WO2024037676A1 - System implemented in a neural network for detecting blood vessels by means of pixel segmentation - Google Patents
System implemented in a neural network for detecting blood vessels by means of pixel segmentation Download PDFInfo
- Publication number
- WO2024037676A1 WO2024037676A1 PCT/CO2023/000016 CO2023000016W WO2024037676A1 WO 2024037676 A1 WO2024037676 A1 WO 2024037676A1 CO 2023000016 W CO2023000016 W CO 2023000016W WO 2024037676 A1 WO2024037676 A1 WO 2024037676A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- artificial intelligence
- blood vessels
- process according
- pixels
- intelligence process
- Prior art date
Links
- 210000004204 blood vessel Anatomy 0.000 title claims abstract description 23
- 238000013528 artificial neural network Methods 0.000 title claims description 5
- 230000011218 segmentation Effects 0.000 title description 2
- 238000000034 method Methods 0.000 claims abstract description 26
- 238000013473 artificial intelligence Methods 0.000 claims abstract description 14
- 230000008569 process Effects 0.000 claims description 11
- 239000011159 matrix material Substances 0.000 claims description 8
- 238000011161 development Methods 0.000 claims description 7
- 230000004913 activation Effects 0.000 claims description 5
- 238000005259 measurement Methods 0.000 claims description 3
- 238000012417 linear regression Methods 0.000 claims description 2
- 238000001514 detection method Methods 0.000 description 7
- 208000017442 Retinal disease Diseases 0.000 description 6
- 206010038923 Retinopathy Diseases 0.000 description 5
- 230000018109 developmental process Effects 0.000 description 5
- 238000004458 analytical method Methods 0.000 description 4
- 230000006870 function Effects 0.000 description 3
- 238000011176 pooling Methods 0.000 description 3
- 206010012688 Diabetic retinal oedema Diseases 0.000 description 2
- 201000011190 diabetic macular edema Diseases 0.000 description 2
- 230000036541 health Effects 0.000 description 2
- 238000003709 image segmentation Methods 0.000 description 2
- 238000010297 mechanical methods and process Methods 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 210000001525 retina Anatomy 0.000 description 2
- 238000012549 training Methods 0.000 description 2
- 206010025421 Macule Diseases 0.000 description 1
- 230000000295 complement effect Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 239000012634 fragment Substances 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 230000002028 premature Effects 0.000 description 1
- 238000004393 prognosis Methods 0.000 description 1
- 230000035755 proliferation Effects 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
Classifications
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B3/00—Apparatus for testing the eyes; Instruments for examining the eyes
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B3/00—Apparatus for testing the eyes; Instruments for examining the eyes
- A61B3/10—Objective types, i.e. instruments for examining the eyes independent of the patients' perceptions or reactions
- A61B3/13—Ophthalmic microscopes
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61F—FILTERS IMPLANTABLE INTO BLOOD VESSELS; PROSTHESES; DEVICES PROVIDING PATENCY TO, OR PREVENTING COLLAPSING OF, TUBULAR STRUCTURES OF THE BODY, e.g. STENTS; ORTHOPAEDIC, NURSING OR CONTRACEPTIVE DEVICES; FOMENTATION; TREATMENT OR PROTECTION OF EYES OR EARS; BANDAGES, DRESSINGS OR ABSORBENT PADS; FIRST-AID KITS
- A61F2/00—Filters implantable into blood vessels; Prostheses, i.e. artificial substitutes or replacements for parts of the body; Appliances for connecting them with the body; Devices providing patency to, or preventing collapsing of, tubular structures of the body, e.g. stents
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61F—FILTERS IMPLANTABLE INTO BLOOD VESSELS; PROSTHESES; DEVICES PROVIDING PATENCY TO, OR PREVENTING COLLAPSING OF, TUBULAR STRUCTURES OF THE BODY, e.g. STENTS; ORTHOPAEDIC, NURSING OR CONTRACEPTIVE DEVICES; FOMENTATION; TREATMENT OR PROTECTION OF EYES OR EARS; BANDAGES, DRESSINGS OR ABSORBENT PADS; FIRST-AID KITS
- A61F9/00—Methods or devices for treatment of the eyes; Devices for putting-in contact lenses; Devices to correct squinting; Apparatus to guide the blind; Protective devices for the eyes, carried on the body or in the hand
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
-
- G—PHYSICS
- G09—EDUCATION; CRYPTOGRAPHY; DISPLAY; ADVERTISING; SEALS
- G09B—EDUCATIONAL OR DEMONSTRATION APPLIANCES; APPLIANCES FOR TEACHING, OR COMMUNICATING WITH, THE BLIND, DEAF OR MUTE; MODELS; PLANETARIA; GLOBES; MAPS; DIAGRAMS
- G09B21/00—Teaching, or communicating with, the blind, deaf or mute
Definitions
- the present invention belongs to the field of health, particularly that of optics in relation to the measurement and detection of ocular blood vessels.
