System and method for classifying breast cancer risk based on impedance-signal processing

Abstract

The present invention relates to a method and to a system configured to carry out the method, for classifying a patient's breast cancer risk. The method comprises: separating bioimpendance components on the basis of bioimpedance measurement data produced using various combinations at different points of each breast for a critical frequency range; determining the differences in each impedance component between one breast and the other for each combination of electrodes; constructing an image for the impedance components measured in each breast individually and a differential image that represents the differences in the impedance components between one breast and the other; feeding the image of each breast and / or differential image into a previously trained neural network to define a breast cancer risk classification; and displaying the breast cancer risk classification through a user interface.

Description

SYSTEM AND METHOD FOR CLASSIFYING THE RISK OF BREAST CANCER OCCURRENCE BASED ON IMPEDANCE SIGNAL PROCESSING FIELD OF TECHNIQUE

[0001] The present invention is directed to computer-implemented systems and methods that classify a patient's risk of developing breast cancer. Specifically, the methods of the present invention perform impedance signal processing and utilize neural networks for these purposes. STATE OF THE ART

[0002] Early detection of breast cancer has become a public health priority worldwide. Traditionally, breast cancer diagnosis has been made through mammograms and biopsy analysis. However, access to these techniques is sometimes limited, particularly for patients living in rural or marginalized areas, for whom access to medical care is restricted. Thus, in the current state of the art, several alternatives to these traditional techniques have been proposed for early detection of tissue abnormalities in a patient's breasts, using impedance measurement, as demonstrated below.

[0003] Document CN112754456A discloses a deep learning-based three-dimensional electrical impedance imaging system, which can be applied to the early detection of female breast cancer. The imaging system comprises a sensor module, a data acquisition control module, and an upper computing module that are connected in sequence; the sensor module is used to form a sensitive field in the space of a measured area; the data acquisition control module is used to measure the surface impedance of the measured area and carry out primary processing and data transmission; and the upper computing module is used to perform image reconstruction based on the processed impedance data. This deep learning-based three-dimensional electrical impedance imaging system can be used for imaging human mammary glands.In this application, the system can obtain three-dimensional surface impedance information of a hemisphere measurement area through a certain excitation acquisition mode and is used for the reconstruction of the conductivity distribution of the three-dimensional hemisphere area using. the difference in electrical properties between mammary gland tissue and cancer tissue.

[0004] International patent application WO2023163720A1 discloses a non-invasive, portable probe for breast cancer detection, having an elongated housing with a sensor head at one end and microelectronics housed within the housing. The microelectronics include a Colpitts oscillator module. A ferrite coil sensor is mounted within the sensor head. The sensor includes a ferrite core wrapped in a multi-turn coil. The coil generates a primary magnetic field and receives a secondary magnetic field generated by the electrical conductivity of biological tissue. The Colpitts oscillator detects a voltage change. The probe detects the suspected presence of a breast tumor that has a higher electrical conductivity than normal breast tissue.

[0005] WO2013027120A2 relates to systems and methods for measuring the composition of a patient's body part, or for otherwise determining a clinically relevant fact using characteristics of tissue in or proximal to said body part. In one embodiment, this disclosure relates to measuring the electrical impedance of an organ or portion of an organ, such as the female breast, to obtain clinically relevant information. In one aspect, the system and method can be used to measure and utilize certain information such as breast density data and other risk factors for determining or classifying a woman's likelihood of developing breast cancer. Other aspects quantify or qualify a woman's responsiveness to a drug or hormonal therapy.

[0006] International application W02009082434A1 mentions methods and systems for noninvasive measurement of subepithelial impedance of the breast and for assessing the risk that a substantially asymptomatic patient will develop or have a substantially increased risk of developing proliferative or precancerous changes in the breast, or may have a subsequent risk of developing precancerous or cancerous changes. A plurality of electrodes are used to measure the subepithelial impedance of a patient's parenchymal breast tissue at one or more locations and at least one frequency, particularly moderately high frequencies. The risk of developing breast cancer is assessed based on measured and expected or estimated values ​​of subepithelial impedance for the patient and based on one or more experience-based algorithms.

