic being determined from a visibility characteristic different from those visibility characteristics used for growth in stage b), and comparing the value of the visibility characteristic for the vis ibility of the area of candidate injury with a predetermined threshold, d) the classification of the area of candidate injury detected in b) with respect to the predetermined threshold as a visible or non-visible lesion, e) optionally the repetition of stages b) to d) until all areas of candidate injuries have been classified.

Description

DESCRIPTION

Evaluation of lesions in an image

5 The present invention relates to a method for assessing the presence or absence of injury or lesions in an image and a system for it, where said image can be any image that potentially includes injuries, in particular a medical diagnostic image. , and more particularly an image of the fundus.

10 Background

The analysis of the images of the fundus of the eye presents several difficulties, such as high image variability, the need for reliable processing against conditions for obtaining non-ideal images and short calculation times. A great variability is observed among the different patients, even if they are healthy, where the situation worsens when there are pathological states. For the same patient, variability is observed under different conditions of obtaining images and during the course of a treatment or simply a long period of time. In addition, the fundus images are often characterized by limited quality, being subject to inadequate lighting, flashes, blurring, loss of focus and artifacts that come from reflection, refraction and dispersion.

twenty

The extraction and automatic analysis of the vascular tree from the images of the fundus is an important task in the analysis of images of the fundus for several reasons. First, the vascular tree is the most prominent feature of the retina, and is present regardless of the state of health. This makes the vascular tree an obvious basis for automated logging and synthesis synthesis algorithms. In addition, the task of automatic and robust localization of the head of the optic nerve and the fovea, as well as the task of automatic classification of veins and arteries in the images of the fundus of the eye can depend strongly on an adequate removal of the vascular tree . Another example is the task of automatic detection of lesions that in many cases resemble blood vessels. An adequately extracted tree of vessels can be a valuable instrument to rule out false positive responses produced by said algorithm, thereby increasing its specificity. Finally, the vessels often show different pathological manifestations in themselves, such as an increase in tortuosity, abnormal changes in caliber and deproliferation. An automatic vessel tracking algorithm would also be the obvious basis for the analysis of these phenomena.

Diabetes is the leading cause of blindness in adults of working age. It is a disease that, among its many symptoms, includes a progressive deterioration of the peripheral vascular system. These changes in the vasculature of the retina cause progressive deterioration of vision and can lead to complete loss of sight. The tragedy of diabetic retinopathy is that in the vast majority of cases, blindness can be prevented with early diagnosis and treatment, although screening programs that could provide early detection are not widespread.

40

Currently there are promising techniques for the early detection of diabetic retinopathy. Researchers have discovered that retinopathy is preceded by visibly detectable changes in blood flow through the retina. Today there are diagnostic techniques that categorize and classify diabetic retinopathy, and together with a series of images of the retina taken at different times, they provide a methodology for the early detection of degeneration. Thus, various medical, surgical and dietary interventions can prevent the evolution towards disease blindness.

Despite the existing techniques to prevent diabetic blindness, only a small fraction of the affected population receives adequate attention and on time, and important barriers separate the majority of patients from 50 technical advances in eye care in case of diabetes . There are a limited number of ophthalmologists prepared to assess retinopathy, and they are mostly in the center of the populations. Many patients cannot afford the costs or time needed to travel to a specialist. In addition, cultural and linguistic barriers often prevent elderly, rural and ethnic minority patients from seeking adequate attention. On the other hand, since diabetes is a persistent disease and diabetic retinopathy is a degenerative disease, an affected patient requires lifelong treatment for the disease, which includes periodic examinations to monitor and record the state of the retina, and a sustained attention by the patient to the medical or behavioral guidelines received. This sustained level of personal responsibility demands a high degree of motivation, and the treatment of the disease forever can be a major burden on the lifestyle. These factors raise the probability that the patient, at least

at some point, you do not receive adequate treatment for your disease, often with catastrophic consequences.

Consequently, it would be convenient to launch a more extensive screening of the degeneration or pathological involvement of the retina, and to positively address the financial, social and cultural barriers to the application of said screening. It would also be convenient to improve the efficiency and quality of the evaluation of the retina.

At this point, accurate knowledge of the location and orientations of the fundus structures is important, which includes the location of the vessels. At present, the exploration of the images of the fundus of the eye is carried out mainly by a clinician who examines each image “manually”. This work not only consumes a lot of time, since even an experienced clinician may need several minutes to evaluate a single image, but it is also prone to error since there may be inconsistencies between the way different clinicians evaluate a given image. Therefore it is convenient to provide ways to automate the process of the analysis of the images of the fundus, using computerized image analysis, so that at least preliminary screening information is provided and also as a diagnostic aid that facilitates the clinician Difficult case analysis.

Next, it is usually convenient to provide a procedure for precise determination, using computerized image analysis techniques, the position of the papilla (the exit point of the optic nerve) and the fovea (the region in the center of the retina, where the retina is more sensitive to light), as well as the vessels in the back of the eye.

twenty

Summary of the invention

The present invention relates to a method for detecting lesions in an image, wherein said image comprises a plurality of vessels. The image can be an image of any subject that comprises the vessels. 25 In particular, the procedure refers to diagnostic imaging in medicine, such as X-rays, scan images, photographs, nuclear magnetic radiation scans, CT scans, as well as other images that potentially include injuries.

In order to perform an automatic detection of various structures in the fundus images, a reliable procedure is used to detect lesions in the fundus images that actually contain the lesions, and that reliably does not detect lesions in other images that do not They include injuries. Current procedures can detect lesions in many images, but procedures are not reliable when applied to images that do not contain lesions.

35 In addition, the procedure must be robust in the sense that it should be applicable to a wide variety of images irrespective of lighting, the presence of disease symptoms and / or image artifacts. The lesions can be any sign of disease or pathological state that is detectable as local episodes in the image.

40 A very important aspect of the invention is the detection of any lesion of the fundus. Retinal lesions normally comprise microaneurysms and exudates, which appear in the images of the back of the eye generally as "dotted" areas (ie, substantially circular). It is interesting to differentiate between these microaneurysms and exudates, and also distinguish them from other lesions or pathological conditions in the image, such as "cotton wool stains" and hemorrhages.

Four. Five

Accordingly, the present invention relates to a method for assessing the presence or absence of injury or injuries in an image comprising

a) the estimation of at least one subset of the image, so that each subset is a candidate lesion zone 50 that has visibility, and the estimation of the background variation of the image in an area surrounding the

Candidate Injury Zone,

b) the correction of the visibility of the candidate lesion zone with the background variation, comparing the corrected visibility with a predetermined visibility threshold for the lesions of that area, or the correction of a threshold

55 of predetermined visibility with the background variation, comparing the visibility of the candidate injury zone with the corrected default visibility threshold,

c) the classification of the area of candidate injury detected in a) with respect to the threshold obtained in stage b) as an injury or not,

d) optionally repeating stages a) to c) until all areas of candidate lesions have been classified.

5 Through the procedure, the lesions detected are corrected with respect to the background substrate, in particular the local background substrate in close proximity to the lesion, in order to detect independent lesions on the background substrate in the specific image, which includes background variations in the images, for example due to the variable illumination of the different parts of the image.

Another aspect of the invention relates to the establishment of the area of candidate injury, by a procedure for assessing the presence or absence of injury or injuries in an image, which comprises

a) the establishment of at least one representative starting point of an image injury,

15 b) the selection of at least one starting point, estimating a subset of the image corresponding to a candidate lesion by isocurves that grow around the starting point, where q is an integer of at least 1, until the periphery of the candidate lesion, the isocurves extending from a vector of one or more visibility characteristics,

C) the determination of the visibility of the area of candidate injury, the visibility being determined from a visibility characteristic different from the visibility characteristics used for growth in stage b), and the comparison of the visibility of the lesion candidate with a predetermined threshold,

d) the classification of the area of candidate injury detected in a) with respect to the threshold obtained in stage b) 25 as an injury or not,

optionally repeating steps b) to d) until all candidate injury zones have been classified.

