Device and method for classifying the grade and / or type of cataracts
The computerized method and system for classifying cataracts using specific light frequencies and image processing techniques address measurement errors in existing methods, offering an objective and accurate evaluation of cataract grade and type for improved surgical planning and lens selection.
Patent Information
- Application Number
- PCT/IB2024/054719
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2023-05-16
- Filing Date
- 2024-05-15
- Publication Date
- 2024-11-21
AI Technical Summary
Current methods for classifying cataracts are prone to measurement errors due to the opacification of the lens, affecting the accuracy of surgical methods and artificial lens selection, as they rely on visual inspection or incomplete automated systems.
A computerized method and system that processes lens images using light radiation between 400nm and 490nm, specifically focusing on pixel values to classify cataracts by distinguishing healthy and deteriorated lens structures, with image acquisition and processing techniques to resize and classify regions of interest.
This approach reduces measurement errors by providing an objective evaluation of cataract grade and type, improving the accuracy of surgical planning and artificial lens selection, with high efficiency in classifying cataracts as demonstrated by statistical validation.
Smart Images

Figure IB2024054719_21112024_PF_FP_ABST
Abstract
Description
[0001] DEVICE AND METHOD FOR CLASSIFYING THE GRADE AND / OR TYPE OF CATARACTS
[0002] DESCRIPTION
[0003] TECHNICAL FIELD
[0004] The present invention refers to a device and a method for classifying the type and / or grade of cataract.
[0005] BACKGROUND
[0006] Cataract is a disease that leads to the progressive opacification of the lens.
[0007] The correct evaluation of the type and grade of cataract is very important, both to determine the surgical method to be applied for the removal of the lens (e.g. by means of the "phacoemulsification" technique), and to determine the choice of the artificial lens to be implanted.
[0008] The characteristics of the artificial lens, in particular its dioptric power, are chosen on the basis of complex mathematical formulas that take into account the conformation of the eye in which it is to be implanted. The measurement of the ocular characteristics that must be taken into account is, however, affected by measurement errors that depend on the presence of cataract. Optimizing the system of classification of type and grade of cataract allows to reduce errors in measurements and to determine more accurately the artificial lens to be implanted.
[0009] To date, cataracts are classified through a visual inspection by the doctor, who, based on the colour of the lens, assigns a value indicative of the severity of the case. Visual classification methods are the Oxford Clinical Cataract Classification and Grading System and the Lens Opacity Classification System (LOCS, in the versions LOCS I, LOCS II and LOCS III).
[0010] In order to assist the doctor in evaluating the grade of cataract, some automated systems have been proposed.
[0011] International patent application W02010052576 describes a method which provides for: (i) creating a three-dimensional image of at least the majority of the lens of an eye affected by cataract starting from information received from an image acquisition system, (ii) comparing the three-dimensional image with a first model to determine the optical density of the lens (measured as backscattered light), and (iii) determining the grade of cataract based on the comparison between the optical density of the lens and the volume of the first model. The solution described provides for the use of an image acquisition system that uses the Scheimpflug principle, but does not provide other information about the light used to illuminate the lens.
[0012] International patent application W02009142601A1 describes a method executed by a computer system for detecting opacity in an image of the lens of an eye. The method provides for detecting a region of interest in an image of the lens and processing the region of interest to produce a modified image using an algorithm that emphasizes opacity associated with a cortical cataract over opacity caused by other types of opacity, such as opacity caused by posterior subcapsular cataract (PSC). The modified image may be used to classify the level of cortical opacity by measuring, in the modified image, the percentage of opacity in at least one area of the region of interest. The described solution focuses on the lens image processing method without giving indications on how such an image was acquired.
[0013] Chinese patent application CN113469936A provides for acquiring images of the lens with Optical Phase Coherence Tomography (OCT) to reconstruct a three-dimensional model of the lens, and classifying the cataract based on the grey values of the tomographic images of the model. The patent application does not go into the merits of how OCT images are acquired, but in general this technique provides for the use of broadband light radiation with frequency in the infrared field.
[0014] OBJECTS AND SUMMARY OF THE INVENTION
[0015] An object of the present invention is to overcome the drawbacks of the prior art.
[0016] In particular, it is an object of the present invention to present a system and method for classifying cataract that is of assistance to the doctor's diagnosis.
[0017] These and other objects of the present invention are achieved by a system and a method incorporating the features of the attached claims, which form an integral part of the present description.
[0018] According to a first aspect, the present invention is directed to a computerized method for classifying a cataract, wherein the method provides for processing the image of a lens to determine the grade of cataract based on the pixel values of the image, and wherein the image of the lens is acquired by focusing on the lens a light radiation with a frequency between 400nm and 490nm, and preferably between 450nm and 470nm and more preferably equal to 460nm.
[0019] This solution offers the advantage that light radiation in this frequency range is absorbed by a healthy lens, while the absorption is progressively reduced as the structures of the lens itself deteriorate, i.e. as the grade of cataract grows. By focusing the light in a narrow zone of the lens, this zone will absorb all the radiation and, therefore, the image acquisition system records a scattered radiation equal to zero, or in any case very low. If, on the other hand, the illuminated zone of the lens has degenerated, then the same zone will diffuse blue light and the image acquisition system detects a non-zero intensity signal whose value is indicative of the grade of cataract.
[0020] The method may be performed on digital images stored in the form of data files in a storage system, or be performed on images acquired by an appropriate image acquisition system.
[0021] Preferably, the image of the lens is acquired by scanning the lens. The light radiation is focused in different zones of the lens and for each zone illuminated by the light radiation, an intensity value of the scattered radiation is acquired. This allows a quality image to be obtained for the evaluation of the grade and / or type of cataract.
[0022] Advantageously, the step of processing the image of the lens comprises: (i) a step of identifying a region of interest, which corresponds to a region of the image occupied by the lens, (ii) a step of generating the image of an equivalent lens, said step providing for resizing the region of interest to a predefined size, (iii) a step of classifying the cataract (in particular grade and / or type) based on the pixel values internal to the resized region of interest.
[0023] This solution offers the advantage that the evaluation of the grade and / or type of cataract is performed on a standard-sized image, thus offering an evaluation as objective as possible.
[0024] Preferably, the cataract classification step provides for identifying the pixels of the resized region of interest that represent a healthy eye. Subsequently, a morphological index My is calculated as the ratio between the number of pixels representing a healthy lens and the area of the resized region of interest, wherein the area of the resized region consists of the number of pixels present in the resized region of interest. Then it is verified whether the morphological index exceeds a first morphological threshold value to , and if so, classify the lens of the image as healthy if the verification has given a positive result.
[0025] Advantageously, if the morphological index does not exceed the first morphological threshold value, a second verification is carried out in which it is verified (a) whether the morphological index Mfexceeds a threshold valuet7associated with a grade j of cataract different from non-zero and (b) whether the average of the values of the pixels representing a healthy eye exceeds a threshold value if associated with said grade of cataract, and (c) whether the average of the values of the pixels representing a diseased eye exceeds a threshold value associated with said grade of cataract. Therefore, the image of the lens in which all three conditions from (a) to (c) are satisfied is classified as a grade j cataract.
[0026] Preferably, the cataract that does not satisfy any of the conditions from (a) to (c) is classified as maximum grade.
[0027] In one embodiment, the step of identifying the region of interest comprises the steps of
[0028] - dividing the pixels of the acquired image into two classes of pixels, generating a binarized image by assigning to the pixels of one class a first value, in particular a value corresponding to the colour black, and assigning to the pixels of the other class a second value, in particular a value corresponding to the colour white,
[0029] - cleaning up the binarized image by removing cavities and filling holes by applying closing and hole-filling operators to the binarized image,
[0030] - identifying the maximum area object within the cleaned binarized image,
[0031] - identifying the centre C and radius r of the circle that best approximates the contours of the maximum area object,
[0032] - identifying the region of interest in the set of pixels of the acquired image that lie within a circle of centre C and radius r.