[0007] Document CA2231038A1 discloses a method and apparatus for diagnosing states pathological conditions by obtaining a plurality of electrical impedance data measurements in organized patterns from two anatomically homologous body regions, one of which may be affected by a disease. A subset of the obtained data is processed, compared, and analyzed by structuring the data values ​​as elements of an nxny impedance matrix, characterizing these matrices by their eigenvalues ​​and eigenvectors. Another subset of the data is processed, compared, and analyzed by alternately plotting the impedance data as chords of two circles representing the two homologous body regions. The impedance chord plots provide a visual indication of certain pathological states and their location. SUMMARY OF THE INVENTION

[0008] The inventors of this application have identified a need in the technical field for new noninvasive, radiation-free systems and methods that provide an alternative for the early classification of a patient's risk of developing breast cancer; that can detect nonpalpable and / or asymptomatic tissue abnormalities; that can identify whether a given tissue alteration is actually related to a pathology; and that are easy to implement so that they can be used in patients residing in rural or marginal areas. They thus identified a need for alternative mechanisms that can assist personnel responsible for health promotion and prevention in the detection of any type of breast tissue alteration or abnormality in order to refer the patient to a diagnostic test for breast cancer (such as mammography) depending on the determined risk classification.

[0009] Thus, the inventors of the present application provide a computer-implemented method for classifying a patient's risk of breast cancer occurrence, which overcomes some of the identified needs, and which comprises: from bioimpedance measurements made using different combinations of electrodes in each breast for a critical range of frequencies, separating the components of said bioimpedance (impedance magnitude and phase); determining the differences in each of the impedance components between one breast and the other, for each combination of electrodes; constructing an image for the impedance components measured in each breast individually and / or a differential image representing the differences in the impedance components between one breast and the other; feeding the image of each breast and / or the differential image to a previously trained to define a breast cancer risk classification; and present the breast cancer risk classification through a user interface.

[0010] Likewise, the inventors of the present application provide a system comprising electrode pairs; bioimpedance readers; processing means; and a user interface; wherein the processing means are configured to: receive the impedance values ​​measured by different combinations of the electrode pairs; and execute the computer-implemented method described above. BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 shows a grayscale coded image for the behavior of the magnitude of the impedance, measured in the left breast of a patient with BI-RADS category 1 (no abnormality found), by means of an embodiment of the method of the present invention. FIG. 2 shows a grayscale coded image for the behavior of the magnitude of the impedance, measured in the right breast of a patient with BI-RADS category 1 (no abnormality found), by means of an embodiment of the method of the present invention. FIG. 3 shows a grayscale coded image for the behavior of the differential measurement of the magnitude of the impedance between both breasts of a patient with BI-RADS category 1 (no abnormality found), using a modality of the method of the present invention. FIG. 4 shows a grayscale image coded for phase behavior, measured in the left breast of a patient with BI-RADS category 1 (no abnormality found), by an embodiment of the method of the present invention. FIG. 5 shows a grayscale image coded for phase behavior, measured in the right breast of a patient with BI-RADS category 1 (no abnormality found), by an embodiment of the method of the present invention. FIG. 6 shows a grayscale coded image for the behavior of the differential phase measurement between both breasts of a patient with BI-RADS category 1 (no abnormality found), using a modality of the method of the present invention. FIG. 7 shows a grayscale coded image for the impedance magnitude behavior, measured in the left breast of a patient with BI-RADS category 4 (abnormality that strongly suggests a malignant finding, so appropriate action should be taken), by an embodiment of the method of the present invention. FIG. 8 shows a grayscale coded image for the impedance magnitude behavior, measured in the right breast of a patient with BI-RADS category 4 (abnormality that strongly suggests a malignant finding, so appropriate action should be taken), by an embodiment of the method of the present invention. FIG. 9 shows a grayscale coded image for the behavior of the differential measurement of the magnitude of the impedance between both breasts of a patient with BI-RADS category 4 (anomaly that strongly suggests that it is a malignant finding, so appropriate actions should be taken), using a modality of the method of the present invention. FIG. 10 shows a grayscale image coded for phase behavior, measured in the left breast of a patient with BI-RADS category 4 (abnormality strongly suggesting a malignant finding, requiring appropriate action), by an embodiment of the method of the present invention. FIG. 11 shows a grayscale image coded for phase behavior, measured in the right breast of a patient with BI-RADS category 4 (abnormality strongly suggesting a malignant finding, requiring appropriate action), using an embodiment of the method of the present invention. FIG. 12 shows a grayscale image coded for the behavior of the phase differential measurement between both breasts of a patient with BI-RADS category 4 (abnormality that strongly suggests that it is a malignant finding, so appropriate action should be taken), using an embodiment of the method of the present invention. FIG. 13 illustrates a diagram that allows validating the predictive capacity of the trained neural network with the differential phase and magnitude components of the impedance measurements between the right and left sine, according to an embodiment of the method of the present invention. FIG. 14 illustrates a diagram that allows validating the predictive capacity of the neural network trained with the differential magnitude components of the impedance measurements between right and left breast, according to an embodiment of the method of the present invention. FIG. 15 illustrates a diagram that allows validating the predictive capacity of the neural network trained with the differential phase components of the impedance measurements between right and left sine, according to an embodiment of the method of the present invention. FIG. 16 illustrates one side of the flower-shaped electrode support of a preferred embodiment of the system of the present invention, according to an embodiment of the method of the present invention. FIG. 17 illustrates the flower-shaped electrode holder of a preferred embodiment of the system of the present invention, according to an embodiment of the method of the present invention. FIG. 18 shows curves generated with independent measurements of the magnitude component of impedance, measured in the left breast of a patient with BI-RADS category 1 (no abnormality found). FIG. 19 shows curves generated with independent measurements of the magnitude component of impedance, measured in the right breast of a patient with BI-RADS category 1 (no abnormality found). FIG. 20 shows curves generated with the differential measurement of the magnitude component between both breasts of a patient with BI-RADS category 1 (no abnormality found). FIG. 21 shows curves generated with independent measurements of the phase component of impedance, measured in the left breast of a patient with BI-RADS category 4 (an abnormality that strongly suggests a malignant finding, so appropriate action should be taken). FIG. 22 shows curves generated with independent measurements of the phase component of impedance, measured in the right breast of a patient with BI-RADS category 4 (abnormality that strongly suggests a malignant finding, so appropriate actions should be taken) appropriate). FIG. 23 shows curves generated with the differential measurement of the phase component between both breasts of a patient with BI-RADS category 4 (an abnormality that strongly suggests a malignant finding, so appropriate action should be taken). DETAILED DESCRIPTION OF THE INVENTION

