ES2311413A1 - System for estimating the titration of breastfeeding - Google Patents

Abstract

The invention relates to a system for estimating the titration of breastfeeding, that is the volume of milk ingested by a newborn during breastfeeding. Said system establishes a fractal model of the human breast with the aim of estimating the total volume of milk stored therein. The system according to the invention uses a system for emitting and receiving ultrasounds for capturing an image before the child starts to breastfeed, and another after. Said images are pre-processed with both DCT and DWT before the estimation. The method for estimating the titration is based on a group of neuronal networks with the aim of reducing error between the actual value and the estimated value by reducing the bias (systematic error) and the variance of the estimating system using an average of the responses of the group of neuronal networks. An algorithm is also used to reduce said systematic error on the basis of image segmentation techniques combined with said fractal system.

Description

System for estimating the volumetry of breastfeeding

Field of the Invention

The present invention develops a system, including a digital ultrasonic scanning device, for the estimation of the volumetry of milk ingested by a newly born during the breastfeeding process, configuring itself as a handheld ultrasonic ultrasound device. He system provides the final measurement through the information of ultrasound obtained by the operating procedures of the same.

Background

The benefits of breastfeeding both for women as for newborns, they are well known by the medical community and this is confirmed by numerous publications in the field of pediatrics. In fact, both the Organization World Health Organization (WHO) as the Children's Fund for The United Nations (UNICEF) recommends that all mothers breastfeed their newborns exclusively during the first six months of life. In the recommendations published in the "Healthy People 2010" report, the US Department of Health sets the goal of achieving a 75% prevalence of breastfeeding during postpartum along with a 50% prevalence during the first six months of life.

Although the current prevalence during the postpartum is relatively high (about 70% in the US, 80% in Australia, 90% in Germany, etc.), less than half of newborns receive breastfeeding at six months of life. Although it is difficult to identify the direct causes of early termination of breastfeeding, the last published pediatric studies suggest that there are numerous difficulties arising from the process of breastfeeding during first weeks of postpartum, which seems be that they have a great correlation with this discontinuity early.

These problems during the first 4 weeks of life are quite common. One third of mothers refer perceive at least one or more problems related to breastfeeding, such as lack of confidence in your ability to give the chest, perception of not having enough milk, problems with suction by the newborn, etc. In turn, they have published other studies that raise the rate of women who suffer Such problems up to 38%.

WO2006054287 patent develops a system which integrates various devices integrated into a bra, with openings that allow breast exposure to breastfeed newborn. Inside said bra are various ultrasound emitting / receiving probes, which use the Doppler Effect to calculate milk flow and perform at from this the estimate of the volume of milk taken by the newborn.

However, such a system does not turn out to be a suitable medium for estimating the volume of milk taken for the newborn, since it is a highly system invasive and disturbing intimacy between mother and newborn born by forcing to breastfeed with a truly garment uncomfortable to be able to perform the volumetric estimation.

US5827191 and WO2006003655 patents develop a system for measuring milk volumetry ingested by a newborn based on a elastic nipple cover that passes through, with the purpose of measuring it, the flow of milk taken.

Both systems have the disadvantage of establish a physical barrier between the mother and the baby disturbing not only the intimacy between the two, but also the very process of breastfeeding, since it makes it difficult for the infant to suck milk production derived from direct nipple stimulation by the child

It persists however without solving the question of  establish an autonomous system for volumetric estimation of breastfeeding that is accurate, comfortable, non-invasive, easy to use and should not be in contact with the mother or the newborn during the entire breastfeeding process. In specifically, the need to establish a system that helps to solve all the problems described above.

That is why one of the objectives main of the present invention is to give solution to the problematic derived from the breastfeeding process of newborns  during the first six months of life providing data on real time to new mothers about the volume of milk ingested For your newborn.

Another objective and advantages obtained with the ease and portability of the present system invention refers to the increase in the quality of life and comfort of new mothers and newborns since it it requires the minimum contact for the realization of the estimate (requires only two pre and post suction scans), being another of The main objectives of the present invention extend the breastfeeding process beyond the first six months of lifetime.