[0033] In another embodiment, the step of identifying the region of interest comprises the steps of generating a normalised image whose pixels have normalised values PN calculated as follows: wherein Pi is the value of the i-th pixel of the acquired image, PN,iis the value of the i-th pixel after normalisation, Pmax is the maximum value among the pixel values of the acquired image, Pmin is the minimum value among the pixel values of the acquired image, R is the number of discrete values that a pixel of the normalised image can assume,
[0034] - dividing the pixels of the normalised image into two classes of pixels,
[0035] - generating a binarized image by assigning to the pixels of one class a first value, in particular a value corresponding to the colour black, and assigning to the pixels of the other class a second value, in particular a value corresponding to the colour white,
[0036] - cleaning up the binarized image by removing cavities and filling holes by applying closing and hole-filling operators to the binarized image,
[0037] - identifying the maximum area object within the cleaned binarized image,
[0038] - identifying the centre C and radius r of the circle that best approximates the contours of the maximum area object,
[0039] - identifying the region of interest in the set of pixels of the acquired image that lie within a circle of centre C and radius r.
[0040] In one embodiment, the step of generating the image of an equivalent lens comprises the steps of:
[0041] - generating a normalised image of the region of interest whose pixels have normalised values PL calculated as follows: wherein Pc,i is the value of the i-th pixel of the region of interest, Pi,i is the value of the i-th pixel of the region of interest after normalisation, Pcmax is the maximum value among the pixel values of the region of interest, Pcmm is the minimum value among the pixel values of the region of interest, R is the number of discrete values that a pixel of the normalised image can assume, - resizing the normalised region of interest to a predefined size.
[0042] In one embodiment, the method comprises (i) a step of identifying a region of interest, said region of interest corresponding to a region of the image occupied by the lens, (ii) a step of generating the image of an equivalent lens, said step providing for resizing the region of interest to a predefined size, (iii) a step of classifying the type of cataract based on the pixel values within the resized region of interest.
[0043] Preferably, the step of classifying the type of cataract provides for dividing the region of interest into concentric sectors, calculating an energy of each sector (7004), wherein the energy of a sector corresponds to the square of the pixels' value in each sector, identifying the sector with the highest energy, classifying the type of cataract based on the identified sector.
[0044] According to a further aspect, the invention is directed to a system for classifying a cataract, comprising
[0045] - a head positioning module suitable for receiving a patient's head and positioning it,
[0046] - a source module capable of emitting electromagnetic radiation to focus it on a lens of the patient, a detector module suitable for acquiring the image of the lens illuminated by electromagnetic radiation,
[0047] - a control unit configured to control the source module and the detector module, and to process the image of the lens to determine the grade of cataract based on the pixel values of the image.
[0048] The source module is suitable for generating electromagnetic radiation with a frequency between 400nm and 490nm, and preferably between 450nm and 470 nm and more preferably equal to 460nm.
[0049] Advantageously, then, the control unit is configured to perform a method as indicated above and as better detailed in the description below.
[0050] In particular, in one embodiment, the control unit is configured to control the source module and the detector module in such a way as to scan the lens, focusing in different zones of the lens the electromagnetic radiation and acquiring, for each zone illuminated by the electromagnetic radiation, an intensity value of the scattered radiation.
[0051] Advantageously, then, the control unit is configured to
[0052] - identify a region of interest within the acquired image, said region of interest corresponding to a region of the acquired image occupied by the lens, generate an image of an equivalent lens by resizing the region of interest to a predefined size,
[0053] - classify a grade of cataract, based on the pixel values within the resized region of interest. Preferably, in order to classify the cataract the control unit is configured to
[0054] - identify the pixels of the resized region of interest representing a healthy lens,
[0055] - calculate a morphological index Mf as a ratio between the number of pixels representing a healthy lens and the area of the resized region of interest, wherein the area of the resized region consists of the number of pixels present in the resized region of interest.
[0056] - verify (6003) whether the morphological index exceeds a first morphological threshold value tg , and if so classify the lens as healthy.
[0057] Advantageously, if the morphological index does not exceed the first morphological threshold value, the control unit is configured to perform a second verification wherein it verifies
[0058] (i) whether the morphological index Mf exceeds a threshold valuer7associated with a grade j of cataract different from non-zero and
[0059] (ii) whether the average of the pixel values representing a healthy eye exceeds a threshold value associated with said grade of cataract, and
[0060] (iii) whether the average of the pixel values representing a diseased eye exceeds a threshold value associated with said grade of cataract,
[0061] The control unit classifies as a grade j cataract the image of the lens in which all conditions from (i) to (iii) are satisfied.
[0062] Advantageously, the control unit is configured to use the threshold values reported in the table below:
[0063] Preferably, the control unit is configured to generate a maximum absorption index, associated with a maximum grade of cataract, which does not satisfy any of the conditions from (i) to (iii).
[0064] According to an embodiment, in order to identify the region of interest, the control unit is configured to
[0065] - divide the pixels of the acquired image into two classes of pixels,
[0066] - generate a binarized image by assigning to the pixels of one class a first value, in particular a value corresponding to the colour black, and assigning to the pixels of the other class a second value, in particular a value corresponding to the colour white,
[0067] - clean up the binarized image by removing cavities and filling holes by applying closing and hole-filling operators to the binarized image,
[0068] - identify the maximum area object within the cleaned binarized image,
[0069] - identify the centre C and radius r of the circle that best approximates the contours of the maximum area object,
[0070] - identify the region of interest in the set of pixels of the acquired image that lie within a circle of centre C and radius r.
[0071] In one solution, in order to identify the region of interest, the control unit is configured to generate a normalised image whose pixels have normalised values PN calculated as follows: wherein Pi is the value of the i-th pixel of the acquired image, PNJ is the value of the i-th pixel after normalisation, Pmax is the maximum value among the pixel values of the acquired image, Pmin is the minimum value among the pixel values of the acquired image, R is the number of discrete values that a pixel of the normalised image can assume,
[0072] - divide the pixels of the normalised image into two classes of pixels,
[0073] - generate a binarized image by assigning to the pixels of one class a first value, in particular a value corresponding to the colour black, and assigning to the pixels of the other class a second value, in particular a value corresponding to the colour white,
[0074] - clean up the binarized image by removing cavities and filling holes by applying closing and hole-filling operators to the binarized image,
[0075] - identify the maximum area object within the cleaned binarized image,
[0076] - identify the centre C and radius r of the circle that best approximates the contours of the maximum area object,
[0077] - identify the region of interest in the set of pixels of the acquired image that lie within a circle of centre C and radius r.
[0078] In one embodiment, for generating the image of an equivalent lens, the control unit is configured to: generate (5010) a normalised image of the region of interest whose pixels have normalised values PL calculated as follows: wherein Pc,i is the value of the i-th pixel of the region of interest, PL,I is the value of the i-th pixel of the region of interest after normalisation, Pcmax is the maximum value among the pixel values of the region of interest, Pcmin is the minimum value among the pixel values of the region of interest, R is the number of discrete values that a pixel of the normalised image can assume,
[0079] - resize the normalised region of interest to a predefined size.
[0080] Advantageously, the control unit is further configured to
[0081] - identify a region of interest, said region of interest corresponding to a region of the image occupied by the lens,
[0082] - generate the image of an equivalent lens, said step providing for resizing the region of interest to a predefined size,
[0083] - classify a type of cataract based on the pixel values within the resized region of interest,
[0084] - wherein the control unit is configured to classify the type of cataract by performing an algorithm comprising the steps of
[0085] - dividing the region of interest into concentric sectors,
[0086] - calculating an energy of each sector, wherein the energy of a sector corresponds to the square of the pixels' value in each sector,
[0087] - identifying the sector with the highest energy,
[0088] - classifying the type of cataract based on the identified sector.