[0011] The invention relates to a computer-implemented method that allows for classifying a patient's risk of developing breast cancer by processing bioimpedance values ​​measured directly from the patient's left and right breasts. Bioimpedance is a non-invasive body analysis technique that allows for simple identification of the composition of breast tissue. These methods are not intended to replace mammography, nor do they constitute screening or diagnostic tests per se. Instead, they seek only to categorize a patient's risk of presenting abnormalities in their breast tissue that may be related to some type of pathology. After this categorization, the patient will undergo an examination such as a mammogram and / or a biopsy to rule out any pathology or receive an effective diagnosis.Thus, the inventors provide a method that can be easily implemented in the population and that can determine the risk category for breast cancer, particularly when the patient presents with a non-palpable and asymptomatic breast tissue alteration or abnormality that could not be easily identified by other mechanisms and would go unnoticed.

[0012] The computer-implemented method of the present invention comprises: from bioimpedance measurements taken on each breast for a critical frequency range, separating the components of said bioimpedance (impedance magnitude and phase); determining the differences in each of the impedance components between one breast and the other; constructing an image for the impedance components measured on each breast individually and / or a differential image representing the differences in the impedance components between one breast and the other; feeding the image of each breast and / or the differential image to a pre-trained neural network to define a breast cancer risk classification; and presenting the breast cancer risk classification through a user interface.

[0013] In one embodiment of the invention, bioimpedance measurements are performed by applying small alternating currents at multiple frequencies to electrodes placed on each breast, resulting in the recording of electrical potentials. This process is repeated for different electrode configurations to generate the readings. A bioimpedance reader is used that allows measurements to be taken at different points (electrodes) on both breasts in different frequency ranges. A breast tissue bioimpedance response meter can also be used at different frequencies. Specifically, these frequencies can be in the alpha region (f less than approximately 1000 Hz) or beta region (f between approximately 1000 Hz and approximately 100 MHz), preferably the frequency can be less than 120 MHz.

[0014] In one embodiment of the invention, measurements are made in two or more frequency intervals. The intervals may be subdivided into one or more subintervals (alpha and beta), preferably between 25 and 100 subintervals. In particular, the intervals may be subdivided into 50 subintervals.

[0015] In one embodiment of the invention, measurements can be made in two frequency ranges, preferably in a first range between 1,000 Hz and 10,000 Hz with 200 Hz steps (alpha) and a second range between 10,000 Hz and 100,000 Hz with 2,000 Hz steps (beta).

[0016] In one embodiment of the invention, a first frequency range between 2000Hz-12000Hz and a second frequency range between 20000Hz and 120000Hz are used.

[0017] In one embodiment of the invention, an even number of electrodes can be used to perform measurements in each breast. Specifically, between 2 and 20 electrodes, preferably between 4 and 10 electrodes, can be used. In one embodiment of the invention, 6 electrodes can be used to perform measurements in each breast.

[0018] In one embodiment of the invention, between 10 and 100 measurements can be made per frequency subrange for each electrode combination. Particularly, between 20 and 50 measurements can be made, more preferably 30 measurements can be made per frequency subrange, per electrode combination.