Through this aspect, the area or zones of candidate lesions grow in the form of isocurves from a starting point, and the starting point can be adjusted for structures of natural occurrence in the image, such as the vessels and the head of the optic nerve. .

In addition, the invention relates to a system for performing the procedures according to the invention, such as

a system to assess the presence or absence of injury or injuries in an image, which comprises

a) an algorithm to estimate at least a subset of the image, so that each subset is a candidate injury zone that has visibility, and to estimate the background variation of the image in a zone that

40 surrounds the area of candidate injury,

b) an algorithm to correct the visibility of the candidate lesion zone with the background variation, comparing the corrected visibility with a predetermined visibility threshold for the lesions of that area, or to correct a predetermined visibility threshold with the background variation , comparing the visibility of the area of injury

45 candidate with the default visibility threshold corrected,

c) an algorithm to classify the area of candidate injury detected in a) with respect to the threshold obtained in stage b) as an injury or not,

50 d) an algorithm to optionally repeat steps a) to c) until all candidate injury zones have been classified

as well as

55 a system for assessing the presence or absence of injury or injuries in an image, comprising

a) an algorithm to establish at least one representative starting point of an image injury,

b) an algorithm to select at least one starting point, estimating a subset of the image

corresponding to a candidate lesion by means of isocurves that grow around the starting point, being an integer of at least 1, until the periphery of the candidate lesion is established, the isocurves extending from a vector of one or more characteristics of visibility,

5 c) an algorithm to determine the visibility of the candidate lesion zone, the visibility being determined from a visibility characteristic different from the visibility characteristics used for growth in stage b), and to compare the visibility of the lesion candidate with a predetermined threshold,

d) an algorithm to classify the area of candidate injury detected in a) with respect to the threshold obtained in stage b) as an injury or not,

e) an algorithm to optionally repeat steps b) to d) until all areas of candidate lesions have been classified.

15 Such systems are capable of incorporating any of the variations of the procedures described herein.

In addition, the invention relates to a method for diagnosing the presence or absence of a disease in a subject from a background image of at least one eye of said subject comprising

twenty

- the evaluation of the presence or absence of at least one injury by the procedure as defined above,

- the categorization of the fundus image with respect to the number and / or size and / or location of the lesions,

25 - the diagnosis of the presence or absence of the disease.

The image can then be classified depending on the size and / or the numbers and / or the location of the lesions in the image, and accordingly, the invention relates to a method for classifying an image of the fundus of the eye comprising

- the evaluation of the presence or absence of at least one injury by the procedure as previously,

- the categorization of the fundus image with respect to the number and / or size and / or lesions,

35 - the classification of the fundus image in at least two classes.

All the procedures described herein are preferably for use in an automatic procedure, for example as included in a computer-readable program.

40 Drawings

Figure 1: Background image of the eye.

Figure 2: An organization chart showing the dividing line procedure.

Figure 3: Example 1-D of a pixel for the process using the dividing line algorithm with a tolerance of 1. 45 Figure 4: The figure shows a partial image of the fundus, where a circle is shown showing the region 75-pixel radio background of an injury.

Figure 5: The background region in the gradient image.

Figure 6: Organization chart for the normalization process.

Figure 7: Organization chart for the growth process.

50 Figure 8: Schematic drawing showing the calculation of the visibility characteristic.

Figure 9: An example of a developed lesion and the band around it representing the background substrate.

Figure 10: An example of overlapping regions. The left panel shows the three regions that grow from seed points located in the proliferation displayed in the right panel. The band located around the largest developed region is used as a background substrate for the three regions.

Definitions

Fovea: The term is used in its normal anatomical meaning, that is, the point of the retina that has a large

how is the location of the

Cane concentration that gives rise to vision. Fovea and the term "yellow spot" are used as synonyms.

Image: The term image is used to describe a representation of the region to be explored, that is, the term image includes 1-dimensional representations, two-dimensional representations, 3-dimensional representations as well as representations in n dimensions. Thus, the term image includes a volume of the region, a matrix of the region as well as a provision of information of the region.

Lesion in images of the fundus: Any pathological alteration present in the fundus, such as microaneurysms, exudates, hemorrhages, cotton wool stains. Preferably, the lesions refer to 10 lesions in a timely manner: microaneurysms and exudates.

Optic nerve head: The term is used in its normal anatomical meaning, that is, the area at the bottom of the eye where the optic nerve enters the retina. They are synonymous with the area, for example, “blind” point, papilla or optical disc.

fifteen

Red-green-blue image: The term refers to the image that has the red channel, the green channel and the blue channel, also called RVA.

Rdl: Region of interest.

twenty

Starting point: The term describes a point or zone to start the search for a subset. The term starting point is not thus limited to a mathematical point, for example it is not limited to a pixel, but simply denotes a location to start the search.

25 Visibility: The term visibility is used in the normal meaning of the word, that is, the extent to which an injury or structure of the region of the fundus is visible compared to the background substrate and other structures / lesions.

Detailed description of the invention

30

Images

The images of the present invention can be any kind of images and presentations of the region of interest. The fundus image is a conventional tool for exploring the retina and can be registered in any suitable medium. In one embodiment, the image is presented in a medium selected from slides, paper photographs or digital photographs. However, the image can be any other kind of representation, such as a presentation in an arrangement of elements such as, for example, a CCD.

The image can be a grayscale image or a color image; In a preferred embodiment the image 40 is a color image.

Subsets

The image establishes at least one subset, where the subset is a candidate injury zone. The term subset is used in its normal meaning, that is, one or more pixels.

The subset can be detected and established by any appropriate procedure, for example by filtering, by template correspondence, by the establishment of starting points and from said points of regions in starting growth and / or by other procedures for searching candidate areas. , and / or 50 combinations thereof. In one embodiment, the zone (s) of candidate regions are detected by establishing starting points, and from the starting points estimating the subset. Two or more subsets can be detected, each representing the same injury, such as overlapping subsets or adjacent subsets.

55 In a preferred embodiment the subset is a connected subset, that is, all the pixels of the subset are connected to at least one of the other pixels, and it is possible to reach any of the pixels from any of the pixels by following the pixels in the subset. In a preferred embodiment, the estimation of the subset of the image includes the establishment of the periphery of the subset. The periphery can be established for example by active contour models (snake) (reference "Snakes: Active contour models" of

M. Kass, A. Witkin and D. Terzopoulos), by staff training or by growth.

Establishment of the starting points

5 The starting points can be established by various suitable procedures and combinations of said procedures. The variability of the images of the fundus is especially relevant in relation to the dynamics of the images; the contrast can vary considerably from one image to another and even from one region to another in the same background image of the eye. An appropriate starting point algorithm must recognize this circumstance and seek to adapt its sensitivity to the image in question. The image can be filtered and / or blurred 10 before establishment or as a part to establish the starting points for the procedure. For example, the low frequencies of the image can be eliminated before setting the starting points. In addition, the image can be filtered to improve the sharpness, for example by filtering by average or by median of the image and subtraction of the filtered result of the image.

15 Regardless of whether the image is filtered or not, the starting points can be established as ends of the image, for example local extremes. However, preferably the image is a filtered image, where the filtering can be linear and / or nonlinear. Depending on the type of lesions evaluated, the extremes may be minimal or maximum or both,

20 In one embodiment the filtering procedure is a template matching procedure, where the template can display any suitable geometry to identify the lesions. Some examples of templates are the circles, where the circles have a fixed radius as a proportion of the expected diameter of the optic nerve head.

Within the scope of the invention is that the image can be filtered with one or more filters before establishing the starting points, or as a part of the stage of establishing the starting points. Thus, in an embodiment of the invention the starting points are established by combining two or more filters.