[0089] In a further embodiment, the system is also configured to acquire BAF or OCT images of the ocular fundus. In this embodiment, the system is configured to modify the ocular fundus image acquisition parameters based on the grade of cataract determined according to the method described above and further detailed in the description below.
[0090] In particular, in one embodiment the system is configured to change the frequency of the electromagnetic radiation used to acquire the ocular fundus image, based on the previously determined grade of cataract. For example, in the case of OCT images of the ocular fundus, the system preferably uses a different frequency for each grade of cataract according to the LOCS III scale. Alternatively, the system uses at least two different frequencies for two different grades of cataract, e.g. 125kHz for grades 0 and 1, 20kHz for grades from 2 to 4. The range of potentially usable frequencies can vary from 250kHz to 10kHz.
[0091] In one embodiment, the system is configured to provide an alert that signals to the user the poor quality of the BAF images of the ocular fundus and the consequent risk of nonuse in clinical practice and in research settings , if the previously determined grade of cataract is greater than a predefined value, for example it is greater than 2 or 3. Alternatively to alerting the operator, in one embodiment the system blocks BAF image acquisition if the previously determined grade of cataract is greater than a predefined value, e.g., greater than 2 or 3.
[0092] Still, in one embodiment the system is configured to correct the ocular fundus images acquired by the system itself, to increase the signal-to-noise ratio thereof by a factor dependent on the grade of cataract measured, in particular a factor fa(for BAF images) or fb (for OCT images) as reported below in Table V.
[0093] Further features and objects of the present invention will be more evident from the description below.
[0094] BRIEF DESCRIPTION OF THE DRAWINGS
[0095] The invention will be described below with reference to some examples, provided for explanatory and non-limiting purposes, and illustrated in the attached drawings. These drawings illustrate different aspects and embodiments of the present invention and reference numerals illustrating structures, components, materials and / or similar elements in different drawings are indicated by similar reference numerals, where appropriate.
[0096] Figure 1 illustrates a block diagram of a system for classifying a cataract.
[0097] Figure 2 illustrates a flowchart of a method for cataract classification implementable through the system of figure 1.
[0098] Figure 3a illustrates an image of a lens.
[0099] Figure 3b illustrates an image of the lens of figure 3a taken at a different instant of time.
[0100] Figure 4 illustrates the model of an equivalent lens.
[0101] Figure 5 illustrates a flowchart of a method for implementing the steps of lens segmentation and of generation of the equivalent lens of the method of Figure 2.
[0102] Figure 6 illustrates a flowchart of a method for implementing the cataract classification step of the method of figure 2.
[0103] Figure 7 illustrates an image of a lens,
[0104] Figure 8 illustrates the image of figure 7 after contrast enhancement,
[0105] Figure 9 illustrates the image of figure 8 after a binarization process,
[0106] Figure 10 illustrates the image of figure 9 after processing by means of a closing operator,
[0107] Figure 11 illustrates the image of figure 10 after processing by means of a hole-filling operator,
[0108] Figure 12 illustrates the image of figure 11 after selection of the maximum area object,
[0109] Figure 13 illustrates the image of figure 1 after application of a circular model calculated on the image of figure 12,
[0110] Figure 14 illustrates the result of an operation of segmentation of the image of figure 1 by means of the circular model calculated on the image of figure 12,
[0111] Figure 15 illustrates the image of figure 14 after contrast enhancement, Figure 16 illustrates the image of figure 15 after a lens image expansion process,
[0112] Figure 17 illustrates the image of figure 16 after a noise removal process,
[0113] Figure 18 illustrates the image of figure 17 and two images obtained from it for cataract classification.
[0114] Figure 19 illustrates a flowchart of a process for classifying the type of cataract.
[0115] Figure 20 schematically illustrates a lens on which an annular mask used in the process of Figure 19 is applied.
[0116] Figure 21 illustrates images of five eyes with different grades of cataract, for each eye an image of the lens acquired with the system of figure 1, a BAF image and an OCT image of the ocular fundus are shown.
[0117] Figure 22 illustrates images of two lenses, acquired with the system of figure 1, of two patients with white cataracts.
[0118] DETAILED DESCRIPTION OF THE INVENTION
[0119] Some preferred embodiments will be described in detail below, although the invention is susceptible to various alternative modifications. It must in any case be understood that there is no intention to limit the invention to the specific embodiment illustrated, but, on the contrary, the invention intends covering all the modifications and equivalents that fall within the scope of the invention as defined in the claims.
[0120] Unless otherwise defined, all the terms of the art, notations and other scientific terms used herein are intended to have the meanings commonly understood by those who are skilled in the art to which this description pertains. In some cases, terms with commonly understood meanings are defined herein for the sake of clarity and / or for ease of reference; the inclusion of such definitions in the present description should thus not be interpreted as representing a substantial difference from what is generally understood in the art.
[0121] The terms "comprising", "having", "including" and "containing" are to be understood as open-ended terms (i.e. the meaning of "comprising, but not limited to") and are to be considered as a support also for terms like "consist essentially of", "consisting essentially of", "consist of" or "consisting of".
[0122] The use of "for example", "etc.", "or" indicates non-exclusive alternatives without limitation, unless otherwise indicated. The use of "includes" means "includes, but not limited to" unless otherwise indicated.
[0123] The term 'pixel' will refer to the smallest portion of a digital image such as, for example, a JPEG or BITMAP or PNG file. In a known way, each pixel is digitally represented by a plurality of bits and has a value that can represent the intensity of a colour (e.g. Red, Green, Blue) or an intensity of light, as in grayscale images. Figure 1 schematically illustrates a system 1 according to an embodiment of the present invention.
[0124] The system comprises a first head positioning module 100 suitable for receiving and positioning a patient's head. The module 100 comprises a lower element 101 for supporting the patient's chin, and an upper element 102 for supporting the patient's forehead 10. Positioning modules of this type are known per se and will not be described in more detail. The purpose of this head positioning module is to allow the correct positioning of the patient and induce them to look towards the image acquisition module 200 that has to capture an image of the lens.
[0125] The image acquisition module 200 comprises a source module 201 capable of emitting electromagnetic radiation with a frequency between 400nm and 490rtm, and preferably between 450nm and 470nm and more preferably equal to 460nm. For example, such a source module may comprise LEDs or a laser capable of emitting radiation in the frequency range indicated above.
[0126] The source module 201 is configured to illuminate the lens 11 of the patient 10, in such a way that the detector module 202 can acquire the image thereof. The detector module comprises one or more sensors, typically an array of sensors, capable of receiving reflected and / or backscattered radiation from the lens. Preferably the detector module is configured to detect radiation with a frequency between 400nm and 550nm.
[0127] The source module 201 and the detector module 202 may be, for example, elements of a confocal microscope capable of acquiring the image of the lens.
[0128] In order to acquire the image of the lens 11, the source module 201 and the detector module 202 may comprise, in a known manner, optical elements (such as lenses, diaphragms, beam-splitters, prisms) which are necessary to direct, and preferably to focus, the radiation emitted by the source module on the lens and allow the acquisition of the image by the detector module.
[0129] In order to allow correct focusing even in the case of patients with a different face, the image acquisition module 200 comprises a movement module 203 capable of moving the image acquisition module along at least two orthogonal axes (indicated with x and y in figure 1), so as to move it closer to or away from the head positioning module 100 and lift or lower it with respect to the latter. Preferably, the movement means 203 allow a micrometric adjustment that can be obtained by manual operation of a knob (not illustrated in the figure), or can be obtained electronically by reading, by a control unit, a command provided by the operator, for example the pressing of a forward or backward button.