[0019] Bioimpedance measures the opposition offered by a biological medium (cells, fluids, tissues or organs) to the passage of an ionic current. If these measurements are made within a certain frequency range, electrical impedance spectroscopy is obtained, which allows the analysis of impedance behavior in a biological medium according to its basic electrical properties: resistance, capacitance, and inductance. It is widely known in the technical field that the bioelectrical parameters estimated by bioimpedance measurement are: electrical resistance, inductive reactance, and capacitive reactance. The literature documents that, from electrical resistance and capacitive reactance, it is possible to calculate components of each impedance measurement, such as magnitude and phase angle.

[0020] Biological tissues have the ability to store charge and therefore develop electrical impedance in response to applied alternating current. Measuring this impedance has found immense applications in imaging different physiological and pathological activities, such as in the detection of breast cancer. Cells are made of a cell membrane and an intracellular medium. Both the extracellular and intracellular medium consist of ionic solutions that are resistive in nature. The cell membrane is made of a lipid bilayer and proteins and is primarily capacitive. The impedance associated with this capacitance depends on the frequency. At lower frequencies, the cell membrane acts as an insulator, and therefore, current flows only through the extracellular space, and the resulting impedance is largely resistive. At higher frequencies, the cell membrane begins to conduct, and therefore, the impedance is lower.Therefore, it is evident that tissues exhibit frequency-dependent electrical behavior. Fat-free tissue is electrically conductive due to its high water (~73%) and electrolyte (ions and proteins) content, whereas fatty tissue is less conductive due to its anhydrous property. These properties allow the measurement of tissue volumes, shapes, or electrical properties through electrical impedance sensors. When the relative permittivity of tissue is plotted as a function of frequency f, three clearly distinguishable regions called relaxation regions can be observed. These are referred to as alpha (f <1000 Hz), beta (1000 Hz < f <100 MHz), and gamma (f > 1 GHz) regions.

[0021] The inventors of the present application have surprisingly found that the alpha and beta regions are particularly useful for tumor detection measurements, since most of the changes between normal and pathological tissue appear to appear in this frequency range.

[0022] The method described here has many advantages over other electrophysiological approaches, since it is non-invasive, allowing measurements to be made without skin damage caused by The electrodes also have a simple design in terms of transducers and instrumentation.

[0023] In different embodiments of the invention, the proposed method can be carried out using as components of each impedance measurement only the phase angle, only the magnitude of the impedance, or a combination of these.

[0024] In one embodiment of the invention, to separate the components of said bioimpedance, a vector can be constructed for each sine for each of the components of said bioimpedance, where the components can be magnitude, phase angle, or both. The vector is a set of values ​​that represent the magnitude and the impedance angle, or both if both parameters are used, corresponding to different excitation frequencies. In one embodiment of the invention, 100 magnitude values ​​and 100 additional phase values ​​can be generated corresponding to a sampling of 100 frequencies. In the case of a neural network that uses both sets of values, the generated vector will contain 200 measurements (the first 100 magnitude and the following 100 phase). In one embodiment of the invention, 100 magnitude values, 100 phase values, and 100 values ​​of the combination can be generated, for a total of 300 values.

[0025] In one embodiment of the invention, based on these vectors, a curve of the measurements of each of the bioimpedance components can be generated at all frequency points for each breast, as seen in Figs. 18, 19, 21 and 22. Particularly, this curve is made by making continuous sweeps of the excitation frequency in the different ranges (regimes) and capturing the impedance vector (magnitude and phase) for each of the frequencies and in the different electrode combinations. In addition, a curve can also be generated for the differential measurement between the components measured for one breast and the other, as shown in Figs. 20 and 23. The curves can be used to identify if there are errors in the values ​​of the measured impedance components, by detecting significantly abnormal values, in order to generate an alert in a user interface, so that the impedance components are measured again.This helps prevent errors in the subsequent processing of the measured values. The generation of these curves for reviewing the measured impedance component values ​​can be carried out before constructing each image for the impedance components measured in each sine individually and for the differential image, which represents the differences in impedance components between one sine and the other.

[0026] In one embodiment of the invention, a differential measurement of the bioimpedance components measured in each breast can be performed (the difference between the components obtained in one breast and those measured in the other breast is calculated). This is done in this way, taking into account that the inventors found that significant differences in the bioimpedance components measured in the right and left breasts imply the presence of alterations in the tissue of one of them.

[0027] In one embodiment of the invention, to determine the differences in each of the impedance components between one sine and the other, the values ​​in the phase and magnitude channels for the same combination of electrodes in the two sines are subtracted point by point at corresponding frequency points. The differential measurement is calculated by subtracting the resulting values ​​from the measurements taken in both sines to obtain the absolute difference.