Brief Explanation of the Invention

According to the background of the invention detailed above, the system for estimating the volumetry in breastfeeding uses a device contact ultrasound using the physical principle of Back-Scatter with which a first one is made scan of the mother's breast providing a first set of parameters that can be included, the total volume of the chest, the estimated number of lobes and lobules (units Functional mammary gland), the number of channels lactophores, tissue acoustic properties, etc.

With the data set obtained previously a fractal model is obtained by which the volume is obtained of stored milk (total capacity) in the lactophores and in the whole of the mammary gland (especially the lobules) This model also allows an estimate of the milk production rate once consumed the reserve volumetric calculated previously with the model.

Once the previous measurements have been carried out together with the first estimate of the volumetry and once the process of breastfeeding the newborn with a breast, we proceed to perform a second scan by contact with the mother's breast with The ultrasonic sensor With this second scan you get the same data set detailed above.

With the available data set, it is inferred a function / system that, from an image input of the breast, before and after breastfeeding be able to provide volume Total milk consumed. To do this, this estimate is modeled with a classification process based on different categories (input data).

For this, a preprocessing of the images prior to the estimation phase. The system presented in this invention can use the discrete transform of cosine (Discrete Cosinus Transform / DCT) or the matrix's own values autocorrelation of the input images. The implementation preferred of the present system performs said preprocessing by Wavelet transforms (Discrete Wavelet Transform / DWT), since captures the specifics of the input ultrasound (i.e. images of the two scans made with the sensor ultrasound) at different levels of resolution, and the DCT (Discrete Cosinus Transform).

Once the preprocessing of the incoming ultrasound is performed, the volume of milk ingested by the newborn is estimated through the implementation of a committee of neural networks of the "Radial Basis Functions" type, which calculate a function that returns the volume ingested from Preprocessed images. In order to reduce the variance of the previous estimate, the neural networks committee is constituted by a set of networks, which is trained from a sampling of the input image, so that by averaging the outputs of said networks the variance of the estimate is compensated and, at the same time, the input of each neural network has a lower dimension. This committee of neural networks can be combined with image processing techniques and, in particular, with the segmentation of preprocessed images to calculate the degree of collapse of the breast tree (ie of all breast tissue) and, as of this degree of compression make an estimate of the volume of milk ingested (this estimate is direct since the total volume of milk available from the fractal model of the
chest).

Brief explanation of the drawings

Figure 1 shows a representation of the lower part of the system for estimating breastfeeding maternal where the ultrasonic emitting device is detailed.

Figure 2 shows a representation of the system profile for breastfeeding estimation where  It is appreciated that it is a non-invasive handheld device.

Figure 3 shows a superior representation of the system for estimating breastfeeding where appreciate the screen where the data of the readings of the volumetry and the temporal evolution of them.

Figure 4 shows the tall architecture global system level for ultrasonic uptake and implementation of estimation algorithms. The components of system include an ultrasonic emitter, a switch transmission / reception, analog converters / digital devices FPGA or DSP and the interconnection elements between plates.

Figure 5 shows the minimum, average and volume maximum milk stored in a breast (BSC) obtained from the Application of the fractal model of the present invention.

Figure 6 shows the filter bank that implement the DWT (Discrete Wavelet Transform) of this invention. The high-pass components are detailed and low-pass filter bank as well as levels of resolution for the calculation of the wavelet transformation.

Figure 7 shows the block diagram for the estimator based committee of neural networks of the Radial type Basis Functions, which perform the estimation of the volumetry of the breastfeeding from image samples preprocessed (preprocessed ultrasound).

Figure 8 shows the block diagram and operations to estimate lactation volumetry maternal combining the fractal model defined herein invention with image segmentation techniques.

Detailed Explanation of the Invention

The present invention consists of a system for the estimation of the volume of milk ingested by a newly born during breastfeeding whose data are evaluated in real time by means of an ultrasound capture device (1) linked to different digital subsystems (7, 8, 9) that implement, respectively, a fractal model (7) of the chest human, a committee of classifiers based on neural networks (8) of the RBF type and, finally, another subsystem (9) that calculates the degree of collapse of the breast tree. Although the implementation Preferred system uses digital fractal subsystems (7) and from the neural network (8) above, the breastfeeding estimator of the present invention can be integrated with the tree collapse measurement sub-system breast (9) to reduce variance of estimates made.