[0130] The image acquisition module 200 further comprises a control unit 204 operatively connected to the source module 201, to the detector module 202 and to the movement module 203 in case this is electronically operable. The control unit 204 is configured to control the source module 201 and the detector module 202. For example, the control unit may be configured to slightly change the direction of emission of the electromagnetic radiation exiting from the source module 201 so as to illuminate different portions of the lens and the detector module may in turn be moved to ensure correct image acquisition. In one embodiment, the control unit 204 is configured to command the modules 201 and 202 to obtain a lens scan. For example, in the case of a pulsed source, the movement of the source module 201 may be done between the emission of one pulse and the other. The complete image of the lens is then generated by merging the images obtained during the scan.
[0131] In the example of figure 1, the image acquisition module 200 further comprises a user interface 205 comprising command input means (such as buttons and / or touch display) by means of which the operator can control the image acquisition module, and audio / video output means. In particular, the user interface 205 may comprise a display (LED, OLED, LDC or other type) on which the acquired image of the lens and other information, such as choice menus and information on the grade of cataract, is displayed.
[0132] In a known manner, the image acquisition module 200 may be connected to one or more external devices, for example by direct wiring to a computer 300. or by connection to a data network 400. To this end, the image acquisition module 200 comprises a data interface 206 which may comprise one or more connectors and / or wireless network cards.
[0133] Both the data interface 206 and the user interface 205 are operatively connected to the control unit 204.
[0134] In one embodiment, the control unit 204 is configured to execute code portions stored in a memory unit 207 accessible by the control unit itself. In the example of figure 1, the memory unit 207 is stored in the image acquisition module 200, however, according to other embodiments, such memory unit may be external to the module 200, for example it may be a memory unit connected to the network 400.
[0135] When the control unit 204 executes the code portions stored in the memory unit 207, the method illustrated hereinbelow is executed with reference to the flowchart of figure 2.
[0136] The method begins (step 2001) with the acquisition of the image (I) of a lens by the image acquisition module 200; to this end, the control unit 204 controls the source module 201 and the detector module 202 as a function of the operator commands received via the user interface 205. For example, in one embodiment the operator will be able to command the focusing of the device to ensure that it has a good image of the lens. The control unit 204 will then command the source module 201 and the detector module 202 to acquire a scan of the lens: the electromagnetic radiation is from time to time focused in different zones of the lens and, for each zone illuminated by the electromagnetic radiation, an intensity value of the scattered radiation is acquired. The image of the lens is then reconstructed by putting together the acquisitions of the different zones of the lens. An example of the image I of the lens is reported in figure 7.
[0137] As visible in figure 3(a) and 3(b), which show the images of a lens at two different time instants, the image of the lens may have different resolution (in the sense that the lens may occupy a more or less large portion of the image) and different brightness. In this example, the images acquired by the image acquisition device 200, and illustrated in figures 3(a) and 3(b), show the negative of the detected brightness. The images are thus represented in 8-bit greyscale, where the value 0 indicates black and the value 255 white. Since the image I is the negative of what is detected by the detector module, the healthy parts of the lens, those absorbing blue radiation emitted by the source module well, will be white or close to white. In other embodiments, the images may have different colouring, for example the pixels of the image representing healthy parts of the lens may be dark and have a value equal to or close to 0.
[0138] The method then provides for a second step (step 2002) which provides for the segmentation of the acquired image I, i.e. for the identification, within the acquired image I, of the region of interest occupied by the lens and for its extraction.
[0139] Subsequently, (step 2003) the method then provides for generating the image E of an equivalent lens, which we will indicate below as ECL (Equivalent Crystalline Lens), on which to make the measurements necessary for classifying the cataract. As shown in figure 4, the image E is a square image of SxS pixels (e.g. 1024x1024 pixels) in which a circular profile ECL lens with radius r equal to half the height (S) of the image is inscribed. Outside the ECL the pixels are all set to 0 and, in this example, they therefore appear to be black.
[0140] The cataract classification is then performed (step 2004) by processing the image E of the equivalent lens to determine the grade of cataract based on the pixel values of the image. Going into more detail, and with reference to the flowchart of figure 5, the acquired image I is first processed to enhance contrast, for example by means of an image histogram equalization method or by means of a normalisation method. In the example considered here, the image is normalised (step 5001), i.e. the pixel values are rescaled in order to use the full range of available values. Hereinbelow we will describe the case in which the acquired image I is an 8-bit grayscale image (illustrated in figure 7), wherein the pixels can assume a value between 0 and 255, where 0 is the black colour and 255 the white colour.
[0141] The normalisation step 5001 then provides for identifying the minimum (Pmin) and maximum (Pmax) values of the pixels present in the image I, in order to reconstruct an image IN whose pixels have normalised values PN calculated as follows:
[0142] Wherein Pi is the value of the i-th pixel of the image I, and P ,I is the value of the i-th pixel after normalisation.
[0143] Obviously, should the pixels of the image I have a different dynamic range, that is, if they could assume discrete values in a range R (calculated as Pmax-Pmin) other than 255, the equation (1) above should be modified by substituting the value of R in place of 255.
[0144] The result of the normalisation is an image IN illustrated in figure 8.
[0145] The normalised image IN is then processed (step 5002) using the Otsu method (known per se) to automatically identify the threshold that allows the pixels of the images to be classified into two distinct classes, minimizing the intra-class variance. Other classification methods can be used, as an alternative to the Otsu method, such as for example one from among the "K-means" method, the "watershed" method, the "graphbased" method, the "region growing" method, and methods based on "Convolutional Neural Network".
[0146] Once the separation threshold between the two classes has been determined, the method provides for generating a binarized image (step 5003) by assigning to the pixels of one class a first value, in particular a value corresponding to the colour black, and by assigning to the pixels of the other class a second value, in particular a value corresponding to the colour white. The result of the binarization is the image reported in figure 9.
[0147] The combination of the methods of Otsu plus binarization allows to discriminate the lens with respect to the background, however, in addition to eliminating the background, with this process also parts internal to the lens are eliminated. For this reason, the image is preferably cleaned in order to identify as precisely as possible the area occupied by the lens.
[0148] In order to clean up the image, the method first of all (step 5004) provides for applying a morphological closing operator. The closing operator dilates the white part and then narrows it making sure that the convex parts are closed; this allows to close small holes present within the image of the lens. In the example considered here, the result of the closing operation on the binarized image of figure 9 is the image reported in figure 10.
[0149] Once the closing operation has been completed, any black holes present in the area of the lens are closed with a hole-filling operation (step 5005) using a filling operation (flood-fill algorithm known per se). The result of the hole-filling operation is illustrated in figure 11.
[0150] The method then provides for eliminating small objects (step 5006), for example small groups of white pixels that are clearly not part of the lens and are the residue of previous processing. In this way, the image of the lens is transformed as much as possible into a white circle. This operation of removing small objects can be performed, for example, by using an opening morphological operation which, as is known, removes small objects and lines from an image while preserving the shape and size of larger objects. Other classification methods, which can be used as an alternative to the opening morphological operator, are for example methods based on connected component labelling (CCL) and the methods based on "Convolutional Neural Network".
[0151] Once the image has been cleaned up, the method provides for identifying (step 5007) the maximum area object within the image, i.e. the lens. This can be done in an automated way using a special algorithm known per se, such as calculating the area (number of pixels) for each object in the image and extracting the object with maximum area. In the case of figure 11, the image after the hole-filling operation has only one object, so even after the operation of removing small objects (step 5006), the result of step 5007 is an image, illustrated in figure 12, with a single object.