- the specific process proposed is implemented through automated methods by machine learning.
- Patent CN111144413 “Iris positioning method and system for its computer-readable storage” was found.
- the invention discloses an iris positioning method and a computer readable storage medium and the method comprises the steps: obtaining an infrared image comprising a human face or human eyes; human eye detection in the infrared image to obtain a human eye detection area; perform clustering of pixels in the detection area of the human eye according to a preset clustering number to obtain a clustering graph; performing a Hough transform on the clustering graph according to a first preset radius range and a second preset radius range to obtain a first circle and a second circle; determining an annular area according to the first circle and the second circle; combining the pooling areas with the first N pixel points in the annular area to obtain a combined pooling area, and N is a preset number; and obtain an iris area according to the human eye detection area and the pooling fusion area.
- the efficiency of the Hough transform on the basis of ensuring the accuracy of the Hough transform,
- Patent US2021259546 “Portable system to identify potential cases of diabetic macular edema using image processing and intelligence” was also identified. artificial”.
- the invention describes a portable system for the detection of diabetic macular edema by capturing a fundus image using a portable ophthalmoscope; Said image is sent by wired or wireless means to an embedded system that has an algorithm based on artificial intelligence, which extracts information from the image and processes it to identify the presence of the condition under study.
- patent CN105118049 “image segmentation method based on superpixel clustering” was found.
- the invention discloses an image segmentation method based on superpixel clustering.
- None of the solutions found in the state of the art provide the technical advantages or specific characteristics presented by the solution that is intended to be protected. It does not consist exclusively of the use of technology for grouping pixels, but rather of complementing this method with other discriminatory and classifying algorithms for the prognosis of retinal diseases, specifically retinopathy.
- the method is not only capable of identifying the presence of blood vessels in the fundus of the eye, but also of dividing the image matrix that represents the ocular area analyzed, to determine the presence of pixels equivalent to blood vessels in it and of This form is capable of determining the stage of development of the blood vessel and in turn predicting the existence of retinopathy. It should be noted that the present method has been validated in neonates, subjects in whom the identification of blood vessels using mechanical methods is particularly difficult due to the incipient development of blood vessels at this age.
- the intended solution solves the problem posed by providing a novel method that includes the use of artificial intelligence (AI) previously trained through pixel clustering that allows the pixel tone to be identified.
- AI artificial intelligence
- the measurement of blood vessel proliferation to determine retinopathy in newborns includes a mechanical method and its interpretation by medical specialists. This makes testing difficult in remote locations with little access to health infrastructure. Likewise, there are specialized devices that are difficult to access due to their cost.
- the identification of retinopathy in newborns is determined by the presence of blood vessels in different areas of the retina.
- the retina is divided into three zones, where zone 1 is the innermost and is located in the macula, then, from the inside out, come zones 2 and 3.
- the method provided offers a solution to the problem posed that includes the use of artificial intelligence for image processing, which allows identifying the presence of blood vessels and determining their extension.
- the mechanized procedure by which direct observation must be carried out to identify the presence and development of these ocular structures is dispensed with; Thus, it is possible to make forecasts remotely.
- the method implemented by artificial intelligence requires the capture of an image that, in a preferred embodiment of the invention, is taken with a camera with an angle of 160 ° in adults and 120° in premature children and a definition of 1080p.
- the intelligence comprises a consistent input. in the normalized image data; eight hidden layers that process the data (black box) and an output that delivers the presence of blood vessels and their extension.