The device (1) consists of a transmitter and ultrasonic receiver (2), of a transmission card responsible for generating the ultrasound pulses of the conformation of the ultrasound emission beam and conversion digital / analog of said pulses and beam shaping ("beaforming") of the issuer. In the preferred implementation of system, said sender / receiver device (2) is implemented using a single DSP type FPGA microcontroller device.

It also includes the device (2) of one or several receiving devices (5) (one for each channel according to the previous beam shaping) whose function is the reception of the pulses emitted by the sending card (4), the conversion analog / digital and the conformation of ultrasound, which they will use as input for the estimation of milk volume ingested In the preferred implementation of the system, said Receiving elements are implemented by a device DSP / FPGA for each channel. Depending on the degree of reliability desired in the final system, the number of ultrasound channels It can vary between 8 and 64.

Finally, the device (1) consists of a connection panel between cards, which interconnects the card station (4) and receivers (5) (8-64 channels) with the controller card (6) which is responsible for not only control the receiving transmitter (2) but also on it They implement the filters and models detailed below. In the preferred implementation of the system, said controller is implemented through a FPGA type DSP device.

For the calculation of breast volume and interpretation of the images obtained by ultrasound the invention has the appropriate means of computation and processing of information that allow you to apply different procedures according to different subsystems of information processing received

Subsystem (7) implements a fractal model described below. In the breast tree, the channels bifurcate dichotomously, reducing their length and diameter systematically. On the other hand, it is estimated that the breast tree ends in approximately 2 10 lobules. Each of these lobules is divided into 4 generations of alveoli, which constitute the mammary gland. Considering the lower part of the breast tree (ie generations 5 to 10) and assuming that milk flows in the canal system following the Poseuille law (in fact it is a very good approximation for the area of the tree considered), structures can be deduced optimal by minimizing viscosity dissipation ("viscous dissipation") within the volume of the tree. In fact, the detailed analysis of this proposition by means of Lagrange multipliers suggests that the best tree structure is fractal with dimension 3. In this ideal tree, the successive segments will be homothetic with a constant size / radius ratio ( h ). In fact, this result can be considered as a particular case of the Hess-Murray law, which models the blood tree.

Assuming that the branches that bifurcate are symmetrical (this assumption turns out to be quite reasonable for the inner part of the chest), the ratio between the diameter and the length between the p-1 generation and the p generation is h p. Denoting V as the volume of a given channel, reduced by a factor h, a channel is multiplied by a factor h ^ {3} for each generation volume. For after p generations, the sizes will have been reduced by a factor h 1 xh 2 x .. xh p so that the volume of a tree of N + 1 generations (indexed from 0 to N ) can be written as follows:

one

If we assume that the reduction factor is constant between generations, the volume can be rewritten how:

2

Rewriting the previous equation as a equation in 5 finite differences (EDF), the volume can be written as follows:

V (n + 1) = V (n) + V_ {0} r ^ n-1}

The term Nth can be written as:

3

Where r = 2h3 . However, until now only the volume of milk stored inside the breast tree has been calculated, so the contribution to the total volume derived from the lobules has not yet been considered. As will be seen below, this volume is much larger than the one calculated above.

As mentioned earlier, there is a one-to-one relationship between the number of lactophores channels in the last iteration ( N ) and the number of lobules. In other words to a breast tree with 2 N branches, it corresponds to 2 N lobules. Therefore, assuming a spherical shape for the lobules with an average radius h , the total chest capacity (BSC / "Breast Storage Capacity") can be written as follows:

4

The value of h can be calculated from the ultrasounds taken in the subsystem (3) since this value is the relationship between the length of the lactophore channel and its radius. In conclusion, the subsystem (4) is responsible for estimating this factor h and calculating the BSC, using the above equation from h and the number of terminal lactophores.

On the other hand, the fractal model presented above not only defines the BSC but also provides the theoretical basis for the subsystem (9) since during a milk intake, the amount of breast tissue must be kept constant. In other words, this implies that any change within the breast during a milk intake is due to a pressure drop (ie the structure of the tree of channels and lobules collapses during the "emptying" of the breast). Therefore, the volume of milk consumed by the newborn can be estimated from the BSC and the degree of breast tissue collapse. Figure 5 shows the minimum, average and maximum BSC for a chest as a function of h .