The ends can be identified individually by one or more of several procedures, such as the following:

The lesions are usually dark areas or light areas in the image, or at least locally the darkest areas or the lightest areas. Thus, a procedure can be to establish at least one extreme of intensity in the image, preferably at least a minimum of intensity or at least a maximum of intensity. 35 Therefore, in a preferred embodiment at least a maximum of local intensity is established. The ends may be set to any image function, for example where the image function is the untreated image to improve the sharpness, the image of the red channel, the image of the green channel, or any combination thereof. In a preferred embodiment the image function is the green channel.

Instead of using the intensity or in addition to using the intensity, the procedure may include the establishment of at least one end of variance in the image, preferably the establishment of at least a maximum of variance in the image. For the same reasons described with respect to intensity, at least a maximum of local variance is established. The ends may be set to any image function, for example where the image function is the untreated image to improve sharpness, the image of the red channel, the image of the green channel 45, or any combination thereof. In a preferred embodiment the image function is the green channel.

Another procedure to establish the starting points may be the random establishment of the starting points, where the final random establishment is the establishment of a starting point in substantially each of the pixels of the image. Naturally, a random establishment can be combined with any of the 50 procedures outlined above.

In a further embodiment, the starting points can be established as the points of a lattice, for example the points of a lattice evenly or unevenly distributed. Again, this procedure can be combined with any of the procedures for establishing endpoints in the image and / or random setting.

In a preferred procedure, the starting points are established by more than one of the procedures described in order to increase the probability of assessing the correct location of the lesions, if they exist, also with respect to images that have a suboptimal illumination or that present other quality forms of

suboptimal image. A problem that increases when the imagery of the fundus of the eye is registered in a non-central way and by a less experienced staff than in the case of specialized hospital services.

In a more preferred embodiment, the starting points are established by locating the minimum and / or the maximum 5 locations of the green channel image function, and making them act as starting points.

Increase

In a preferred embodiment the subset is established by developing a subset from a point of departure. The growth of an object is used to segment an object form with respect to the background substrate. The procedure can be used to develop dark and bright objects, the algorithm being for one simply an inversion of the algorithm of the others. The most essential part of the growth procedure is to limit the object with respect to the background substrate. This limitation can be done by any means, for example by examining the visibility characteristic as described below, for a wide variety of 15 isocurves, or depth of the object, and then simply selecting the depth, which produces the highest visibility characteristic. possible.

Thus, in this embodiment the establishment of subsets can be explained as the growth of isocurves based on at least one characteristic of the growing area around the starting point, where an integer number of at least 1 is q, until the periphery of the area of candidate injury. That is, for each starting point, a series of isocurves is developed, where each isocurve can represent a candidate lesion zone, starting from the starting point. In other words, the growth process can lead to the extraction of more than one subset, corresponding for example the number of subsets to uniformly separated isocurves. Preferably, the smallest subset exceeds that of the starting point in sf, and the area of the largest subset is greater than a predetermined value. However, normally only one of the isocurves establishes the periphery of the lesion, said isocurve being the isocurve that is most likely to be a candidate lesion zone. The probability may be, for example, the highest visibility as described below. The subset of the image thus comprises the region contained by an isocurve resulting from the growth process, and the isocurve in sf implies the periphery of the subset.

30

The growth algorithm is initialized at the starting point for the subset. By increasing the height in equidistant levels a series of developed isocurves is obtained. The depth of the stage can be set arbitrarily, although for practical reasons it is usually chosen as equal to 1, since pixel levels come from octet images, which have discrete integer values. In principle, the algorithm can continue 35 for the entire image from each starting point. However, again for practical reasons and because it is known that the lesions evaluated in the image have certain normal intervals of size, it is appropriate to apply at least a limitation to growth, specifically that areas of candidate lesions are not allowed above of a certain predetermined zone. In addition, another limitation may be applied additionally or alone, specifically that the candidate injury is limited by a minimum and maximum number of isocurves.

40

The predetermined value described above is preferably in the range of 0.1 to 1.0, for example in the range of 0.2 to 0.8, for example in the range of 0.3 to 0.6.

Figure 7 shows the growth procedure by means of an organization chart.

Four. Five

In particular in the present invention growth can be applied by using the dividing line procedures. The dividing line algorithm was introduced by Lantuejoul and Beucher to facilitate segmentation. The idea of dividing lines is understood by considering an image as a topographic surface. Image intensity (gray level) is considered as an altitude with this view. A minimum regional 50 is a connected plateau from which it is impossible to reach a lower gray level point through an always downward path. When the surface of the image is submerged, part of the flood zones (catchment basins) will tend to merge. When two or more different flood zones touch each other, infinitely high dikes (dividing lines) are built between them. When finished, the resulting networks of the dikes define the dividing line of the image. In other words, the dividing lines divide the image into areas of non-intersection, called catchment basins. Since each area contains only a minimum regional, the number of areas is equal to the number of regional minimums in the image.

In a preferred embodiment, the pixel, with minimum value and which is closer to the center of mass of the region, becomes the origin of the growth algorithm.

An example of the growth process is shown in the flow chart of Figure 2. The processing steps shown are described by the flowchart shown above with respect to Figure 3.

5 Stages 1, 3, 5 and 8:

Find minimal unprocessed pixel, and include adjacent pixel with the same value. Limits do not touch other regions.

Stages 2 and 6:

10 Include adjacent pixels that do not deviate more than the tolerance with respect to the starting pixel. The limits do not touch other regions with which a new region is assigned.

Stages 4 and 9:

Include adjacent pixels that do not deviate more than the tolerance with respect to the starting pixel. The limits touch 15 other regions with which they are expanded.

Stages 7, 10, 11, 12 and 13:

Find minimal unprocessed pixel, and include adjacent pixel with the same value. The limits touch other regions with what they expand.

twenty

The sensitivity of the dividing line algorithm can be adjusted by modifying the tolerance level, which makes it possible to exclude basins with an insignificant depth.

After having established the periphery of the candidate lesion zone, the zone can be filled, for example, by simply filling in flooding from the starting point to the periphery.

Visibility

The term visibility is used with the normal meaning of the word, that is, the extent to which an injury or a structure of the region of the fundus is visible is compared to the background substrate and other structures / lesions. The visibility of an area can be determined as a feature vector, which includes the intensity, visibility of the candidate lesion compared to the visibility of the vessels, the visibility of the edge of the candidate lesion, the color information of the candidate lesion, the measure of the variance of a part of the image and / or a measure of the variance of the image. In a preferred embodiment, the visibility of the edge of the candidate lesion 35 was calculated as the edge gradient of the targeted candidate injury area, in particular a weighted edge gradient.

Thus, the visibility characteristic can be based on a sum of gradient pixels of the weighted region boundary depending on the orientation. In particular, the gradient pixels must be weighted according to their orientation, a, towards the developed region, applying for example the following formula in the area of candidate injury:

1 N

Visibility = —V V

one

V, = c (a,) • || V (Intensity;) ||

c (a,) = jc ° s (a) • a

1-1, in other cases

where N is the number of pixels in the profile of the lesion. Let r, ■ the vector between the center of mass of the lesion and the image point in Vi then ai be the angle between rt and the orientation at the point.

Examples of the vectors are shown in Figure 8.

The visibility characteristic can also be calculated from an image where the vessels have been "removed." 50 For this purpose, the image of the vessel V (r, c) is subtracted from the original image I (r, c), and an image of the vessel is produced

"Restored", that is, an image where interpolated background substrate values have been introduced instead of the vessels, by

1VRes (r, C)

5

The interpolated value can be produced by

B (r, c, w (r, c)) I (r, c)

hvis V (r, c) = 1 hvis V (r, c) = 0

B (r, c, w)

2 I (r ', c') (1 -V (r ', c')) h (|| (r, c) - (r ', c') || / w)

(r ', c') __________________________________________

2 (l - V (r ', c')) h (|| (r, c) - (r ', c') || / w)

(r ', c')

10 where h (x) = exp (-x2 / 2). The amplitude of the core w (r, c) is set as the distance from the pixel (r, c) to the closest pixel of the substrate. By the same principle, other features of the image can be "removed". This "elimination" is done to avoid false positives of lesions in proximity to features such as, for example, the vessels.