[0028] In one embodiment of the invention, a differential image is constructed that represents the comparison between the impedance components measured in each breast, based on the differences in each of the impedance components between one breast and the other. To construct this image, grayscale encoded images of the vectors of each component (phase, magnitude, or both) of each breast are generated. FIGS. 1, 2, 7 and 8 illustrate the encoded images for the impedance vectors of the right breast and the left breast of two patients, while FIGS. 4, 5, 10 and 11 illustrate the encoded images for the phase vectors of the right breast and the left breast of two patients. The frequency values ​​at which the measurements were made are shown on the x-axis of these graphs, while the number of the electrode combination is shown on the y-axis.For example, if 6 electrodes are used, they can be combined in 30 different ways, so there would be 30 measurements for those 30 electrode combinations. These 30 combinations are represented on the y-axis of these graphs. The color intensity of each x,y coordinate corresponds to the value of the impedance or phase magnitude measured in each sine.

[0029] In one embodiment of the invention, to obtain images of each impedance component for each breast, measurements are taken at different points of each breast with electrodes located in the breast and in different combinations with each other. Each measurement is taken in a predetermined frequency sweep and in a forward and return direction in the measurement of impedance between electrodes. These measurements allow the generation of bioimpedance signals for a critical frequency range. These signals are separated into their impedance magnitude and phase components to construct grayscale coded images. The images generated from the measurements can be: of each sine separately and of the differential values ​​between the sines for both the impedance and phase values ​​at each measured frequency point and for each electrode combination.

[0030] The final differential image represents the difference in measurements between a patient's left and right breasts (differential images) for each measurement component analyzed. FIGS. 3, 6, 9, and 12 illustrate differential images for phase and magnitude for two patients. The x-axis of these graphs shows the frequency values ​​at which the measurements were made, while the y-axis shows the number of electrode combinations. For example, if 6 electrodes are used, they can be combined in 30 different ways, so there would be 30 measurements for those 30 electrode combinations. These 30 combinations are represented on the y-axis of these graphs. The color intensity of each x,y coordinate corresponds to the value of the difference in impedance magnitude or phase between one breast and the other.

[0031] In one embodiment of the invention, the neural network into which the images are fed to define a breast cancer risk classification has been pre-trained using a machine learning technique that uses a feedforward neural network with a hidden layer to process the differential image. The hidden layer may have half the number of neurons as the input layer. In one embodiment of the invention, the input layer may be composed of 30 neurons, while the hidden layer contains half this number. Preferably, the learning technique requires a training set of approximately 10 observations per parameter, considering both the number of neurons and the connections (30 * 15 * 10 = 4500 observations in total).

[0032] Alternatively, additional learning strategies such as support vector machine classifiers can be employed. In the presence of a large dataset, feedforward neural networks with a larger number of layers can be considered as the preferred modality, as well as 1D convolutional neural networks for processing individual signals or 2D neural networks for manipulating images generated from impedance signals, along with facial data, or their combinations, as well as additional data on potentially relevant characteristics and behaviors of the subject. patient.

[0033] In one embodiment of the invention, the neural network can be trained to classify patients as “healthy” or “unhealthy.” These two risk categorizations can be performed based on the Breast Imaging Reporting and Data System (BI-RADS). Specifically, the neural network is trained to classify patients as “healthy” if the generated images are similar to those of BI-RADS categories 1 and 2 with which it was trained, and as “unhealthy” if the generated images are similar to those of BI-RADS categories 0, 3, 4, 5, and 6 with which it was trained. To train the neural network, mammogram and ultrasound results that yield certain BI-RADS category results can be used, along with images generated by the techniques described above for various patients.In this way, the neural network is able to determine whether the images generated for a specific patient correspond to the "healthy" or "unhealthy" categories, based on previous training performed with these images for other patients and on the results obtained from mammograms and ultrasounds for the same patients. The neural network can be trained to perform risk classification using only the magnitude component of impedance, only the phase component of impedance, or both.

[0034] In one embodiment of the invention, the neural network's definition of breast cancer risk classification consists of classifying patients as "healthy" or "unhealthy" according to the previously explained BI-RADS classifications. This risk classification can be presented through a user interface.

[0035] The method described above can be executed through the processing means known in the state of the art for these purposes.

[0036] In another aspect, the present invention relates to a system for classifying a patient's risk of developing breast cancer, which executes the computer-implemented method described above. This system comprises electrodes; bioimpedance readers; processing means; and a user interface. The processing means are configured to receive the impedance values ​​measured in each of a patient's breasts and subsequently execute the computer-implemented method described above.

[0037] In one embodiment of the invention, the electrodes of the system are connected to a support that is shaped like a flower, the petals of which are connected to the electrodes and the center of which is a disc to accommodate the patient's nipple, as exemplified in FIGs. 16 and 17.