Once the first scan of ultrasound (ultrasound) from which the data detailed above (Basically BSC, h and number of lactophores channels), a second scan is performed (ultrasound) after taking for automatic real-time estimation of the milk taken by the newborn.

On the other hand, the subsystem inputs (8) consist of two images obtained from the echographs of the receiving transmitter (2) made before and after breastfeeding. The present subsystem consists of two distinct phases. One defines the preprocessing unit, which is designed with the purpose of obtaining invariant characteristics / data regarding translations, rotations and / or scaling (both in frequency and time) from images / ultrasound. The second phase is based on " Machine Learning " techniques and computes the total volume from the data obtained from the preprocessed images / ultrasound.

The preprocessing phase of the images (10) It aims to transform the input image into a vector, that captures the characteristics associated with the estimation problem of the volumetry of breastfeeding. This vector is invariant. in relation to the rotation of the input images, changes in the measurement angle (or ultrasound capture), common characteristics related to milk production that are independent of the person, etc ...

In the preferred implementation of said phase of preprocessed images (10) two different techniques are used which can be used independently or in conjunction with the objective of obtaining the data vector described above. The First technique uses the Discrete Cosine Transform (Discrete Cosinus Transform or DCT) and the second uses the Two-dimensional Wavelet Transform (Discrete Wavelet Transform or DWT)

In the present invention the transformed DCT because it computes a representation of the image as a combination of horizontal frequencies and vertical Therefore, with this representation, the details of the image that possess certain frequency characteristics, they will appear along a circle depending on the angle that said  feature have with respect to horizontal axes and vertical. Therefore, the representation of the input image in the transformed domain DCT will be invariant with respect to the rotations that can be done on the original image. For other part, only part of the image in the transformed domain will be relevant, so you can select the region of interest of the image in the transformed domain in order to reduce the redundancy while preserving all the information contained in the original image.

The second approach to preprocessing (10) of the input image is based on the DWT. The application of this preprocessing of the input images in conjunction with the DCT or independently is considered since the scattering properties of the input ultrasound can be uniquely characterized by their spatio-temporal properties. Normally, these properties have been studied by the Discrete Two-Dimensional Fourier Transform (2D-DFT) or by the STFT (Short-Term Fourier Transform). It is a well-known result that both 2D-DFT and STFT cannot simultaneously represent the temporal and frequency properties at the same time, and with different degrees of resolution, of an input image.

The DWT presents the ideal framework for the problem of estimating the volume of breastfeeding since provides a representation of the images in which it is separated the relevant information (presence of tubes, layers of tissue, etc.) of information related to the smooth change of scales Gray, artifacts due to wave diffraction, the effect image scaling and rotation because each measurement is performed under slightly different conditions, etc.

DWT coefficients describe the correlation between the selected wavelet (in the implementation preferred wavelets of the present invention have been selected Haar and Daubechines) and multi-scale image / resolutions (i.e. the degree of similarity between the image and the wavelet is measured for a  time-discrete combination and position). In others words, the calculated coefficients provide the amplitudes of the series of wavelets on a set of scales and translations, which must be added in order to obtain the original image. Since this perspective, DWT analysis can be understood as a search on the image of interest of characteristics that resemble the selected wavelet. This search is done on Several scales and several wavelet sizes. For this reason, they have selected Haar Wavelets to identify channels Lactophores while Daubechines Wavelets are used both for the detection of lobules and for the elimination of artifacts derived from the absorption and scattering of initial images.

The implementation of this analysis in the preprocessing (10) is carried out by means of a filter bank (figure 6). With this set of filters, the input image is divided into various spectral components (" sub-band coding ") in order to obtain the high-pass, low-pass and pass-band components of the input signal. These components are obtained by applying filters h (high pass) and g (low pass). The application of filter h is analogous to the application of the selected wavelet to the input image while the application of filter g is analogous to the application of a scaling or smoothing function to the input signal.

The subsystem (8), apart from the phase of preprocessed (10) described above, also includes a module (11) whose objective is to infer a general function from a set of available examples (ultrasounds stored in a Database). From this general function and from both preprocessed images volume estimation can be performed of ingested milk.