- the proposed method allows prediction based on the analysis of a digitally captured image, which is transformed into three matrices of pixel values for machine understanding, which are normalized using a standard normalization filter to homogenize their size to 576 rows and columns in such a way that they match the size of the matrices used during training.
- the first matrix is the original image as captured by the camera; a second matrix where the image is segmented by pixels of interest; and a third matrix where the pixels are grouped in a region of interest based on the application of a mask.
- the intelligence fragments the pixels into “interest” and “non-interest”, where each pixel consists of a point of one tone on the matrix in such a way that it only graphs the blood vessels of the eye, generating a new binarized white and black image. black, to reference the length of the blood vessels in the backs of the eyes, which determines their stage of development.
- the intelligence is a multilayer neural network that comprises an input node to identify the tone palette of the referenced pixel, which in the embodiment provided by the invention identifies as positive - presence of blood vessels - the white pixels of the referenced image.
- the other four hidden layers are “Dropout” to avoid overfitting in the analysis due to the number of pixels per image and variation in tones in them, so that the output layer optimizes its precision and are also complemented with a function of Relu activation.
- the algorithm determines the stage of development of the blood vessel according to its extension by the cluster of grouped positive pixels present in the eye area. The level of extension of the blood vessel stages in the areas of the eye will determine if the subject has retinopathy.
Landscapes
- Health & Medical Sciences (AREA)
- Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Animal Behavior & Ethology (AREA)
- Theoretical Computer Science (AREA)
- Heart & Thoracic Surgery (AREA)
- Veterinary Medicine (AREA)
- Public Health (AREA)
- Ophthalmology & Optometry (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- General Physics & Mathematics (AREA)
- Surgery (AREA)
- Medical Informatics (AREA)
- Vascular Medicine (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Artificial Intelligence (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Transplantation (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Cardiology (AREA)
- Multimedia (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Business, Economics & Management (AREA)
- Educational Administration (AREA)
- Educational Technology (AREA)
- Eye Examination Apparatus (AREA)
Abstract
The claimed solution solves the problem addressed by providing a new method involving the use of an artificial intelligence (AI) previously trained using pixel clustering, allowing pixel tone identification. In this way, the system can segment blood vessels and differentiate them from the rest of the fundus of the eye.
Description
SISTEMA IMPLEMENTADO EN UNA RED NEURONAL PARA LA DETECCIÓN DESYSTEM IMPLEMENTED IN A NEURAL NETWORK FOR THE DETECTION OF
VASOS SANGUINEOS A TRAVES DE SEGMENTACION DE PIXELES BLOOD VESSELS THROUGH PIXEL SEGMENTATION
Campo de la invención field of invention
La presente invención pertenece al campo de la salud, particularmente el de la óptica en lo relacionado con la medición y detección de vasos sanguíneos oculares. El proceso específico que se propone está implementado a través de métodos automatizados por aprendizaje de máquina. The present invention belongs to the field of health, particularly that of optics in relation to the measurement and detection of ocular blood vessels. The specific process proposed is implemented through automated methods by machine learning.