[0152] Subsequently, once the maximum area object has been identified, the method (step 5008) provides for identifying the circle that best approximates the identified object. This operation can be performed, for example, by determining the centroid of the maximum area object obtained in step 5007:
[0153] Where C is the centroid of the object, xtare all the ordinates of the pixels belonging to the object, yt are the abscisses of the pixels belonging to the object, A is the area of the object. The centroid represents the centre of the circle. The radius of the circle is determined by the area A of the object:
[0154] This operation therefore makes it possible to identify the centre C and the radius r of the circle that approximates the maximum area image identified above, i.e. the lens present in the image.
[0155] Once the circle that approximates the lens has been identified, the data belonging to the circle of centre C and radius r is extracted (step 5009) from the image I . In other words, the area occupied by the lens and identified by the circle determined in step 5008 is extracted from the image I. This extraction process is also called lens segmentation. Figure 13 allows to appreciate, visually, the application of the circular model to the image I, whereas figure 14 represents the segmentation result, i.e. an image with black background and with a single circular object inside it.
[0156] The lens data extracted from the image I are processed with a known contrast enhancement technique (e.g., but not only, image normalisation method, image equalization method, etc.). For example, we consider applying the normalisation method (step 5010) so that a wider dynamic range can be used. The maximum (Pcmax) and minimum (Pcmax) values of the data extracted in step 5009 are then determined, i.e. the grayscale data that make up the lens extracted from the image I. Subsequently, the normalisation is performed, thus generating an image of the lens LN whose pixels have values PL,; determined as follows: (2)
[0157] Wherein Pc,i is the value of the i-th pixel of the image I belonging to the circle of centre C and radius r, and PL,; is the value of the i-th pixel after normalisation. Similarly to what was said for equation (1), in the case of an image with a dynamic range R other than 255, equation (2) should be rewritten by replacing R to the value 255.
[0158] After normalisation, the image of figure 14 assumes the appearance of figure 15.
[0159] At this point, the method provides for resizing (step 5011) the image of the lens in such a way as to obtain the image of the equivalent lens, i.e. to obtain an image of the lens with a radius equal to half the height of the final image. This operation can be done with known image oversampling algorithms, such as, for example, cubic spline interpolation algorithms. Visually, the resized image is illustrated in figure 16.
[0160] Finally, the method preferably, but not necessarily, provides for performing a step (5012) of reducing the noise from the image. This step can, operationally, be performed using one or more filters for the reduction of noise in the images chosen from among the known techniques. For example, the image obtained in step 5011 could be filtered with an adaptive Wiener filter. Figure 17 illustrates the image of figure 16, after this noise reduction operation with Wiener filter.
[0161] Once the image of the equivalent lens has been extracted, the method provides for classifying the cataract, for this purpose a classification algorithm is implemented that we illustrate below with reference to figure 6. The algorithm of figure 6 may then be used to implement step 2004 of the method of figure 2.
[0162] After starting (step 6000), the classification algorithm provides for initializing the values of some parameters of the algorithm ), as follows.
[0163] First of all, the image of the equivalent lens is separated into two images IB (Bright) and ID (Dark) by means of a known thresholding algorithm (e.g. local adaptive thresholding) or manually set by the operator (e.g. global thresholding, local adaptive thresholding with manual choice of thresholding sensitivity). Figure 18 makes it possible to appreciate, one next to the other, the image of the equivalent lens (1800), the image IB (1801) and the image ID (1802).
[0164] The image IB contains the pixels classified as bright (i.e. with value above the adaptive threshold) in the lens, the image ID contains the pixels classified as dark (i.e. below the adaptive threshold, local or global, used for the generation of the two images IB and ID) in the lens. The pixels of the image IB therefore represent, for the purposes of the algorithm, pixels considered as representative of a healthy eye, while the dark pixels of the image ID are considered as representative of a diseased eye. Mf is a morphological index and indicates the bright portion of the lens (and therefore assumed to be healthy) with respect to the surface of the lens. Mf is then calculated as the ratio between the number of pixels representing a reasonably healthy eye, i.e. with brightness above a certain threshold value, and the total number of pixels constituting the image of the equivalent lens. The index Mf can then be calculated as ' '
[0165] Where "area" is an operator that counts the number of pixels of the object contained in the image (i.e. the number of total pixels minus the background pixels, i.e. those with a 0 value).
[0166] EBis an index of the average value of the pixels of the image IB; mathematically EBit can be expressed as follows:
[0167] (4) where Bi indicates the value of the i-th pixel of the image IB.
[0168] EDis an index of the average value of the pixels of the image ID; mathematically EDit can be expressed as follows:
[0169] <5> where Di indicates the value of the i-th pixel of the image ID.
[0170] The parameters , with integer j between 0 and 4, are a set of fifteen predetermined threshold parameters. Empirical tests conducted by the Applicant on different training images, identified as preferred the following values of the threshold parameters-able to maximize the probability of correct classification of the grade of cataract (grade 0, 1, 2, 3, or 4). These values could be varied to adapt as the images subject to analysis validated by medical report grow.
[0171] Table I j
[0172] 0 1 2 3 4 b 0.97 0.96 0.92 0.82 0.72 b - 0.7 0.65 0.6 0.5 b - 0.5 0.45 0.4 0.3
[0173] The threshold parameters indicated above may be varied by more or less than 5%.
[0174] Once the initialization of the parameters necessary for the algorithm has been ended, it continues by setting to 0 the value of j (step 6002) and by verifying (step 6003) whether the shape index Mf is higher than the value of the shape threshold parameter provided for grade 0 cataract (i.e. healthy eye).
[0175] If the condition is satisfied, the algorithm returns (step 6004) a value 0 for the indicator CG of the grade of cataract, and the algorithm ends (step 6005).
[0176] If, on the other hand, the condition M is not satisfied, the algorithm provides for increasing the value of j by one unit (step 6006) and verifying (step 6007) whether the following three conditions are satisfied simultaneously: (6)
[0177] If the above reported three conditions are satisfied, then the algorithm returns (step 6008) the value j for the indicator CG of the grade of cataract, and the algorithm ends (step 6005).
[0178] If the above reported three conditions are not satisfied, then the algorithm verifies (step
[0179] 6009) if j has reached the maximum value (4 in this case). If so, the algorithm returns (step
[0180] 6010) a value 4 for the indicator CG of the grade of cataract, and then ends (step 6005). If not, the algorithm returns to step 6006, j is incremented by one unit, and the above reported three conditions in (6) are verified again.
[0181] In one embodiment, the control unit 204, in addition to performing the method illustrated above for determining the grade of cataract, is also configured to perform a method for performing a first classification of the type of cataract.
[0182] In this embodiment, after generating the image of the equivalent lens (step 5011) and possibly removing noise as illustrated above with reference to step 5012, the control unit performs a method for classifying the type of cataract illustrated below with reference to figure 19.
[0183] The method for classifying the type of cataract is based on the analysis of the radial distribution of the pixel intensity levels. This distribution is calculated by considering a set of concentric rings, of thickness t, of the image of the equivalent lens. Figure 20 shows a ring 2001 of the equivalent lens 2000 and the parameters of interest used in the method for classifying the type of cataract described below. The ring has a thickness t and is defined by the inner circumference of radius a^ = Xi — t / 2 and the outer circumference of radius bi= xi+ t / 2, where x, indicates the radius of the circumference passing in the middle of the ring.