Since the dependence estimate between the volume and the two ultrasounds available before and after the breastfeeding is quite complicated, that the dimensionality of the entries are really high and the number of examples available low, the use of a neural network of the Radial Basis type is proposed Function (RBF) to solve the problem described above. The RBF neural network is based on the computation of centroids (the most intuitive function that can be computed is an interpolation between values using a "lookup table"). Said computation of centroids and interpolation assumes that the underlying function that should estimate is smooth (low variability) so that the effect of noise in the input data (ultrasound) results in a uncertainty in the estimated volume, which will be around centroid, but said effect will be limited since said volume of uncertainty is not too high in the system presented.

On the other hand, the criteria to evaluate the Estimated function quality will be the Mean Square Error (MSE)  applied to a control or validation sequence on a set of elements of the original Ultrasound Database. The performance of the RBF learning algorithms, which infer the function of volume from the examples and entries available, can be improved by various techniques. These techniques include the combination of classifiers and / or training of classifiers with distributions non-uniform data, coding of the output of the classifiers or exploitation of the properties of Kernels due to the high dimensionality of the data. These Methods for improving estimator performance are analyzed by hoping for the input / output data and through the relationship between the dimensionality of the input and the number of samples.

The error on the measurement of the module (11) can  decompose as the sum of the squares of the bias of the estimation and variance of it. By aggregating neural networks, the estimation error can be controlled by the decrease of both bias and variance. Thus, if you train the set of neural networks so that the output of each network is a random variable, the effect of computing the average of all outputs will result in a reduction of the variance and, therefore, of the estimation error. Bliss reduction will be a function of the number of neural networks that are use in the system. In the preferred system implementation, Depending on the degree of reliability required, they can be used between 1 and 100 neural networks.

The neural networks of the present invention are  organized in a committee type structure so that, given a common input, each network uses a sampled version of the common input and the output of each of them is calculated in parallel. The effect of sampling the common input is that it Enter a different systematic error in each network. Thus, when you add the output of a group of neural networks, put that each network has a different bias, the estimation error it decreases even more because the error is now taken as average systematic and to the reduction of the variance described previously.

In the preferred implementation of this invention, it is proposed to train a committee of neural networks RBF with sampled input images (or segmented images) of so that each neural network has as input an image different so that when you add all the outputs of the neural networks by calculating the mean of all outputs of the neural networks, the systematic error will be less that the one that would be obtained with RBF trained with the whole image and not with sampled versions (or image segments) of it. The Figure 7 shows the block diagram of this system.

Finally, in order to further reduce the systematic error of volumetric estimation, the system (8) can be combined with the system (9) (figure 8) where the input images are segmented by image processing techniques, which include the opening, smoothing, " overlay ", calculation of the perimeter / segmentation and estimation of the area of the previous perimeter. The ratio of areas calculated by this subsystem applied to the fractal model (BSC) described above provides a second estimate of the completely incorrect volume with which it has been calculated with the module (11) figure 7 so, if the average between the two is taken estimates, the systematic error decreases.

It is planned that the handheld device implemented in the system of the invention can function in a way  autonomous thanks to incorporating batteries (12). The system generates a alarm when a malfunction occurs, for example in the capture of an ultrasound or by default or low load of the batteries (12).

It is understood that in the present case they can be variable how many alterations of finish or form not modify the essence of the invention.

Claims (24)