Estado de la técnica State of the art
Se encontró la patente CN111144413 “Método de posicionamiento del iris y sistema par su almacenamiento legible por computador”. La invención divulga un método de posicionamiento del iris y un medio de almacenamiento legible por computadora y el método comprende los pasos: obtención de una imagen infrarroja que comprende un rostro humano u ojos humanos; detección del ojo humano en la imagen infrarroja para obtener un área de detección del ojo humano; realizar un agrupamiento de píxeles en el área de detección del ojo humano de acuerdo con un número de agrupamiento preestablecido para obtener un gráfico de agrupamiento; realizar una transformada de Hough en el gráfico de agrupamiento de acuerdo con un primer rango de radio preestablecido y un segundo rango de radio preestablecido para obtener un primer círculo y un segundo círculo; determinando un área anular según el primer círculo y el segundo circulo; combinando las áreas de agrupación con los primeros N puntos de píxeles en el área anular para obtener un área combinada de agrupamiento, y N es un número preestablecido; y obtener un área de iris de acuerdo con el área de detección del ojo humano y el área de fusión de agrupamiento. Según la invención, sobre la base de garantizar la precisión de la transformada de Hough, la eficiencia de la transformada de Hough es mejorada, y se mejora la robustez de un algoritmo de posicionamiento del iris. Patent CN111144413 “Iris positioning method and system for its computer-readable storage” was found. The invention discloses an iris positioning method and a computer readable storage medium and the method comprises the steps: obtaining an infrared image comprising a human face or human eyes; human eye detection in the infrared image to obtain a human eye detection area; perform clustering of pixels in the detection area of the human eye according to a preset clustering number to obtain a clustering graph; performing a Hough transform on the clustering graph according to a first preset radius range and a second preset radius range to obtain a first circle and a second circle; determining an annular area according to the first circle and the second circle; combining the pooling areas with the first N pixel points in the annular area to obtain a combined pooling area, and N is a preset number; and obtain an iris area according to the human eye detection area and the pooling fusion area. According to the invention, on the basis of ensuring the accuracy of the Hough transform, the efficiency of the Hough transform is improved, and the robustness of an iris positioning algorithm is improved.
También se identificó la patente US2021259546 “Sistema portátil para identificar casos potenciales de edema macular diabético usando procesamiento de imágenes e inteligencia
artificial”. La invención describe un sistema portátil para la detección de edema macular diabético mediante la captura de una imagen de fondo de ojo utilizando un oftalmoscopio portátil; dicha imagen es enviada por medios alámbricos o inalámbricos a un sistema embebido que cuenta con un algoritmo basado en inteligencia artificial, el cual extrae información de la imagen y la procesa para identificar la presencia de la condición en estudio. Patent US2021259546 “Portable system to identify potential cases of diabetic macular edema using image processing and intelligence” was also identified. artificial". The invention describes a portable system for the detection of diabetic macular edema by capturing a fundus image using a portable ophthalmoscope; Said image is sent by wired or wireless means to an embedded system that has an algorithm based on artificial intelligence, which extracts information from the image and processes it to identify the presence of the condition under study.
Finalmente se encontró la patente CN105118049 “método de segmentación de imágenes basado en agrupamiento de superpíxeles”. La invención divulga un método de segmentación de imágenes basado en agrupamiento de superpíxeles. Finally, patent CN105118049 “image segmentation method based on superpixel clustering” was found. The invention discloses an image segmentation method based on superpixel clustering.
Ninguna de las soluciones encontradas en el estado de la técnica provee las ventajas técnicas o características específicas que presenta la solución que se pretende proteger. La misma no consiste exclusivamente en el uso de tecnología para el agrupamiento de pixeles, sino en la complementación de este método con otros algoritmos discriminatorios y clasifícatenos para el pronóstico de enfermedades retínales, específicamente de la retinopatía. El método, no sólo es capaz de identificar la presencia de vasos sanguíneos en el fondo del ojo, sino de dividir la matriz de la imagen que representa la zona ocular analizada, para determinar la presencia de pixeles equivalentes a vasos sanguíneos en la misma y de esta forma es capaz de determinar el estadio de desarrollo del vaso sanguíneo y a su vez pronosticar la existencia de retinopatía. Cabe anotar, que el presente método ha sido validado en neonatos, sujetos en los cuales la identificación de vasos sanguíneos utilizando métodos mecánicos es particularmente difícil debido al desarrollo incipiente de los vasos sanguíneos a esta edad. None of the solutions found in the state of the art provide the technical advantages or specific characteristics presented by the solution that is intended to be protected. It does not consist exclusively of the use of technology for grouping pixels, but rather of complementing this method with other discriminatory and classifying algorithms for the prognosis of retinal diseases, specifically retinopathy. The method is not only capable of identifying the presence of blood vessels in the fundus of the eye, but also of dividing the image matrix that represents the ocular area analyzed, to determine the presence of pixels equivalent to blood vessels in it and of This form is capable of determining the stage of development of the blood vessel and in turn predicting the existence of retinopathy. It should be noted that the present method has been validated in neonates, subjects in whom the identification of blood vessels using mechanical methods is particularly difficult due to the incipient development of blood vessels at this age.