15 Validation of subsets

Subsets can be validated before being corrected with respect to the background substrate. Validation means that each subset undergoes a validation stage to determine whether or not the candidate zone should be classified as a candidate injury zone before correcting it with the background variation. The validation is preferably carried out by means of a characteristic different from the one or characteristics in growth.

In one embodiment the validation stage includes the calculation of the visibility of the area of candidate injury.

Background variation

25

One of the main problems in detecting lesions in the images of the fundus lies in the enormously variable background substrates in which the lesions are found. Some areas of the fundus have an almost constant background substrate while some areas are extremely variable, for example areas with a visible layer of nerve fibers or choroidal structures. In accordance with one aspect of the invention, an advantage has been found in correcting, for example, normalizing, the visibility characteristic values obtained for each candidate injury zone according to an estimated robust background variation. The standardization procedure can be performed in at least two stages

estimating at least one subset of the image, so that each subset is a candidate injury zone 35 that has visibility, and after having assigned visibility characteristics to the candidate injury zone

estimating the background variation, and the correction of the visibility of the area of candidate injury.

40 However, in many embodiments the estimation of the subsets and the estimation of the background variation are carried out in one stage for each subset.

The background variation can be estimated by any suitable measure, whereby the background variation can be selected from the spatial and / or distribution properties of the original image, or any transformation thereof, such as a gradient image, a curvature image or a Laplace image.

The spatial properties can be based, for example, on a Fourier transformation, an occurrence matrix and a fractal dimension, and the distribution properties can be moments such as the mean, the asymmetry of the variance or the kurtosis.

Lesions can be described, for example, by a visibility characteristic as set forth above,

which is based on the observations of weighted limit of injury gradients in realization, it has been shown that it is advantageous to normalize the lesions characteristic of the average estimation and the typical deviation of the background gradient.

5 Thus, in a preferred embodiment the background variation is estimated by sequential identification example by

c1) the estimation of the mean and the standard deviation of the pixels of magnitude of the gradient of a defined area that surrounds the area of candidate injury, determining a lower threshold or above an upper threshold for 10 pixels of magnitude of the gradient ,

c2) the iterative elimination of a pixel of gradient magnitude of atypical value below a lower threshold or above a higher threshold, and the new estimation of the mean and the typical deviation of the remaining gradient magnitude pixels, the determination of a second lower threshold and a second upper threshold for 15 pixels of magnitude of the gradient, until no pixel of gradient magnitude of atypical value is found.

c3) the estimation of the background variation with respect to the mean and the standard deviation estimated in c2).

In this embodiment, the upper and lower thresholds are determined as a constant multiplied by the typical deviation, for example as the typical deviation multiplied by at least 2, for example at least 3, for example at least 4, for example at least 5 or for example at least 6. It is preferred that at each stage of iteration c2) at most one pixel is eliminated.

25 The defined area surrounding the candidate injury zone may include or exclude the candidate injury zone itself. In a preferred embodiment, the pixels of the magnitude of the gradient in step c1) include pixels of the area of candidate injury.

The area surrounding the candidate zone is normally selected in the range of 0.25 to 1.0 of the expected zone 30 of the optic nerve head, for example 0.5 to 1.0 of the expected zone of the head of the optic nerve, for example 0.6 to 1.0 of the expected area of the head of the optic nerve. Normally said zone corresponds to a number of pixels in the range of 100 to 100,000 pixels, for example in the range of 400 to 64,000 pixels, for example in the range of 1,000 to 50,000 pixels, for example in the range of 5,000 to 25,000 pixels. .

35 A preferred procedure for estimating background variation is described below, referring to the flow chart of Figure 6:

The first stage of normalization consists in estimating the background gradient of the lesion to correct. This estimate is made by the initial collection of the pixels within a given radius from the center of mass 40 of the lesion. The number of pixels is fixed according to the resolution of the image evaluated. For most purposes, the number of pixels is set at a radius of 50 to 100 pixels, see, for example, Figure 4.

Transverse vessels and / or other lesions could influence the estimation of a gradient of the background substrate, which requires a robust estimate of the background gradient. The background region in the gradient image is shown in Figure 5, from which the influence of the transverse vessels and / or other lesions is clear.

An example of robust estimation consists in the continuous elimination of atypical values. Therefore the gradient pixels in the defined background region are collected in a matrix, which is sorted according to their values. Next, the mean and typical deviation of this matrix are calculated. The value of each end 50 of the matrix is then compared with the mean and the standard deviation calculated and in case the most deviated of the two is an atypical value, it is eliminated from the matrix.

After having eliminated this observation, the mean and the standard deviation are recalculated, and the extremes are checked again. This cut is continued until no more atypical values are found. In this implementation, an atypical value is defined as a value that deviates more than two standard deviations from the mean.

After estimating the robust mean and the standard deviation of the background substrate of the lesion, the visibility of the lesion can be normalized using the standard formula:

orientation, and in this visibility with a

of atypical values, for

V ^ Gradient

^ Gradient + 1

Other robust procedures may consist of filtering the image before collecting the intensities or using robust estimators, such as the median instead of the mean and the absolute deviation of the mean instead of the typical deviation.

Once the background variation has been estimated, it is possible to correct the visibility of the candidate lesion zone with the background variation, and compare the corrected visibility with a predetermined visibility threshold for injuries in that area, or correct a threshold of predetermined visibility with the background variation, and compare the visibility of the candidate lesion zone with the corrected default visibility threshold. Through any of these stages it is possible to assign a local threshold for the lesions, thus increasing the specificity as well as the sensitivity of the procedures for assessing the presence and / or absence of an injury.

After comparison with the threshold obtained in stage b) the area of candidate injury is classified as 15 or not.

The steps of the procedures can be carried out in sequence or in parallel for all subsets.

Some of the structures of natural occurrence of the image can influence the evaluation of the lesions in a disadvantageous way. Such structures are for example vessels, and the head of the optic nerve of an image of the fundus, since these structures have dark / bright areas in the image. Therefore, a certain adjustment of the structure is preferred.

Adjustment with respect to the vessels

25

Several methods are known by which it is possible to isolate the vascular system from the rest of the image content.

Another procedure to keep track of the vessels is a procedure where the fact that the vessels are linear in the adjacent local area where different filter matrices have different orientations is used. The location and orientation of said linear elements can be determined using a template matching approach sometimes referred to as correspondence filters).

Other procedures for monitoring the vessels known to the person skilled in the art can be found in

Subhasis Chaudhuri et al., "Detection of Blood Vessels in Retinal Images Using Two-Dimensional Matched Filters", IEEE Transactions on Medical Imaging, Vol. 8, No. 3, September 1989.

40 Tolias et al. “A fuzzy vessel tracking algorithm for retinal images based on fuzzy clustering, IEEE Transactions on Medical Imaging, April 1998, IEEE; USA vol. 17, No. 2, pages 263-273, ISSN: 0278-0062.

Akita et al .: "A computer method of understanding ocular fundus images" Pattern Recognition, 1982, UK, vol. 15, n ° 6, pages 431-443, ISSN: 0031-3203 chapter 4.

Four. Five

A preferred procedure for tracking vessels is to track individual vessels from representative starting points of the vessels, and iteratively grow the network of retinal vessels. A preferred embodiment thereof is described in a PCT patent application under processing entitled "Vessel Tracking" of RETINALYzE A / S.

fifty

In some of the embodiments according to the invention, it is preferred that the estimation of starting points and / or the estimation of the subsets be adjusted with respect to the vessels that appear in the image. For many of these embodiments, it is still more preferred that the estimation of the areas of candidate lesions is preceded by the detection of the vessels in the image.