1. System for estimating breastfeeding volumetry characterized by - incorporate a handheld device (1) no invasive for said estimation of the volume of milk ingested by a  neonate - use an ultrasonic receiver emitter (2) integrated in said handheld device (1) for pickup of an ultrasound before breastfeeding and another after same - having said device (1) of means of computation, data storage and processing thereof, suitable for the treatment of the images obtained - have operating procedures intended to estimate the volume of milk ingested by the neonate in real time in measures of cubic centimeters, milliliters, ounces, etc. - present the estimated volumetry reading on a screen (3) of said device (1). 2. System for estimating breastfeeding volumetry, according to the preceding claim, characterized in that said operative procedures can simultaneously establish different subsystems that perform independent estimates of the total volume of milk stored in the breast (BSC / Breast Storage Capacity), said combination of subsystems aimed at reducing the systematic error in the measurements of said total volume of milk consumed and increasing its degree of precision. 3. System for estimating the volumetry of breastfeeding, according to the preceding claims, characterized in that said device (1) uses a contact ultrasonic sensor to capture ultrasound before and after breastfeeding. 4. System for estimating breastfeeding volumetry, according to the preceding claims, characterized in that said ultrasound capture is carried out by means of a transmitter card (4) and an ultrasound receiver card (5). 5. System for estimating the volumetry of breastfeeding, according to the preceding claims, characterized in that the system for ultrasonic uptake is implemented in said handheld device (1) by means of DSP devices, and / or microcontrollers, for example performed by FPGA chips. 6. System for estimating breastfeeding volumetry, according to the preceding claims, characterized in that the system for ultrasonic uptake (2) can be implemented with approximately 1 to 10 FPGA devices according to the

 \ hbox {number of channels of
desired reception (for example, in a range of 8 to
64).} 

7. System for estimating the volumetry of breastfeeding, according to the preceding claims, characterized in that the system for ultrasonic capture (2) has a coordinating card (6) for the conformation of the images (ultrasound) captured by the emission cards (4) and reception (5) of ultrasound. 8. System for estimating the volumetry of breastfeeding, according to the preceding claims, characterized in that the emission cards (4) and reception (5) are connected to the coordinating card (6) by means of a serial connection panel. 9. System for estimating breastfeeding volumetry, according to the preceding claims, characterized in that said subsystems for estimating volumetry are stored in said coordinating card (6). 10. System for estimating breastfeeding volumetry, according to the preceding claims, characterized in that the coordinating card is implemented with a DSP device, and / or microcontrollers, in turn made for example by FPGA chips. 11. System for estimating the volumetry of breastfeeding, according to the preceding claims, characterized in that the images obtained are preprocessed before estimation by the transformed DCT (Discrete Cosinus Transform). 12. System for estimating breastfeeding volumetry, according to the preceding claims, characterized in that the images obtained can also be preprocessed before estimation by means of the DWT (Discrete Wavelet Transform) transform. 13. System for estimating breastfeeding volumetry, according to the preceding claims, characterized in that the wavelets selected for preprocessing are the Haar and Daubechines wavelets. 14. System for estimating the volumetry of breastfeeding, according to the preceding claims, characterized in that the procedure for estimating the volume of ingested milk is performed from two preprocessed images (before and after breastfeeding). 15. System for estimating the volumetry of breastfeeding, according to the preceding claims, characterized in that said estimation of the volume of ingested milk is performed by a subsystem with a fractal model (7) of the human breast. 16. System for estimating the volumetry of breastfeeding, according to the preceding claims, characterized in that said estimation of the volume of milk ingested is performed by neural networks. 17. System for estimating the volumetry of breastfeeding, according to the preceding claim, characterized in that the anterior neural networks are approximately configured as a committee of 1 to 100 classifiers. 18. System for estimating the volumetry of breastfeeding, according to claims 16 and 17, characterized in that to reduce the systematic error of the system, each element of the classifier committee has as input a sampled version of the preprocessed images. 19. System for estimating breastfeeding volumetry, according to the preceding claims, characterized in that to reduce the systematic error of the system image segmentation techniques can be used to estimate the degree of collapse of the breast tree. 20. System for estimating the volumetry of breastfeeding according to the preceding claim, characterized in that the volume of milk ingested can be estimated by a subsystem (9) from the BSC and the degree of compression of the breast tree. 21. System for estimating the volumetry of breastfeeding, according to the preceding claims, characterized in that said device (1) works with batteries (12). 22. System for estimating breastfeeding volumetry, according to the preceding claims, characterized in that the system generates an alarm in the event of an error (batteries (12), ultrasound capture, etc.). 23. System for estimating breastfeeding volumetry, according to the preceding claims, characterized in that the system can have the appropriate memory registers and be configured to store the data of various readings and give an evolution on the shots along weather. 24. System for estimating the volumetry of breastfeeding, according to the preceding claims, characterized in that the system can have the appropriate means of communication and be configured to dump the readings to a computer for analysis.