Breve descripción de la invención Brief description of the invention
La solución pretendida, resuelve el problema planteado al aportar un método novedoso que comprende el uso de una inteligencia artificial (IA) previamente entrenada a través de un clustering de pixeles que permite identificar el tono de los mismos. Así, es capaz de segmentar y diferenciar vasos sanguíneos del resto del fondo del ojo.
Breve descripción de las figuras The intended solution solves the problem posed by providing a novel method that includes the use of artificial intelligence (AI) previously trained through pixel clustering that allows the pixel tone to be identified. Thus, it is capable of segmenting and differentiating blood vessels from the rest of the fundus of the eye. Brief description of the figures
Figura 1. Muestra un flujograma del proceso Figure 1. Shows a flowchart of the process
Descripción detallada de la invención Detailed description of the invention
La medición de la proliferación de los vasos sanguíneos para la determinación de retinopatía en recién nacidos, comprende un método mecánico y su interpretación por especialistas médicos. Ello dificulta las pruebas en ubicaciones recónditas y con escaso acceso a infraestructuras de salud. Asimismo, existen dispositivos especializados de difícil acceso por su costo. The measurement of blood vessel proliferation to determine retinopathy in newborns includes a mechanical method and its interpretation by medical specialists. This makes testing difficult in remote locations with little access to health infrastructure. Likewise, there are specialized devices that are difficult to access due to their cost.
La identificación de la retinopatía en recién nacidos, se determina a partir de la presencia de vasos sanguíneos en las diferentes zonas de la retina. Para esos efectos, la retina está dividida en tres zonas, donde la zona 1 es la más interior y se encuentra en la mácula, luego, de adentro hacia afuera, vendrían las zonas 2 y 3. The identification of retinopathy in newborns is determined by the presence of blood vessels in different areas of the retina. For these purposes, the retina is divided into three zones, where zone 1 is the innermost and is located in the macula, then, from the inside out, come zones 2 and 3.
Por ello, el método provisto ofrece una solución al problema planteado que comprende el uso de la inteligencia artificial para el procesamiento de imágenes, que permite identificar la presencia de vasos sanguíneos y determina la extensión de los mismos. Con ello, se prescinde del procedimiento mecanizado por el cual se debe realizar observación directa para identificar la presencia y desarrollo de estas estructuras oculares; así, es posible la realización de pronósticos de manera remota. Therefore, the method provided offers a solution to the problem posed that includes the use of artificial intelligence for image processing, which allows identifying the presence of blood vessels and determining their extension. With this, the mechanized procedure by which direct observation must be carried out to identify the presence and development of these ocular structures is dispensed with; Thus, it is possible to make forecasts remotely.
El método implementado por inteligencia artificial requiere la captura de una imagen que, en una modalidad preferida de la invención se toma con una cámara con angular de 160 0 en adultos y 120° en niños prematuros y una definición de 1080p la inteligencia comprende una entrada consistente en los datos normalizados de la imagen; ocho capas ocultas que procesan los datos (caja negra) y una salida que entrega la presencia de vasos sanguíneos y su extensión. The method implemented by artificial intelligence requires the capture of an image that, in a preferred embodiment of the invention, is taken with a camera with an angle of 160 ° in adults and 120° in premature children and a definition of 1080p. The intelligence comprises a consistent input. in the normalized image data; eight hidden layers that process the data (black box) and an output that delivers the presence of blood vessels and their extension.
Para el entrenamiento de la inteligencia artificial se utilizan grupos de imágenes donde se usan matrices de las mismas dimensiones y se coloca una máscara de determinada tonalidad sobre los píxeles de interés para focalizar el entrenamiento. Los píxeles de interés
comprenden información acerca de su tono específico que le diferencian del resto. Con esa información la inteligencia artificial aprende a distinguir los píxeles que le corresponde evaluar cada vez. To train artificial intelligence, groups of images are used where matrices of the same dimensions are used and a mask of a certain tone is placed over the pixels of interest to focus the training. Pixels of interest They understand information about your specific tone that differentiates you from the rest. With this information, artificial intelligence learns to distinguish the pixels that it should evaluate each time.