[0184] After starting (step 7000), the method for classifying the type of cataract provides for initializing (step 7001) the values of some parameters (x0, d, t), as follows:
[0185] • t is the thickness of the ring measured in number of pixels, e.g. t=10 pixels
[0186] • x0is the initial radius, also measured in pixels, and is preferably equal to x0= t / 2 such that the first ring considered is, actually, a circle centered in C and of radius x0• • d is the difference between the radii of adjacent rings, preferably d = t = xi— xi-1, so as to have adjacent rings; however, d may be different from t and two consecutive rings may be partially superimposed or separated.
[0187] The values of these parameters may be entered by a system operator through the user interface 205, or be predefined and then retrieved by the control unit from a local or remote memory area.
[0188] Subsequently, step 7002, the method provides for initializing the variables i and r for the ring analysis cycle:
[0189] • i is the ring counter and, in this example, it assumes integer values between 1 and N
[0190] • r is the radius of the i-th ring
[0191] Once the process parameters and the variables have been initialized, the method, step 7003, calculates a binary mask M of the ring, that is, an image with pixels equal to 1 in the ring area (grey area in figure 20) and equal to 0 for the other pixels.
[0192] Once the mask has been defined, the method provides, step 7004, for using it to calculate the energy of the pixels of the image within the i-th ring considered. In detail, for each i- th ring, the energy Ei is calculated as the summation of the square of the level of pixels that overlap the binary mask. Mathematically, this can be expressed as follows: where Mj indicates the value (1 or 0) of the j-th pixel of the mask M and Pj indicates the value of the pixel (in the example considered here between 0 and 255).
[0193] The value of Ei is entered as the i-th element of a vector ETof size N.
[0194] Once the energy Ei has been calculated, the method provides for (step 7005) incrementing the counter i by one unit to move on to processing the ring i+1 that follows.
[0195] Before moving on to processing the i+l-th ring, the method verifies (step 7006) whether the counter i is higher than the maximum number (indicated with N in figure 19) of rings to be processed. If not (i<N) then (step 7007) the method increments the radius corresponding to the ring that follows by a quantity d, i.e. the new value of r is set to r+d and the cycle is repeated from step 7003.
[0196] Once the maximum number of rings has been reached, i=N, the method (step 7008) provides for partitioning the vector ETinto Z vectors , preferably of equal size, and, for each partition, calculating the total energy of the partition , as the sum of the elements of the partition vector. For example, if and Z = 3, the partitions are and the total energy per partition is
[0197] Being the energy of each individual partition known, the method (step 7009) provides for determining the type of cataract as the index of the partition with minimum energy, that is, the index p e [1,Z] corresponding to the minimum of E . Compared to the previous example, the type of cataract (CT) will be
[0198] The method ends at step 7010 by providing in output the type of cataract, for example by giving a visual representation thereof via the user interface. In one embodiment the system may colour the pixels of the selected partition.
[0199] The value of CT provides an indication of the type of cataract with respect to the radial distribution of the amplitude of the pixel levels. In particular, CT = 1 indicates the presence of greater opacification in the central zone of the lens and may be associated with a nuclear cataract, CT=Z indicates greater opacification at the periphery of the lens and may be associated with a cortical cataract. Obviously, the diagnosis will then be left to the doctor.
[0200] EXPERIMENTAL RESULTS
[0201] In order to validate the cataract classification method described above, images of 4 healthy eyes (indicated by a grade of cataract CG=0) and of 30 eyes with different grades of cataracts (from 1 to 4) were acquired. The images were classified using the computerized method that returned the GC classification values reported in the right column of Table II below. The same images were subjected to visual evaluation by an ophthalmologist, who in turn classified the grade of cataract as reported in the first left column of Table II.
[0202] The test was conducted on 8-bit images, i.e. images in which each pixel can assume a grayscale value between 0 and 255. The images are in negative, that is, the bright pixels correspond to the region of the lens and the dark ones to zones of degeneration of the lens or to the background of the image.
[0203] As can be appreciated, from the results of Table II, the grade automatically calculated by the algorithm described above corresponds to the grade assigned by the ophthalmologist in 31 out of 34 cases, therefore with an efficiency of 91.2% which is very high.
[0204] The three errors, highlighted in bold in Table II, concerned "mild" cataracts, of grade 1 or 2, and saw the algorithm underestimate the cataract by one grade compared to the ophthalmologist's visual evaluation. Table II
[0205] Disregarding the three cases in which the evaluation of the algorithm was different from that of the ophthalmologist (in which the images were of lower quality than the others and therefore unreliable), statistical analyses of the values Mf, Eb, Ed were performed. Statistical analyses were performed using frequency analysis, the two-sided T-test and the one-way Anova test with Bonferroni correction for multiple comparisons (SPSS, Chicago, Illinois, USA), with statistical significance set at p<0.05. Tau-Kendall correlation analysis (SPSS, Chicago, Illinois, USA) was used to assess the relationship among all the variables considered. The results of the statistical analyses are reported in Table III below, where for each subgroup of LOCS III grading and for each of the indices of the values Mf, Eb, Ed, the average value E, the standard deviation S, the standard error s.e., and the significance p are reported. Table III
[0206] The significant result concerned the values Mf, which were statistically different between all subgroups, with the only exception of the comparison between cataracts with gradation LOCS III 0 and 1 where the p-significance was greater than 0.05.
[0207] The values My and Eb showed statistically significant correlations with LOCS III gradation (Tau-Kendall Coeff. -0.86; p<0.001 and Tau-Kendall Coeff. -0.51; p<0.001, respectively).
[0208] The images of the lens acquired with the system described above were compared with BAF (Blue-Light Fundus Autofluorescence) and OCT (Optical Coherence Tomography) images of the fundus of the same eye. A visual comparison is shown in figure 21, where for each cataract of grade LOCS III from 0 to 4, the image 210 acquired with the system described above, the BAF image 211 of the fundus of the eye and the OCT image 212 of the fundus of the eye are shown.
[0209] Table IV below reports the average value E and the standard deviation S of the signal-to- noise ratio (SNR) measured for the BAF and OCT images of the fundus of the eye for each gradation LOCS III from 0 to 4. The same table also reports the significance value (p) measured.
[0210] Table IV
[0211] The data in Table IV show that as the grade of cataract increased, the SNR decreased significantly for both the BAF images of the fundus and for the OCT images of the fundus. The BAF image of the fundus was not acquired for grade 4 LOCS III cataracts. This phenomenon was also highlighted by the statistical correlation values. In detail, both the SNR values of the BAF images of the fundus and those of the OCT images were significantly correlated with the values of Mf (Tau-Kendall Coeff. 0.89; p<0.001 and 0.87; p<0.001, respectively) and with the values of Eb (Tau-Kendall Coeff. 0.69; p<0.001 and 0.67; p<0.001, respectively) calculated for the images acquired with the system described herein. The data of Table IV are particularly interesting as they show a correlation between the value of the ratio SNR of the image of the ocular fundus (either BAF or OCT) and the LOCS III grade of cataract. This data can then be used in a variant of the system 1 for classifying the cataract.
[0212] In this variant, the system 1 is configured to acquire an image of the lens and process it according to one or more of the methods described above, in order to determine (possibly to propose it to a doctor) a cataract index, for example a LOCS III index. Subsequently, the system 1 uses the determined cataract value to assist a medical operator in the analysis of the ocular fundus.
[0213] Preferably, if the grade of cataract is equal to or higher than 2 on the LOCS III scale, the system is configured to alert the operator, through a visual image, a sound or a message, not to take BAF images of the ocular fundus because these might be not reliable due to the low SNR that could lead to artifacts that could mislead the operator in the diagnosis. For example, in the image 2113 of figure 21, the physiological foveal hypoautofluorescent area (in the centre of the retina), visible instead in figures 2110. 2111 and 2112, is not present. The absence of this area, which, in this case, depends on the grade 3 cataract, could instead be interpreted in a pathological sense by an operator who should judge the ocular fundus only on the basis of the image 2113. In order to alert the user, the system 1 uses, in a manner known per se, the computer 300 and its elements, such as loudspeakers and screen.