In one embodiment, adjustment of the starting points means that the starting points located in the vessels are removed from the plurality of representative starting points of an injury. In another embodiment the subsets of the

image that have at least a part of said subset located in a vessel are rejected as a candidate injury zone.

Another procedure for adjustment with respect to the vessels is that once the vessels of the image have been detected, the vessels that appear in the image are masked before establishing the starting points. The vessels can be masked by any suitable procedure, for example by masking a number of pixels along the vessel, for example a number in the range of 1 to 10 pixels.

Once the blood vessels are identified in the image, it may be convenient to distinguish between veins and 10 arteries within the blood vessels. This may be important, for example in the diagnosis of venous rosary and focal arteriolar stenosis.

The vascular system observed in the images of the fundus is by nature a two-dimensional projection of a three-dimensional structure. In principle it is quite difficult to distinguish veins from arteries, simply looking at isolated segments of the vessels. However, it has been discovered that effective separation can be achieved by making use of the fact that, individually, the structure of the arteries and the structures of the venous vessels are perfect trees (that is, there is a unique trajectory along the vessels from the heart to each capillary and vice versa).

20 In the retina, the structures of the arteries and veins are filled with surface, so that all tissues are irrigated or drained by specific arteries or veins, respectively.

A procedure for distinguishing veins from arteries is described in WO-00 / 65,982 for Torsana Diabetes Diagnostic A / S, which is based on understanding that crossings of vessel segments always occur, for practical purposes, between a vein and an artery (that is, for practical purposes, crossings between arteries and arteries or between veins and veins are non-existent).

Masking of the optic nerve head

30 Another structure capable of interfering with the evaluation of the lesions is the head of the optic nerve. Unlike the vessels, the optic nerve head is not necessarily present in all images, and depends on the region captured by the camera or the CCD.

Thus, in a preferred procedure the presence or absence of the optic nerve head area is evaluated by a robust procedure before assessing the lesions. Said procedure is described for example in a PCT patent application in process entitled "Detection of the head of the optic nerve in an image of the fundus" for RETINALYZE A / S.

In some of the embodiments according to the invention, it is preferred that the estimation of starting points and / or the estimation of the subsets be adjusted with respect to the head of the optic nerve that appears in the image. For many of these embodiments, it is still more preferred that the estimation of the areas of candidate lesions is preceded by the detection of the optic nerve head in the image.

In one embodiment, adjustment of the starting points means that the starting points located at the head of the optic nerve are removed from the plurality of starting points representative of an injury. In another embodiment, subsets of the image that have at least a portion of said subset located at the head of the optic nerve are rejected as a candidate injury zone.

Another procedure for adjustment with respect to the head of the optic nerve is that, when the head of the optic nerve has been detected in the image, the head of the optic nerve that appears in the image is masked before establishing the points of departure. The head of the optic nerve can be masked by any suitable procedure, for example by masking a number of pixels around the head of the optic nerve, for example a number corresponding to a constant multiplied by the diameter of the head of the optic nerve, optionally by a diameter expected from the head of the optic nerve, said constant being in the range of 1.1 to 2.0, preferably about 1.5.

In a further embodiment the method according to the present invention includes the weighting of visibility in relation to the variation of local intensity around the lesion in order to reduce false positives of the lesions due for example to layers of nerve fibers, vessels not monitored and reflections in the

glassy body. A common feature of these false positives is that the variation in local intensity around the lesion is relatively large, unlike most of the true lesions, which are located in homogeneous areas.

5 When two developed lesions overlap, one must be integrated into the other, so that the isointensity curves that define the limits of the lesions cannot cross. In this case, the background substrate can be defined slightly differently, in order to prevent the large lesion from being interpreted as a background substrate when the smallest interior lesion is evaluated.

10 When considering overlapping lesions, the “foreground” may be the complete connected component in the image of the lesion consisting of positive visibility pixels. The background substrate will be defined as before, but with respect to this region in the foreground. Thus, all overlapping lesions will have the same background region. The principle is illustrated in Figure 10.

15 If the background substrate is not defined in this way, the developed regions around the fovea and the large lesions can be classified erroneously since they usually have overlapping developed regions.

Therefore, it is preferred that visibility be weighted with a measure of the homogeneity of the local background substrate. The local or immediate background substrate can be defined as the band of pixels that are more than Bentrada and as maximum 20 Bentrada + BSaiida pixels from the injury. The distance between a point and the injury is defined as the smallest distance between the point and a pixel in the injury. Thus Bsaiida is the width of the background band substrate around the lesion, and Bentrada is the width of the band that separates the lesion and the background substrate. The principle is illustrated in the first page figure and in Figure 9. The parameters Bentrada and Bsaiida must be scaled according to the image scale.

25

Pixels can be excluded from the background substrate if they are located

1. In a follow-up glass,

2. Outside the Rdl,

30 3. Closer than Bentrada from another injury, which has a visibility greater than Tv, where v is the visibility of the current injury and T is a tolerance parameter.

The argument on which the exclusion of pixels is based on the first and second criteria must be clear. The third criterion is used to prevent true lesions that are located close to each other from influencing each other in their background substrates, and parameter T can be used to control the tolerance of this restriction. It has been found that setting T = 0 is significantly better than not using the criteria at all (which corresponds to T = ~); Choosing a T value of approximately 1.0 prevents injuries with small visibility of the background band being removed, which seems most prudent in practice.

In one embodiment, the ratio between the average intensity of the green channel in the background substrate and in the lesion can be used to discriminate true and false lesions. For example, a fixed threshold seems most appropriate,

'' green, exit> go

\ mbral

'green entrance

45 where IRumbrai is preferably less than 1.1 for example between 1.01 and 1.09, preferably between 1.04 and 1.08 to discriminate a true lesion from a false lesion, with a false lesion having a lower average intensity ratio to IRumbral. In this case pverde, saiida and pverde, entry are the means of the green channel in the immediate background substrate and in the lesion, respectively.

50 In another embodiment the measurement of the variance of a part of the image is estimated by

c4) the definition of a band of pixels of a predetermined width and that are at a predetermined distance from the area of candidate injury,

55 c5) the estimation of the mean and the standard deviation of the intensity of the band, and

c6) the estimation of the measure of the variance of an image part with respect to the mean and the standard deviation estimated in c5.

This can be illustrated by the proportion of the standard deviation in the background substrate and in the lesion that is used to detect lesions of typical false positives, in particular those located in the vessels and near the reflections. It has been found that, in fact, a measure of visibility weighted in the variance offers a useful approximation,

v =

v

(

to

2

poly exit

\

v

to

2

Poly J entry

■ 1/2

10 In this case v is the usual standardized visibility measure, and dpou, input and respectively dp0u, saiida is the variance of the intensities in the poly-smoothed image within the lesion and in the background substrate, respectively.

The measure of weighted visibility in variance is then compared with the predetermined visibility threshold as described above. In a preferred embodiment, the one that is weighted is the corrected visibility 15 as described above or the weighted visibility is compared with the corrected threshold.

Applications

In the following examples different applications of the procedure according to the invention are set forth.

twenty

Once the presence or absence of the lesions has been evaluated, the information regarding the injury can be used for different purposes.

Accordingly, the present invention further relates to a method for diagnosing the presence or absence of a disease in a subject from an image of the fundus of at least one eye of said subject comprising

- the evaluation of the presence or absence of at least one injury by the procedure as defined above,

30 - the categorization of the fundus image with respect to the number and / or size and / or location of the lesions,

- the diagnosis of the presence or absence of the disease.

In particular, this procedure refers to the diagnosis and prognosis of diabetic retinopathy.

35

In another aspect the invention relates to a method for classifying an image of the fundus comprising

- the evaluation of the presence or absence of at least one injury by the procedure as defined above,

- the categorization of the fundus image with respect to the number and / or size and / or location of the lesions,

- the classification of the fundus image in at least two classes.