El método propuesto permite la predicción a partir del análisis de una imagen capturada digitalmente, que se transforma en tres matrices de valores de pixel para el entendimiento de la máquina, que son normalizadas mediante un filtro de normalización estándar para homogeneizar el tamaño de las mismas a 576 filas y columnas de tal forma que coincidan con el tamaño de las matrices usadas durante el entrenamiento. The proposed method allows prediction based on the analysis of a digitally captured image, which is transformed into three matrices of pixel values for machine understanding, which are normalized using a standard normalization filter to homogenize their size to 576 rows and columns in such a way that they match the size of the matrices used during training.
En la primera matriz está la imagen original según se captura por la cámara; una segunda matriz donde se segmenta la imagen por píxeles de interés; y una tercera matriz donde se agrupan los píxeles en una región de interés a partir de la aplicación de una máscara.In the first matrix is the original image as captured by the camera; a second matrix where the image is segmented by pixels of interest; and a third matrix where the pixels are grouped in a region of interest based on the application of a mask.
Finalmente, la inteligencia fragmenta los píxeles en “interés” y “no interés”, donde cada pixel consiste en un punto de un tono en la matriz de tal forma que sólo gráfica los vasos sanguíneos del ojo, generando una nueva imagen binarizada a blanco y negro, para referenciar la longitud de los vasos sanguíneos en los fondos de los ojos, lo que determina su estadio de desarrollo. Finally, the intelligence fragments the pixels into “interest” and “non-interest”, where each pixel consists of a point of one tone on the matrix in such a way that it only graphs the blood vessels of the eye, generating a new binarized white and black image. black, to reference the length of the blood vessels in the backs of the eyes, which determines their stage of development.
La inteligencia es una red neuronal multicapa que comprende un nodo de entrada para identificar la paleta de tonos del pixel referenciado, que en la modalidad provista de la invención identifica como positivos - presencia de vasos sanguíneos -- los píxeles blancos de la imagen referenciada. The intelligence is a multilayer neural network that comprises an input node to identify the tone palette of the referenced pixel, which in the embodiment provided by the invention identifies as positive - presence of blood vessels - the white pixels of the referenced image.
Presenta ocho capas ocultas que se complementan entre sí mejorando la eficiencia del análisis, donde las primeras cuatro capas ocultas generan los pesos necesarios para identificar el pixel como positivo o negativo, mediante un algoritmo de regresión lineal complementado con una función de activación Relu. Las otras cuatro capas ocultas son “Dropout” para evitar sobreajuste en el análisis debido a la cantidad de píxeles por imagen y variación de tonos en los mismos, de tal forma que la capa de salida optimice su precisión y también se complementan con una función de activación Relu. It presents eight hidden layers that complement each other, improving the efficiency of the analysis, where the first four hidden layers generate the weights necessary to identify the pixel as positive or negative, using a linear regression algorithm complemented with a Relu activation function. The other four hidden layers are “Dropout” to avoid overfitting in the analysis due to the number of pixels per image and variation in tones in them, so that the output layer optimizes its precision and are also complemented with a function of Relu activation.
También comprende un nodo de salida con una función de activación “Sigmoid” que entrega el resultado del análisis y determina si se trata de un vaso sanguíneo u otro elemento del fondo del ojo.
Además, el algoritmo determina el estadio de desarrollo del vaso sanguíneo según su extensión por el cluster de píxeles positivos agrupados presentes en la zona del ojo. El nivel de extensión de los estadios del vaso sanguíneo en las zonas del ojo, determinará si el sujeto tiene retinopatía.
It also includes an output node with a “Sigmoid” activation function that delivers the analysis result and determines whether it is a blood vessel or another element of the fundus. Furthermore, the algorithm determines the stage of development of the blood vessel according to its extension by the cluster of grouped positive pixels present in the eye area. The level of extension of the blood vessel stages in the areas of the eye will determine if the subject has retinopathy.
Claims
1. Proceso de inteligencia artificial aplicado a una red neuronal de identificación y medición de vasos sanguíneos oculares, caracterizada porque comprende una red neuronal multicapa, prealimentada con una capa de entrada, al menos ocho capas ocultas y una capa de salida dentro de un entorno de desarrollo integrado. 1. Artificial intelligence process applied to a neural network for identification and measurement of ocular blood vessels, characterized in that it comprises a multilayer neural network, pre-fed with an input layer, at least eight hidden layers and an output layer within an environment of integrated development.