[0038] In one embodiment of the invention, the system comprises between 2 and 20 electrodes, preferably between 4 and 10 electrodes. In one embodiment of the invention, 6 electrodes can be used to perform measurements in each breast. Examples Example 1 - Application of the method of the invention in a BI-RADS IV patient

[0039] Measurements were performed in accordance with the present invention on a BI-RADS IV patient, obtaining about 6000 measurements in the alpha and beta frequency range. Six electrodes were used according to FIGs. 16 and 17, and impedance magnitude and phase measurements were obtained for the combination of electrodes 1-2, 1-3, 1-4, 1-5, 1-6, 2-1, 2-3, 2-4, 2-5, 2-6, 3-1, 3-2, 3-4, 3-5, 3-6, 4-1, 4-2, 4-3, 4-5, 4-6, 5-1, 5-2, 5-3, 5-4, 5-6, 6-1, 6-2, 6-3, 6-4, 6-5, for a total of 30 electrode combinations. For each combination, measurements were made at frequencies 2000, 2200, 2400, 2600, 2800, 3000, 3200, 3400, 3600, 3800, 4000, 4200, 4400, 4600, 4800, 5000, 5200, 5400, 5600, 5800, 6000, 6200, 6400, 6600, 6800, 7000, 7200, 7400, 7600, 7800, 8000, 8200, 8400, 8600, 8800, 9000, 9200, 9400, 9600, 9800, 10000, 10200, 10400, 10600, 10800, 11000, 11200, 11400, 11600 and 11800 Hz.Measurements were also made for each combination at frequencies 20000, 22000, 24000, 26000, 28000, 30000, 32000, 34000, 36000, 38000, 40000, 42000, 44000, 46000, 48000, 50000, 52000, 54000, 56000, 58000, 60000, 62000, 64000, 66000, 68000, 70000, 72000, 74000, 76000, 78000, 80000, 82000, 84000, 86000, 88000, 90000, 92000, 94000, 96000, 98000, 100000, 102000, 104000, 106000, 108000, 110000, 112000, 114000, 116000 and 118000 Hz.

[0040] Table 1 presents the first 10 measurements for the right breast and for the left breast, out of the 6000 performed for the frequency range of 2000 to 3800 Hz, as an example. Table 1. First 10 results of impedance measurements performed on a BI-RADS IV patient for the right and left breast

[0041] This table shows the magnitude and phase values ​​obtained for each sine for the crossing of electrodes 1-2.

[0042] Based on the results obtained for phase and magnitude for each sine and for each frequency measurement, the images in FIG 7 and FIG 8 were obtained for the magnitude of the impedance of each sine and FIG 10 and FIG 11 for the phase of each sine. The x-axis of these graphs shows the frequency values ​​at which the measurements were made, while the y-axis shows the number of the electrode combination (since measurements were made for 30 electrode combinations, this axis goes from 0 to 30). The intensity of the color of each x,y coordinate corresponds to the magnitude value of the impedance or phase obtained.

[0043] In addition, from the measured values, the difference between each component of the impedance between one sine and the other was calculated. Based on the results, the differential image according to FIG 9 for impedance magnitude and the differential image according to FIG 12 for the phase. The x-axis of these graphs shows the frequency values ​​at which the measurements were made, while the y-axis shows the number of the electrode combination (since measurements were made for 30 electrode combinations, this axis ranges from 0 to 30). The intensity of the color of each x,y coordinate corresponds to the value of the difference in impedance magnitude or phase obtained between one sine and the other.

[0044] These images were fed into a neural network trained according to the present invention, and a classification of the patient as "unhealthy" was obtained. The classification was subsequently confirmed with the patient's mammogram result. For the purposes of the study, the mammogram was taken after the impedance measurement, but on the same day to ensure comparable results. During the validation process, the classification determined by the mammogram was kept hidden and was only revealed once the coded image of the patient had been generated, fed to the neural network, and it had generated its own classification. At that point, classifications were compared between both examinations to determine the accuracy of the impedance measurement compared to mammography. Example 2 - Application of the method of the invention in a BI-RADS I patient