45 Normally several classes are used, where the images are categorized with respect to the number of injuries and with respect to the distance of the lesions to the fovea.

System

50 In another aspect the invention also relates to a system for assessing the presence or absence of lesions in an image of the fundus. Thus, the system according to the invention can be any system capable of performing the procedure as described above as well as any combination thereof within the scope of the invention. Consequently, the system may include algorithms to perform any of the procedures described above.

A graphical user interface module can act in conjunction with a monitor display screen

display. The graphical user interface can be implemented as part of the processing system to receive data and input commands from a conventional keyboard and mouse through an interface and present the results on a display monitor. To simplify the explanation, many components of a conventional computer system such as address buffers, memory buffers and other standard control circuits 5 have not been exposed since these elements are well known in the art and a detailed description of the same to understand the present invention.

Pre-acquired image data can be supplied directly to the processing system through a network interface and stored locally on a mass storage device and / or in a memory. 10 In addition, image data can also be supplied on a network, through a portable mass storage medium such as a removable hard disk, optical discs, tape drives or any other type of data transfer and / or storage devices. They are known in the art.

One skilled in the art will recognize that a parallel computer platform that has multiple processors is also a hardware platform suitable for use with a system according to the present invention. Such configuration may include, but is not limited to, parallel machines and workstations with multiple processors. The processing system can be a single computer, or several computers can be connected through a communications network to create a logical processing system.

20 Any of the system algorithms described above can be adapted to the various variations of the procedures described above.

This system allows the evaluator, who is the person who normally categorizes the images, to identify the lesions more quickly and safely. In addition, the present system allows a detection

25 automatic lesions and other pathological states of the retina without interference from the vessels, again as

Help instrument for the traditional evaluator.

Using the present system it is also possible to organize the registration of the images in one place and their study in another place. For example the images can be recorded by any optician or doctor or other professional and be

30 transported to the examining specialist, in the form of photograffas or similar or in digital media. In

Consequently, using the present system the need to decentralize image registration centers can be met, while maintaining less expert evaluators.

On the other hand, in addition to the communication of images and medical information between people involved in the procedure, the network can carry data signals that include control signals or adjustment of the images by which the expert examining the images in the scanning unit directly controls the acquisition of images that takes place at the place of registration, that is, the acquisition unit. In particular, said command signals, such as zoom magnification, orientation settings and wavelength of the field illumination can be selectively modified remotely to achieve the desired effect of the image. Thus, the structures of 40 questionable tissues that require greater magnification or a different perspective for their elucidation can be quickly resolved without ambiguity by modifying said control parameters. In addition, changing the illumination wavelengths can be taken selectively to represent different layers of tissue, or to accentuate the image of vasculature characteristics and blood flow. In addition, when a specialized study is undertaken, such as fluorescence imaging, the control signals may include time variation signals to initiate stimulation with certain wavelengths of light, in order to initiate the study of image at certain times after stimulation or application of a dye or drug, or other similar imaging protocols with precise control. The digital data signals for these operations can be interconnected with the ophthalmic equipment in a relatively simple way, provided that such equipment already has initiation switches or internal digital circuits to control the specific parameters involved, or is able to easily adapt electrical controls to said control parameters such as system approach, lighting and the like.

In addition, the expert examiner might be able to apply some treatment in the same way remotely. It will be understood that in this case the instrumentation of the imaging study and the ophthalmic treatment will generally include an orientation and stabilization system that keeps the two instruments aligned and stabilized in the structures that appear in the field of vision. However, in view of the small but not negligible time delays that still intervene between the acquisition of the image and the start of the diagnostic or therapeutic activity at the exploration site, in this aspect of the invention, the invention contemplates that the system control also includes a software for identification and correlation of images that

allow the ophthalmologist located in the place to identify specific positions in the field of vision of the retina, such as precisely locate certain structures of vessels or tissues, and the image acquisition computer includes image recognition software that allows identifying patterns in the video frames and correlate the position identified with each image frame when it is acquired at the acquisition site. For example, the image recognition software can be set in a pattern of retinal vessels. Thus, in spite of the presence of sacks and other sudden eye movements of the small field of the retina that could occur during relatively short intervals of time, the ophthalmic instrumentation is directed to the place identified in the field of vision and remote treatment is achieved.

10 In addition to the previous operation, the invention also contemplates that the images provided by the acquisition unit are processed for photogrammetric analysis of the characteristics of the tissues and optionally the characteristics of the blood flow. This can be achieved as follows. An image acquired in the registration unit is sent to a scanning unit, where it is displayed on the screen. As schematically indicated in the figure, said image may include a network of blood vessels that have different diameters and lengths. These vessels include arterial and venous capillaries that constitute the blood supply and the return network. In the scanning unit, the workstation can be equipped with a photogrammetric measurement program that, for example, can allow the technician to place a cursor in a glass object of an image, and move the cursor along the glass while pressing the mouse button, with which the software automatically determines the width of the vessel and the sub-vessels to which it is connected, as well as the coordinates of the 20.

The software to write down the coordinates from the pixel positions and that link the characteristics shown in a register, as well as the sub-modules that determine the vessel's capabilities and the like, are simple and easy to build from photogrammetric program techniques. 25 workstation protocols can also be implemented to automatically map the vasculature as described above, or to compare two images taken at historically different times and identify or record the changes that have occurred, highlighting features such as erosion of the operator. the vessels, the tissue that has changed color or other differences. In addition, a graphical user interface allows the specialist to write diagnostic indications linked to the image, or to a specific feature that appears in a place in the image, so that the image or the processed version of it becomes more useful.

Thus, a very accurate and well annotated medical record can be easily compiled and compared with a previously taken view to obtain detailed evidence of the changes over a period of time, or it can be compared, for example, with immediately previous angiographic views in order to evaluate the actual degree of blood flow that occurs therein. Regarding the ophthalmologist's notes entries in the scanning unit, the measurements' entries in the scanning unit are converted into an annotated image register and stored in the central library as part of the patient's record.

Unlike a simple medical records system, the present invention modifies the dynamics of the patient's access to care, and the efficiency of the supply of ophthalmic knowledge in a way that solves a very important dilemma of current healthcare, specifically , the obstacle to access to a universal screening suitable for diabetic retinopathy. Having thus disclosed and described a basic embodiment of the invention, those skilled in the art will contemplate additional variations and modifications, and all these variations and modifications are within the scope of the invention as defined in the attached claims.

ELEMENTS

1. A procedure to assess the presence or absence of injury or injuries in an image, which

50 comprises

a) the estimation of at least one subset of the image, so that each subset is a candidate lesion zone that has visibility, and the estimation of the background variation of the image in an area surrounding the lesion zone candidate,

55 b) the correction of the visibility of the candidate lesion zone with the background variation, comparing the corrected visibility with a predetermined visibility threshold for the lesions of that area, or the correction of a predetermined visibility threshold with the variation of background, comparing the visibility of the candidate injury zone with the corrected default visibility threshold,

c) the classification of the area of candidate injury detected in a) with respect to the threshold obtained in stage b)

as an injury or not,

d) optionally repeating stages a) to c) until all areas of candidate lesions have been classified.

5 2. The procedure according to item 1, where the image is an image of the fundus.

3. The procedure according to any of the above elements, where the image is presented

in a medium selected from slides, photograffas on paper or digital photograffas.

10 4. The procedure according to any of the above elements, where the image is a

color image.

5. The procedure according to any of the above elements, where the green channel is used to assess the presence or absence of injury or injuries.

fifteen

6. The procedure according to any of the foregoing elements, where the subset (s) are estimated by establishing a plurality of starting points, said starting points being representative of injuries, and estimating each subset around a starting point.

20 7. The procedure according to any of the above elements, where the starting points

They are set at the ends of the image.

8. The procedure according to any of the above elements, where the subset of the image is a connected subset.

25

9. The procedure according to any of the above elements, where the subset of the image is estimated by filtering the image.