2. Proceso de inteligencia artificial de acuerdo a la reivindicación 1 , en donde la capa de entrada corresponde a la paleta de tonos del pixel referenciado como positivo obtenido a partir de la imagen. 2. Artificial intelligence process according to claim 1, wherein the input layer corresponds to the tone palette of the pixel referenced as positive obtained from the image.
3. Proceso de inteligencia artificial de acuerdo a la reivindicación 1 , donde las capas ocultas fueron entrenadas de tal forma que las cuatro primeras capas ocultas usan un algoritmo de regresión lineal complementado con una función de activación Relu y las cuatro segundas capas ocultas son “dropout” con una función de activación Relu. 3. Artificial intelligence process according to claim 1, wherein the hidden layers were trained in such a way that the first four hidden layers use a linear regression algorithm complemented with a Relu activation function and the second four hidden layers are “dropout”. ” with a Relu activation function.
4. Proceso de inteligencia artificial de acuerdo a la reivindicación 1 , en donde la capa de entrada acondiciona la imagen segmentándola en tres matrices; donde la primera matriz comprende la imagen original, la segunda matriz comprende la imagen segmentada por píxeles de interés y la tercera matriz comprende los píxeles de interés agrupados en regiones. 4. Artificial intelligence process according to claim 1, wherein the input layer conditions the image by segmenting it into three matrices; where the first matrix comprises the original image, the second matrix comprises the image segmented by pixels of interest and the third matrix comprises the pixels of interest grouped into regions.
5. Proceso de inteligencia artificial de acuerdo a la reivindicación 1 , en donde la agrupación de los píxeles en regiones de interés se produce a partir de la aplicación de una máscara. 5. Artificial intelligence process according to claim 1, wherein the grouping of pixels into regions of interest occurs from the application of a mask.
6. Proceso de inteligencia artificial de acuerdo a la reivindicación 1 , en donde la capa de salida, por medio de una función sigmoide, entrega un pronóstico de presencia o ausencia de vasos sanguíneos en el fondo del ojo. 6. Artificial intelligence process according to claim 1, wherein the output layer, through a sigmoid function, provides a forecast of the presence or absence of blood vessels in the fundus of the eye.
7. Proceso de inteligencia artificial de acuerdo a la reivindicación 1 , en donde, si la capa
de salida identifica presencia de vasos sanguíneos en el fondo del ojo, determina su estadio de desarrollo a partir de su extensión en píxeles.
7. Artificial intelligence process according to claim 1, where, if the layer output identifies the presence of blood vessels in the fundus of the eye, determines their stage of development based on their extension in pixels.
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CONC2022/0011610A CO2022011610A1 (en) | 2022-08-17 | 2022-08-17 | Method implemented by machine learning for the detection of blood vessels through pixel clustering |
CONC2022/0011610 | 2022-08-17 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2024037676A1 true WO2024037676A1 (en) | 2024-02-22 |
Family
ID=89940836
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/CO2023/000016 WO2024037676A1 (en) | 2022-08-17 | 2023-08-16 | System implemented in a neural network for detecting blood vessels by means of pixel segmentation |
Country Status (2)
Country | Link |
---|---|
CO (1) | CO2022011610A1 (en) |
WO (1) | WO2024037676A1 (en) |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20190130566A1 (en) * | 2015-04-06 | 2019-05-02 | IDx, LLC | Systems and methods for feature detection in retinal images |
US20200178794A1 (en) * | 2017-06-20 | 2020-06-11 | University Of Louisville Research Foundation, Inc. | Segmentation of retinal blood vessels in optical coherence tomography angiography images |
US11213197B2 (en) * | 2017-05-04 | 2022-01-04 | Shenzhen Sibionics Technology Co., Ltd. | Artificial neural network and system for identifying lesion in retinal fundus image |
US20220012859A1 (en) * | 2020-07-09 | 2022-01-13 | Henan University Of Technology | Method and device for parallel processing of retinal images |
US20220130037A1 (en) * | 2019-02-12 | 2022-04-28 | National University Of Singapore | Retina vessel measurement |