[0045] Measurements were performed in accordance with the present invention on a BI-RADS I patient, obtaining about 6000 measurements in the alpha and beta frequency range. Six electrodes were used according to FIGs. 16 and 17, and impedance magnitude and phase measurements were obtained for the combination of electrodes 1-2, 1-3, 1-4, 1-5, 1-6, 2-1, 2-3, 2-4, 2-5, 2-6, 3-1, 3-2, 3-4, 3-5, 3-6, 4-1, 4-2, 4-3, 4-5, 4-6, 5-1, 5-2, 5-3, 5-4, 5-6, 6-1, 6-2, 6-3, 6-4, 6-5, for a total of 30 electrode combinations. For each combination, measurements were made at frequencies 2000, 2200, 2400, 2600, 2800, 3000, 3200, 3400, 3600, 3800, 4000, 4200, 4400, 4600, 4800, 5000, 5200, 5400, 5600, 5800, 6000, 6200, 6400, 6600, 6800, 7000, 7200, 7400, 7600, 7800, 8000, 8200, 8400, 8600, 8800, 9000, 9200, 9400, 9600, 9800, 10000, 10200, 10400, 10600, 10800, 11000, 11200, 11400, 11600 and 11800 Hz.Measurements were also made for each combination at frequencies 20000, 22000, 24000, 26000, 28000, 30000, 32000, 34000, 36000, 38000, 40000, 42000, 44000, 46000, 48000, 50000, 52000, 54000, 56000, 58000, 60000, 62000, 64000, 66000, 68000, 70000, 72000, 74000, 76000, 78000, 80000, 82000, 84000, 86000, 88000, 90000, 92000, 94000, 96000, 98000, 100000, 102000, 104000, 106000, 108000, 110000, 112000, 114000, 116000 and 118000 Hz.

[0046] Table 2 presents the first 10 measurements for the right breast and for the left breast, out of almost 6000 performed for the frequency range of 2000 to 3800 Hz, as an example. Table 2. First 10 results of impedance measurements performed on a BI-RADS I patient for the right and left breast

[0047] This table shows the magnitude and phase values ​​obtained for each sine for the crossing of electrodes 1-2.

[0048] Based on the results obtained for phase and magnitude for each sine and for each frequency measurement, the images in FIG 1 and FIG 2 were obtained for the magnitude of the impedance of each sine and FIG 4 and FIG 5 for the phase of each sine. The frequency values ​​at which the measurements were made are shown on the x-axis of these graphs, while the y-axis shows the number of the electrode combination (since measurements were made for 30 electrode combinations, this axis ranges from 0 to 30). The intensity of the color of each x,y coordinate corresponds to the magnitude value of the impedance or phase obtained.

[0049] In addition, the difference between each impedance component from one sine to the other was calculated from the measured values. Based on the results, a differential image was constructed according to FIG. 3 for impedance magnitude, and a differential image according to FIG. 6 for phase. The x-axis of these graphs shows the frequency values ​​at which the measurements were made, while the y-axis shows the number of the electrode combination (since measurements were made for 30 electrode combinations, this axis ranges from 0 to 30). The color intensity of each x,y coordinate corresponds to the value of the difference in impedance magnitude or phase obtained between one sine and the other.

[0050] These images were fed into a neural network trained according to the present invention, and a "healthy" patient classification was obtained. The classification was subsequently confirmed with the patient's mammogram results. For the purposes of the study, the mammogram was taken after the impedance measurement, but on the same day to ensure comparable results. During the validation process, the classification determined by the mammogram was kept hidden and was only revealed once the coded image of the patient had been generated, fed to the neural network, and it had generated its own classification. At that point, classifications were compared between both examinations to determine the accuracy of the impedance measurement compared to mammography. Example 3 - Comparison between the classification results of the method of the present invention and the results of mammograms.

[0051] The neural network was trained with measurements from 910 patients, and the method of the present invention was used to obtain risk classifications according to "healthy" and "unhealthy" for a percentage of these 910 patients. These same patients underwent mammography to categorize them as "healthy" or "unhealthy."

[0052] As a result of the analysis of the image of each sine and of the differential image in the neural network, the graphs illustrated in FIGs. 13, 14 and 15 were obtained. FIG. 13 illustrates a diagram obtained by using the phase and magnitude components (combined) of the measured impedance. Alternatively, FIG. 14 illustrates a diagram obtained by using only the magnitude of the measured impedance and FIG. 15 illustrates a diagram obtained by using only the phase of the measured impedance.

[0053] These diagrams compare the neural network output and mammogram classification results for the same patient, with both readings performed on the same day. This allowed us to determine the accuracy of the results obtained using the neural network.

[0054] In the diagrams, the “Label” axis indicates the classification between “healthy = 0” and “unhealthy = 1” dictated by the patient’s actual mammogram. The classification between “healthy” and “unhealthy” is determined by the BI-RADS classification of the mammogram. BI-RADS 1 and 2 indicate a “healthy = 0” classification. BI-RADS 0, 3, 4, 5, and 6 indicate an “unhealthy = 1” classification.