10. The procedure according to any of the above elements, where the estimation of the 30 subset of the image comprises the establishment of the periphery of the subset.

11. The procedure according to any of the above elements, where the subset is validated before the correction with respect to the background variation.

35 12. The procedure according to any of the above elements, where the subset of the

Image is estimated by developing an area around the starting point.

13. The procedure according to item 12, where the isocurves based on at least one

Characteristics of the growing area develop around the starting point, where an integer number of

40 at least 1, until the periphery of the candidate injury zone is established.

14. The procedure according to item 13, where the visibility of the area in the area is estimated

inside the isocurves, and the isocurve that has the highest visibility establishes the periphery of the lesion zone

candidate

Four. Five

15. The procedure according to any of elements 12 to 14, where the validation of the subset is performed by a characteristic different from the one or the characteristics in growth.

16. The procedure according to any of the above elements, where the visibility of a zone is determined as a feature vector that includes the intensity, visibility of the candidate injury

compared to the visibility of the vessels, the visibility of the edge of the candidate lesion, the color information of the candidate lesion, the measure of the variance of a part of the image and / or a measure of the variance of the image.

55 17. The procedure according to any of the above elements, where the subset of the

Image is estimated by an active contour model.

18. The procedure according to any of the previous elements 6 to 17, where the identification

The starting point is adjusted with respect to the vessels that appear in the image.

19. The procedure according to any of the above elements, where the estimate of

Subset is adjusted with respect to the vessels that appear in the image.

5 20. The procedure according to any of the above elements, where the estimate of the

The subsets are preceded by the detection of the vessels in the image.

21. The procedure according to any of the above elements 6 to 20, where the points of

Departures located in the vessels are removed from the plurality of starting points representative of an injury.

10

22. The procedure according to any of the above elements, where the subsets of the image having at least a part of said subset located in a vessel is rejected as a candidate injury zone.

15 23. The procedure according to any of the above elements, where the vessels that

they appear in the image are masked before setting the starting points.

24. The procedure according to any of the previous elements 2 to 23, where the identification of the starting point is adjusted with respect to the head of the optic nerve that appears in the image.

twenty

25. The procedure according to any of the previous elements 2 to 24, where the estimate

The subset is adjusted with respect to the head of the optic nerve that appears in the image.

26. The procedure according to any of the above elements, where the estimate of the

25 the subsets and / or the identification of a starting point is preceded by the detection of a region that

It comprises the head of the optic nerve.

27. The procedure according to any of the preceding elements 6 to 26, where the starting points located at the head of the optic nerve are removed from the plurality of starting points representative of a

30 injury

28. The procedure according to any of the preceding elements 2 to 27, where the subsets of the image having at least a part of said subset located at the head of the optic nerve are rejected as a candidate injury zone.

35

29. The procedure according to any of the above elements 2 to 28, where the region of the head of the optic nerve is masked before the estimation of the subset and / or the establishment of the starting point.

40 30. The procedure according to any of the above elements, where the area surrounding the

Candidate injury zone includes the candidate injury zone.

31. The procedure according to any of the above elements, where the area surrounding the candidate injury zone excludes the candidate injury zone.

Four. Five

32. The procedure according to any of the above elements, where the area surrounding the candidate zone is selected within the range of 0.25 to 1.0 the expected area of the optic nerve head, for example 0, 5 to 1.0 the expected area of the optic nerve head, for example 0.0.6 to 1.0 the expected area of the optic nerve head.

fifty

33. The procedure according to any of the above elements, where the background variation is selected from the spatial and / or distribution properties of the original image, or any transformation thereof, such as a gradient image, an image of curvature or an image of Laplace.

55 34. The procedure according to any of the above elements, where the background variation

It is estimated by

c1) the estimation of the mean and the standard deviation of the pixels of magnitude of the gradient of a defined area that surrounds the area of candidate injury, determining a lower threshold or above a higher threshold for

the magnitude pixels of the gradient,

c2) the iterative elimination of a pixel of gradient magnitude of atypical value below a lower threshold or above a higher threshold, and the new estimate of the mean and typical deviation of the remaining gradient magnitude pixels, the determination of a second lower threshold and a second upper threshold for 5 pixels of magnitude of the gradient, until no pixel of gradient magnitude of atypical value is found.

c3) the estimation of the background variation with respect to the mean and the standard deviation estimated in c2).

35. The procedure according to any of the preceding elements 16 to 34, where the measure of variance of a part of the image is estimated by

c4) the definition of a band of pixels of a predetermined width and that are at a predetermined distance from the area of candidate injury,

c5) the estimation of the mean and the standard deviation of the intensity of the band, and 15 c6) the estimation of the measure of the variance of an image part with respect to the average and the standard deviation estimated in c5.

36. The procedure according to element 28, where the upper and lower thresholds are determined as a constant multiplied by the standard deviation.

twenty

37. The procedure according to element 28 or 29, where the pixels of magnitude of the gradient in step c1) include the pixels of the area of candidate injury.

38. The procedure according to any of the elements 28 to 30, where a maximum of one pixel is eliminated in step c2).

39. A procedure for assessing the presence or absence of injury or injuries in an image, comprising

30 a) the establishment of at least one representative starting point of an image injury,

b) the selection of at least one starting point, estimating a subset of the image corresponding to a candidate lesion by isocurves that grow around the starting point, where q is an integer of at least 1, until the periphery is established of the candidate lesion, the isocurves extending from a vector of one or several visibility characteristics,

C) the determination of the visibility of the candidate lesion zone, the visibility being determined from a visibility characteristic different from the visibility characteristics used for growth in stage b), and the comparison of the visibility of the lesion candidate with a predetermined threshold,

d) the classification of the area of candidate injury detected in a) with respect to the threshold obtained in stage b) as an injury or not,

E) optionally the repetition of stages b) to d) until all areas of candidate lesions have been classified.

40. The procedure according to element 39, where the image is an image of the fundus.

45 41. The procedure according to any of the above elements 39 to 40, where the image is

Presents in a medium selected from slides, photograffas on paper or digital photograffas.

42. The procedure according to any of the above elements 39 to 41, where the image is a color image.

fifty

43. The procedure according to any of the above elements 39 to 42, where the green channel is used to assess the presence or absence of injury or injuries.

44. The procedure according to any of the above elements 39 to 43, where the points of departure are set at the ends of the image.

45. The procedure according to any of the preceding elements 39 to 44, where the subset of the image is a continuous subset.

46. The procedure according to any of the above elements 39 to 45, where the subset of the image is estimated by filtering the image.

47. The procedure according to item 46, where the visibility of the area in the area is estimated

5 inside the isocurves, and the isocurve that has the highest visibility establishes the periphery of the lesion zone

candidate

48. The procedure according to any of the above elements 39 to 47, where visibility

of an area is determined as a feature vector that includes the intensity, visibility of the injury

10 candidate compared to the visibility of the vessels, the visibility of the edge of the candidate injury, the color information of the candidate injury, the measure of the variance of a part of the image and / or a measure of the variance of the image .

49. The procedure according to any of the above elements, which also includes the correction of the visibility of the area of candidate injury with the background variation, comparing the visibility

corrected with a predetermined visibility threshold for injuries in that area, or the correction of a predetermined visibility threshold with the background variation, comparing the visibility of the candidate injury zone with the corrected predetermined visibility threshold, classifying the lesion zone candidate detected with respect to the threshold obtained as an injury or not.

twenty

50. The procedure according to any of the above elements 39 to 49, where the identification of the starting point is adjusted with respect to the vessels that appear in the image.

51. The procedure according to any of the above elements 39 to 50, where the estimate 25 of the subset is adjusted with respect to the vessels that appear in the image.

52. The procedure according to any of the previous elements 39 to 51, where the estimation of the subset (s) is preceded by the detection of the vessels in the image.

30 53. The procedure according to any of the above elements 39 to 52, where the points of

Departures located in the vessels are removed from the plurality of starting points representative of an injury.