US20220160228A1 (en) * | 2019-03-20 | 2022-05-26 | Carl Zeiss Meditec Ag | A patient tuned ophthalmic imaging system with single exposure multi-type imaging, improved focusing, and improved angiography image sequence display |
-
2022
- 2022-08-17 CO CONC2022/0011610A patent/CO2022011610A1/en unknown
-
2023
- 2023-08-16 WO PCT/CO2023/000016 patent/WO2024037676A1/en unknown
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20190130566A1 (en) * | 2015-04-06 | 2019-05-02 | IDx, LLC | Systems and methods for feature detection in retinal images |
US11213197B2 (en) * | 2017-05-04 | 2022-01-04 | Shenzhen Sibionics Technology Co., Ltd. | Artificial neural network and system for identifying lesion in retinal fundus image |
US20200178794A1 (en) * | 2017-06-20 | 2020-06-11 | University Of Louisville Research Foundation, Inc. | Segmentation of retinal blood vessels in optical coherence tomography angiography images |
US20220130037A1 (en) * | 2019-02-12 | 2022-04-28 | National University Of Singapore | Retina vessel measurement |
US20220160228A1 (en) * | 2019-03-20 | 2022-05-26 | Carl Zeiss Meditec Ag | A patient tuned ophthalmic imaging system with single exposure multi-type imaging, improved focusing, and improved angiography image sequence display |
US20220012859A1 (en) * | 2020-07-09 | 2022-01-13 | Henan University Of Technology | Method and device for parallel processing of retinal images |
Also Published As
Publication number | Publication date |
---|---|
CO2022011610A1 (en) | 2024-02-26 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US20210090736A1 (en) | Systems and methods for anomaly detection for a medical procedure | |
US10441160B2 (en) | Method and system for classifying optic nerve head | |
US20190191988A1 (en) | Screening method for automated detection of vision-degenerative diseases from color fundus images | |
US11741608B2 (en) | Assessment of fundus images | |
US20170209078A1 (en) | Method and apparatus for aiding in the diagnosis of otitis media by classifying tympanic membrane images | |
CN113646805A (en) | Image-based detection of ophthalmic and systemic diseases | |
Valente et al. | Automatic diagnosis of strabismus in digital videos through cover test | |
US11974808B2 (en) | Anterior eye disease diagnostic system and diagnostic method using same | |
Pathak et al. | A robust automated cataract detection algorithm using diagnostic opinion based parameter thresholding for telemedicine application | |
Jaworek-Korjakowska et al. | Eskin: study on the smartphone application for early detection of malignant melanoma | |
KR20190048340A (en) | Electronic device and method for determining hyperemia grade of eye using the same | |
US20220198831A1 (en) | System for determining one or more characteristics of a user based on an image of their eye using an ar/vr headset | |
WO2019204700A1 (en) | Neonatal pain identification from neonatal facial expressions | |
Lu et al. | Automated strabismus detection for telemedicine applications | |
Sujaini et al. | Comparing the performance of linear regression versus deep learning on detecting melanoma skin cancer using apple core ML | |
Akram et al. | Segmentation and classification of skin lesions using hybrid deep learning method in the Internet of Medical Things | |
Brügge et al. | Automated motor tic detection: A machine learning approach | |
WO2024037676A1 (en) | System implemented in a neural network for detecting blood vessels by means of pixel segmentation | |
Abdani et al. | Iris segmentation method of pterygium anterior segment photographed image | |
Gibertoni et al. | Vision-Based Eye Image Classification for Ophthalmic Measurement Systems | |
Chauhan et al. | Deep learning based fully automatic efficient burn severity estimators for better burn diagnosis | |
CN112487980B (en) | Micro-expression-based treatment method, device, system and computer-readable storage medium | |
Yıldız et al. | Segmentation and classification of skin burn images with artificial intelligence: Development of a mobile application | |
CN115375611A (en) | Model training-based refraction detection method and detection system | |
Threlfall et al. | A publicly available vessel segmentation algorithm for SLO images |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 23854571 Country of ref document: EP Kind code of ref document: A1 |