[0214] In addition to or as an alternative to the above alert, the system 1 may be configured to generate a second alert to the operator via the computer 300. In particular, in one embodiment, the system 1 is configured to suggest to the operator the frequency of acquisition of the OCT image of the ocular fundus. The suggestion on the OCT frequency to be used, which may be provided via a text or audio message through the computer 300, depends on the grade of cataract previously determined. For example, the system may suggest using a frequency of 125kHz if the grade of cataract (on the LOCS III scale) is between 0 and 2, whereas it may suggest using a frequency of 20kHz if the grade of cataract is equal to 3 or higher on the LOCS III scale. Alternatively, the system 1 may suggest acquiring OCT images with a different frequency for each grade of cataract, e.g. a frequency of 125 kHz for grade 0 LOCS III, a frequency of 85 kHz for grade 1 LOCS III, a frequency of 65 kHz for grade 2 LOCS III, a frequency of 45 kHz for grade 3 LOCS III, a frequency of 20 kHz for grade 4 LOCS III. The range of potentially usable frequencies varies from 250kHz to 10kHz. Greater frequencies provide greater acquisition speed but lower penetration, and can be recommended for transparent lenses or mild-moderate cataracts. Lower frequencies increase acquisition times, but guarantee greater penetration, and can be recommended for medium-advanced cataracts.
[0215] Preferably, then, in one embodiment, the system 1 is able to suggest to the operator a correction factor to be used to improve the display of an image of the ocular fundus. In particular, Table V below reports the preferred corrective factors, faand fb, which can be used to post-process digital images from the ocular fundus:
[0216] Table V
[0217] The factors faand fb reported in Table V for the different grades of cataract (LOCS III grades from 0 to 4) correspond, except for rounding, to the ratio between the average SNR value (reported in Table IV) of the images of the fundus for a grade 0 value and the average value of the corresponding LOCS III grade actually measured with the method described above. For example, in the case of a BAF image of the ocular fundus, if the eye has a grade 2 cataract, then the factor fais calculated as 10.11 / 4.22 =2.4.
[0218] In step of post-processing an image of the ocular fundus, this can be modified, in a manner known per se, in order to increase the signal-to-noise ratio by a factor fa(if BAF image) or fb (if OCT image).
[0219] In the examples mentioned above, the system 1, after determining the grade of cataract, provides suggestions and / or alerts to an operator who, through another system or device, is about to acquire the image of an ocular fundus. In an alternative embodiment, however, the system 1 is also provided with an ocular fundus image acquisition module. For example, for BAF acquisition of the ocular fundus the system 1 can be provided with another optic to focus on the fundus of the eye, instead of on the lens, a light with frequency between 400nm and 490nm, and preferably between 450nm and 470nm and more preferably equal to 460nm. For OCT acquisition of the ocular fundus the system may be provided with one or more additional light sources for focusing light beams with frequency between 10kHz and 250kHz on the ocular fundus.
[0220] In the event that the system 1 is also capable of acquiring BAF and / or OCT images of the ocular fundus, the system 1, in addition to alerting or making a suggestion to the operator, may use the data of Tables IV and V in order to control the ocular fundus image acquisition module. For example, as a function of the grade of cataract determined, the source to be used for OCT images is chosen and / or controlled, e.g. 125 kHz or 20kHz or other frequency chosen for the particular grade of cataract, as illustrated above. In case of a grade 3 or 4 LOCS III cataract, the system may block BAF image acquisition of the ocular fundus to prevent the ophthalmologist from being misled by poor quality images. Finally, the images of the ocular fundus (BAF or OCT) can be corrected to increase the signal-to-noise ratio of the factor faor fb reported in Table V.
[0221] As a further test of the cataract classification system, 4 eyes (4 patients; mean age 80+4 years) with white cataract and 2 eyes (2 patients, age 41 and 38 years, respectively) with posterior polar cataract were tested.
[0222] With regard to white cataracts, the image acquired with the system of the invention was considerably darker than the other types of cataracts analysed. The algorithm provided the following values, results significantly different from those of the cataracts discussed above:
[0223] - Mf(0.39±0.03; p<0.001),
[0224] - Eb(0.35±0.04; p<0.001),
[0225] - Ed (0.33±0.08; p<0.001).
[0226] In this case, p<0.001 can be referred to the comparison between the value of the white cataract metrics with respect to the values of all the LOCS III grades previously described.
[0227] For the eyes with white cataract it was not possible to obtain OCT and BAF images of the fundus due to the extremely high medial opacity.
[0228] Regarding the posterior polar cataract, the images acquired with the system of the invention, and shown in figure 22, showed characteristics of the lens within normal limits, with the sole exception of a central black spot in the image B of figure 22, compatible with the specific type of cataract.
[0229] VARIANTS
[0230] In light of what is set forth above, it is clear how the described method allows to provide a very efficient aid to the diagnosis of the ophthalmologist in the classification of the type and grade of cataract by providing appropriate indices, indicated above with CT and GC, dependent on the absorption of the lens of light radiation in the frequency range between 400nm and 490nm, and preferably between 450nm and 470nm and more preferably equal to 460nm.
[0231] It is also apparent that the invention is not limited to the specific embodiments described above. For example, the cataract grading algorithm can be applied on images captured with devices and / or systems other than those described above with reference to figure 1, as long as the images are acquired by illuminating the lens with a blue light, in the frequency range between 400nm and 490nm, and preferably between 450nm and 470nm and more preferably equal to 460nm.
[0232] For example, the source module 201 and the detector module 202 of figure 1 may be arranged in a single device, or be two separate devices.
[0233] Furthermore, for the acquisition of the images of the lens it is possible to use a device usually used for auto-fluorescence diagnostics of the ocular fundus, but by making appropriate modifications that make it suitable for acquiring images of the lens. These devices, in fact, are configured to illuminate the ocular fundus with a cobalt blue light and shield the return fluorescence by means of a yellow filter. By modifying the support for the patient's face to move it away from the lens, it is possible to make these devices focus on the crystalline lens instead of the ocular fundus. In addition, by adding a suitable lens between the device lens and the patient's eye, it is possible to achieve magnification which is necessary for the resulting image to have acceptable sizes.
[0234] In general, the modules represented in the block diagrams and in the flowcharts may be differently merged or separated while implementing functions similar to those described above.
[0235] Some of the steps of the above-described methods may then be omitted, for example, in generating the image E of the equivalent lens, the normalisation operations described above with reference to steps 5001 and 5010 may be omitted.
[0236] Similarly, the operation 5006 for small object removal may be omitted.
[0237] The number of grades of cataract can be modified and the parameters of the decision thresholds of the grade of cataract (parameters described above) updated with a training algorithm.
[0238] Further, as much as it is preferable to implement both the method for classifying the grade of cataract (for example as described above with reference to figures 2 and 5) and the method for classifying the type of cataract (for example as described above with reference to figures 19 and 20), it is also possible to implement only one of the two classification methods, for example only the method for classifying the grade of cataract or only the method for classifying the type of cataract.
Claims
CLAIMS1. A system for classifying a cataract, comprising a head positioning module (100) suitable for receiving a patient's head and positioning it,- a source module (201) capable of emitting electromagnetic radiation to focus it on a lens of the patient,- a detector module (202) suitable for acquiring the image of the lens illuminated by electromagnetic radiation, a control unit (204) configured to control the source module (201) and the detector module (202), and to process the image of the lens to determine the grade of cataract based on the pixel values of the image of the lens, characterized in that the source module is suitable for generating electromagnetic radiation with a frequency between 400nm and 490nm, and preferably between 450nm and 470 nm and more preferably equal to 460nm.