54. The procedure according to any of the preceding elements 39 to 53, where the subsets of the image having at least a part of said subset located in a vessel are rejected.

35 as a candidate injury zone.

55. The procedure according to any of the above elements 39 to 54, where the vessels that appear in the image are masked before establishing the starting points.

40 56. The procedure according to any of the above elements 40 to 55, where the

Starting point identification is adjusted with respect to the head of the optic nerve that appears in the image.

57. The procedure according to any of the above elements 40 to 56, where the estimate

The subset is adjusted with respect to the head of the optic nerve that appears in the image.

Four. Five

58. The procedure according to any of the previous elements 40 to 57, where the estimation of the subset (s) and / or the identification of a starting point is preceded by the detection of a region comprising the head of the optic nerve.

50 59. The procedure according to any of the above elements 40 to 58, where the points of

Departures located at the head of the optic nerve are removed from the plurality of starting points representative of an injury.

60. The procedure according to any of the above elements 40 to 59, where

55 subsets of the image that have at least a portion of said subset located in the head of the nerve

Optically rejected as a candidate injury zone.

61. The procedure according to any of the above elements 40 to 60, where the region of the head of the optic nerve is masked before the estimation of the subset and / or the establishment of the point of

departure.

62. A procedure for diagnosing the presence or absence of a disease in a subject from an image of the fundus of at least one eye of said subject comprising

5

- the evaluation of the presence or absence of at least one injury by the procedure as defined in any of items 1 to 61,

- the categorization of the fundus image with respect to the number and / or size of the lesions,

- the diagnosis of the presence or absence of the disease.

10

63. The procedure according to element 62, where the disease is a diabetic retinopathy.

64. A procedure for classifying a fundus image comprising the eye

15 - the evaluation of the presence or absence of at least one injury by the procedure as defined in any of items 1 to 60,

- the categorization of the fundus image with respect to the number and / or size of the lesions,

- the classification of the fundus image in at least two classes.

20 65. A system for assessing the presence or absence of injury or injuries in an image, comprising

a) an algorithm to estimate at least a subset of the image, so that each subset is a candidate lesion zone that has visibility, and to estimate the background variation of the image in an area surrounding the lesion zone candidate,

25 b) an algorithm to correct the visibility of the candidate lesion zone with the background variation, comparing the corrected visibility with a predetermined visibility threshold for the lesions of that area, or to correct a predetermined visibility threshold with the variation of background, comparing the visibility of the candidate injury zone with the corrected default visibility threshold,

c) an algorithm to classify the area of candidate injury detected in a) with respect to the threshold obtained in stage b) as an injury or not,

d) an algorithm to optionally repeat steps a) to c) until all areas of candidate lesions have been classified.

66. A system for assessing the presence or absence of injury or injuries in an image, comprising

35

a) an algorithm to establish at least one representative starting point of an image injury,

b) an algorithm to select at least one starting point, estimating a subset of the image corresponding to a candidate lesion using isocurves that grow around the starting point, where q is an integer of at least 1, until the periphery of the candidate injury, extending the isocurves

40 from a vector of one or more visibility characteristics,

c) an algorithm to determine the visibility of the candidate injury zone, the visibility being determined from a visibility characteristic different from the visibility characteristics used for growth in stage b), and to compare the visibility of the candidate injury with a predetermined threshold,

d) an algorithm to classify the area of candidate injury detected in a) with respect to the threshold obtained in stage b) as an injury or not,

e) an algorithm to optionally repeat steps b) to d) until all candidate injury zones have been classified

Claims (12)

I. A procedure for assessing the presence or absence of injury or injuries in an image, comprising 5 a) the establishment of at least one representative starting point of an image injury, b) the selection of at least one starting point, estimating a subset of the image corresponding to a candidate lesion by isocurves that grow around the starting point, where q is an integer of at least 1, until the periphery is established from the area of candidate injury, extending the isocurves from 10 a vector of one or more visibility characteristics, c) the determination of a value of the visibility characteristic for the visibility of the candidate lesion zone, the value of the visibility characteristic being determined from a visibility characteristic different from the visibility characteristics used for the growth in the stage b), and comparing the value of the visibility characteristic for the visibility of the area of candidate injury with a predetermined threshold, D) the classification of the area of candidate injury detected in b) with respect to the predetermined threshold as a visible or not visible lesion, e) optionally the repetition of stages b) to d) until all areas of candidate lesions have been classified. The method according to claim 1, wherein the image is an image of the fundus. 3. The method according to any of the preceding claims, wherein the points of heading are set at the ends of the image. 25 4. The method according to claim 1, wherein the visibility of the area in the area is estimated. inside the isocurves, and the isocurve that has the highest visibility establishes the periphery of the area of candidate injury. 5. The method according to any of the preceding claims, wherein the visibility of A zone is determined as a feature vector that includes the intensity, the visibility of the candidate lesion zone compared to the visibility of the blood vessels, the visibility of the border of the candidate lesion zone, the color information of the zone of candidate injury, the measure of the variance of a part of the image and / or a measure of the variance of the image. The method according to any of the preceding claims, which comprises in addition to the correction of the visibility of the area of candidate injury with the background variation, comparing the corrected visibility with a predetermined visibility threshold for the lesions of that area, or the correction of a predetermined visibility threshold with the background variation, comparing the visibility of the candidate injury zone with the corrected predetermined visibility threshold, classifying the candidate injury zone detected 40 with respect to the threshold obtained as an injury or not. 7. The method according to any of the preceding claims, wherein the identification The starting point or points are adjusted with respect to the blood vessels that appear in the image. The method according to any of the preceding claims, wherein the estimation of the Subset conforms to the blood vessels that appear in the image. 9. The method according to any of the preceding claims, wherein the starting points located in the blood vessels are removed from the plurality of representative starting points of a 50 injury 10. The method according to any of the preceding claims, wherein the subsets of the image having at least a portion of said subset located in a blood vessel are rejected as a candidate injury zone. 55 II. The method according to any of the preceding claims, wherein the identification of the starting point or points is adjusted with respect to a head of the optic nerve that appears in the image. 12. The method according to any of the preceding claims, wherein the estimation of the Subset is adjusted with respect to an optic nerve head that appears in the image. 13. The method according to any of the preceding claims, wherein the starting points located at the head of the optic nerve are removed from the plurality of starting points representative of a 5 injury 14. The method according to any of the preceding claims, wherein the subsets of the image having at least a portion of said subset located at the head of the optic nerve are rejected as a candidate injury zone. 10 15. The method according to any of the preceding claims, wherein at least one of the one or more visibility characteristics used for the growth of isocurves is based on a sum of gradient pixels of the weighted region boundary depending on the Weighted orientation according to its orientation towards the developed region. fifteen 16. A procedure for classifying a fundus image comprising - the evaluation of the presence or absence of at least one injury by the procedure as defined in any of claims 1 to 15, 20 - the categorization of the fundus image with respect to the number and / or size of the lesions, - the classification of the fundus image in at least two classes. 17. A system for assessing the presence or absence of injury or injuries in an image, comprising 25 a) an algorithm to establish at least one representative starting point of an image injury, b) an algorithm to select at least one starting point, estimating a subset of the image corresponding to a candidate lesion using isocurves that grow around the starting point, where q is an integer of at least 1, until the periphery of the candidate lesion zone, the isocurves extending from a vector of one or more visibility characteristics, C) an algorithm to determine a value of the visibility characteristic for the visibility of the area of candidate injury, determining the value of the visibility characteristic from a visibility characteristic different from the visibility characteristics used for growth in step b), and comparing the value of the visibility characteristic for the visibility of the area of candidate injury with a predetermined threshold, d) an algorithm to classify the area of candidate injury detected in b) with respect to the predetermined threshold 35 as a visible or not visible lesion, e) an algorithm to optionally repeat steps b) to d) until all areas of candidate lesions have been classified.