[0055] The "Prediction" axis indicates the classification between "healthy = 0" and "unhealthy = 1" given by the canonical neural network trained on the differential phase components and impedance magnitude between the right and left breasts of the patient being examined. Interpretation of the diagram allows us to identify, from a defined sample, the total number of correct matches of the neural network's prediction with the actual mammographic classification. The value indicated in the center of each quadrant indicates the number of patients whose mammographic classification falls within that quadrant.

[0056] The upper left quadrant of the diagrams in Figs. 13 and 14 indicates the number of patients in the sample that the neural network classified as “healthy = 0” and that successfully matched the mammogram classification also as “healthy = 0”. The lower right quadrant indicates the number of patients in the sample that the neural network classified as “unhealthy = 1” and that successfully matched the mammogram classification also as “unhealthy = 1”. The adjacent cases indicate the number of patients whose neural network classification result does not match the actual mammogram classification result. The upper right quadrant indicates the number of patients in the sample that the neural network classified as “unhealthy = 1” and that do not match the actual mammogram classification which was “healthy = 0”.Finally, the lower left quadrant indicates the number of patients in the sample that the neural network classified as “healthy = 0” and that do not match the classification of the actual mammogram which was “unhealthy = 1”.

[0057] The usefulness of this diagram is to allow verification of the accuracy of the neural network when its output is directly compared to that of a mammogram. Mammography currently

Claims

It is recognized as the Gold Standard in screening and therefore it is pertinent to use it as a reference. [0058] The conclusions of the diagram are interpreted with the results in each quadrant and their proportion of the total sample. The possible results indicate the sensitivity and specificity of the neural network and in turn the proportion of false negatives and positives in the given sample. [0059] The method used in the diagram is comparative between the result of the neural network and mammography classification results in the same patient with both readings performed on the same day. [0060] These results demonstrate that both graphs work and can be adapted to different types of neural networks depending on the volume of patients available for training the neural network. In all these results, a large number of coincidences can be seen between the prediction of the method of the present invention and the mammography / ultrasound results, thus demonstrating that said method does work as a valid mechanism for making an initial risk classification, so that the patient can subsequently go to the confirmatory diagnostic tests. CLAIMS 1. A computer-implemented method for classifying the risk of breast cancer in a patient, comprising: using bioimpedance measurements taken using different combinations of electrode pairs at different points in each breast for a critical frequency range, separating the components of said bioimpedance into impedance magnitude and phase; determining, for each electrode combination, the differences in each of the impedance components between one breast and the other; constructing an image for the impedance components measured in each breast. individually and a differential image representing the differences in the impedance components between one breast and the other; feeding the image of each breast and / or the differential image to a pre-trained neural network to define a breast cancer risk classification; and presenting the breast cancer risk classification through a user interface; wherein the images for the impedance components measured in each breast and the differential image present on the x-axis the frequency at which the impedance measurements were made, on the y-axis the combination of the electrodes and in the x-coordinates, and a color intensity that represents the value of the impedance components or the difference between the impedance components measured between one breast and the other.

2. The method according to claim 1, wherein the components of the bioimpedance measured in each breast are selected from magnitude, phase and / or a combination thereof.

3. The method according to claim 1, wherein the frequency range of the bioimpedance measurements is equal to or less than 120MHz.

4. The method according to claim 1, wherein the frequency range of the bioimpedance measurements is divided into a first frequency range between 2000 Hz and 12000 Hz and a second frequency range between 12000 Hz and 120000 Hz.

5. The method according to claim 1, wherein each frequency interval is divided into between 2 and 100 measurements.

6. The method according to claim 1, wherein to separate the bioimpedance components, a vector is constructed for each sine for each of the components of said bioimpedance.

7. The method according to claim 1, wherein from bioimpedance measurements made for each combination of electrodes in each breast for a critical range of frequencies, a curve of the measurements of each of the bioimpedance components at all frequency points for each breast is also generated.

8. The method according to claim 1, wherein the differences in each of the impedance components between one sine and the other for each combination of electrodes are determined by subtracting point by point the values ​​of each component for the same combination of electrodes in the two sines at corresponding frequency points.

9. The method according to claim 1, wherein the classification of risk of occurrence of breast cancer has two categories: “healthy” for BI-RADS categories 1 and 2 and “unhealthy” for BI-RADS categories 0, 3, 4, 5 and 6.

10. A system for classifying a patient's risk of breast cancer, comprising: electrode pairs; bioimpedance readers; processing means; and a user interface; wherein the processing means are configured to: receive the impedance values ​​measured by different combinations of electrode pairs; and execute the computer-implemented method of any one of claims 1 to 9.

11. The system according to claim 10, wherein the electrodes are connected to a support that is shaped like a flower, the petals of which are connected to the electrodes and the center of which is a disc to accommodate the patient's nipple.

12. The system according to claim 10, wherein the system comprises between 2 and 20 electrodes.

Images