2. System according to claim 1, wherein the control unit (204) is configured to control the source module and the detector module in such a way as to scan the lens, focusing in different zones of the lens the electromagnetic radiation and acquiring, for each zone illuminated by the electromagnetic radiation, an intensity value of the scattered radiation.
3. System according to claim 1 or 2, wherein the electronic control unit is configured to identify (2002) a region of interest within the acquired image, said region of interest corresponding to a region of the acquired image occupied by the lens, generate (2003) an image of an equivalent lens by resizing the region of interest to a predefined size, classify (2004) a grade of cataract, based on the pixel values within the resized region of interest.
4. System according to claim 3, wherein in order to classify the cataract the control unit is configured to identify the pixels of the resized region of interest representing a healthy lens, calculate a morphological index Mf as a ratio between the number of pixels representing a healthy lens and the area of the resized region of interest, wherein the area of the resized region consists of the number of pixels present in the resized region of interest. verify (6003) whether the morphological index exceeds a first morphological threshold valueand if so classify the lens as healthy.
5. System according to claim 4, wherein if the morphological index does not exceed the first morphological threshold value, the control unit is configured to perform a second verification wherein it verifies (6003)(i) whether the morphological index Mf exceeds a threshold valuetj7associated with a grade j of cataract different from non-zero and(ii) whether the average of the pixel values representing a healthy eye exceeds a threshold value associated with said grade of cataract, and(iii) whether the average of the pixel values representing a diseased eye exceeds a threshold value associated with said grade of cataract,and wherein the control unit classifies as a grade j cataract the image of the lens in which all conditions from (i) to (iii) are satisfied.
6. System according to claim 5, wherein the control unit is configured to use the following threshold values: j0 1 2 3 40.97 0.96 0.92 0.82 0.72 t? - 0.7 0.65 0.6 0.5 b - 0.5 0.45 0.4 0.
37. System according to claim 5 or 6, wherein the control unit is configured to generate a maximum absorption index, associated with a maximum grade of cataract, which does not satisfy any of the conditions from (i) to (iii).
8. System according to any one of claims 3 to 7, wherein in order to identify the region of interest, the control unit is configured to- divide (5002) the pixels of the acquired image into two classes of pixels,- generate a binarized image (5003) by assigning to the pixels of one class a first value, in particular a value corresponding to the colour black, and by assigning to the pixels of the other class a second value, in particular a value corresponding to the colour white,- clean up the binarized image by removing cavities (5004) and filling holes (5005) by applying closing and hole-filling operators to the binarized image,- identify the maximum area object within the cleaned binarized image,- identify the centre C and radius r of the circle that best approximates the contours of the maximum area object, identify the region of interest in the set of pixels of the acquired image that lie withina circle of centre C and radius r.
9. System according to any one of claims 3 to 7, wherein in order to identify the region of interest, the control unit is configured to generate a normalised image (5001) whose pixels have normalised values PN calculated as follows:wherein Pi is the value of the i-th pixel of the acquired image, PNJ is the value of the i-th pixel after normalisation, Pmax is the maximum value among the pixel values of the acquired image, Pmin is the minimum value among the pixel values of the acquired image, R is the number of discrete values that a pixel of the normalised image can assume,- divide (5002) the pixels of the normalised image into two classes of pixels,- generate a binarized image (5003) by assigning to the pixels of one class a first value, in particular a value corresponding to the colour black, and by assigning to the pixels of the other class a second value, in particular a value corresponding to the colour white,- clean up the binarized image by removing cavities (5004) and filling holes (5005) by applying closing and hole-filling operators to the binarized image,- identify (5007) the maximum area object within the cleaned binarized image,- identify (5008) the centre C and the radius r of the circle that best approximates the contours of the maximum area object,- identify (5009) the region of interest in the set of pixels of the acquired image that are within a circle of centre C and radius r.
10. System according to any one of claims 3 to 9, wherein, for generating the image of an equivalent lens, the control unit is configured to:- generate (5010) a normalised image of the region of interest (5010) whose pixels have normalised values PL calculated as follows:wherein Pc,i is the value of the i-th pixel of the region of interest, PL,i is the value of the i-th pixel of the region of interest after normalisation, Pcmax is the maximum value among the pixel values of the region of interest, Pcmin is the minimum value among the pixel values of the region of interest, R is the number of discrete values that a pixel of the normalised image can assume,- resize (5011) the normalised region of interest to a predefined size.
11. System according to any one of the preceding claims, wherein the control unit is further configured toidentify a region of interest, said region of interest corresponding to a region of the image occupied by the lens, generate the image of an equivalent lens, said step providing for resizing the region of interest to a predefined size, classify a type of cataract based on the pixel values within the resized region of interest, wherein the control unit is configured to classify the type of cataract by performing an algorithm comprising the steps of dividing the region of interest into concentric sectors (7003), calculating an energy of each sector (7004), wherein the energy of a sector corresponds to the square of the pixels' value in each sector, identifying the sector with the highest energy (7008, 7009), classifying the type of cataract based on the identified sector (7010).
12. Computerized method for classifying a cataract, the method comprising the steps of:- processing (2002,2003,2004) an image of a lens to determine a grade of cataract based on the pixel values of the image, characterized in that the image of the lens is acquired by focusing on the lens a light radiation with a frequency between 400nm and 490nm, and preferably between 450nm and 470nm and more preferably equal to 460nm.
13. Method according to claim 12, wherein the step of processing the image of the lens comprises a step (2002) of identifying a region of interest, said region of interest corresponding to a region of the image occupied by the lens, a step (2003) of generating the image of an equivalent lens, said step providing for resizing the region of interest to a predefined size, a step (2004) of cataract classification based on the pixel values within the resized region of interest, wherein the cataract classification step provides for identifying the pixels of the resized region of interest that represent a healthy eye, calculate a morphological index Mf as a ratio between the number of pixels representing a healthy lens and the area of the resized region of interest, wherein the area of the resized region consists of the number of pixels present in the resized region of interest.verifying (6003) whether the morphological index exceeds a first morphological threshold value to , and if so, classifying the lens of the image as healthy if the verification has given a positive result.
14. Method according to claim 13, wherein if the morphological index Mf does not exceed the first morphological threshold value, a second verification is performed wherein it is verified (6003) whether the morphological index My exceeds a threshold valuetj7associated with a grade j of cataract different from non-zero and whether the average of the pixel values representing a healthy eye exceeds a threshold valueassociated with said grade of cataract, and whether the average of the pixel values representing a diseased eye exceeds a threshold valueassociated with said grade of cataract, and wherein the image of the lens in which all three conditions are satisfied is classified as a grade j cataract.
15. Method according to any one of claims 12 to 14, further comprising a step of identifying a region of interest, said region of interest corresponding to a region of the image occupied by the lens, a step of generating the image of an equivalent lens, said step providing for resizing the region of interest to a predefined size, a step of classifying the type of cataract based on the pixel values within the resized region of interest, wherein said step of classifying the type of cataract comprises the steps of dividing the region of interest into concentric sectors (7003), calculating an energy of each sector (7004), wherein the energy of a sector corresponds to the square of the pixels' value in each sector, identifying the sector with the highest energy (7008, 7009), classifying the type of cataract based on the identified sector (7010).
Citation Information
Patent Citations
Ophthalmic imaging apparatus and ophthalmic information processing apparatus
US10105052B2
Method of and system for automated machine-assisted detection of ocular disease conditions in human eyes captured using visible illumination light sources and digital camera systems
US20220313077A1
Methods and devices for the measurement of the degradation of image quality on the retina of the human eye due to cataract
US5579063A
