JP4882052B2 - Pulse wave diagnosis system using self-organizing map, self-organizing map generating program and generating method - Google Patents

Description

  The present invention relates to a pulse wave diagnostic apparatus using a self-organizing map (SOM).

  The self-organizing map (hereinafter referred to as “SOM”) is an unsupervised neural network devised by Kohonen of Helsinki University of Technology, using a neural network as a hint for displaying high-dimensional data compressed in two dimensions. It has been realized. According to this SOM method, it is possible to project multidimensional data on a two-dimensional map, and the data can be easily classified visually. Therefore, the application of SOM has been studied in various fields. (Non-Patent Document 1).

  For example, there is an invention disclosed in Patent Document 1 regarding the application of SOM in the field of health examination. In Patent Document 1, SOM is generated using data on various health conditions of a subject as an input vector. It is possible to easily visually recognize the patient's health status by inputting data related to the patient's health status and highlighting the corresponding node position on the SOM for the SOM obtained in this way. It becomes.

In general, information used for diagnosis of a patient's health condition includes blood pressure, cholesterol, neutral fat, pulse wave, and the like. In particular, pulse waves have been used for diagnosis of the degree of arteriosclerosis and the like, since they are easy to collect and contain a lot of information on vascular dynamics from the center to the periphery. Patent Document 2 discloses a method for generating a health check SOM using data related to a pulse wave. According to Patent Document 2, it is possible to obtain an SOM that can estimate information on a human body other than a pulse wave from pulse wave data, and it is possible to acquire data relating to a health condition in a short time.
Heizo Tokutaka, Kikuo Fujimura, supervised by Yamakawa Retsu, “Self-Organizing Map Application Examples-Visualization Information Processing by SOM”, Kaibundo, October 2002 JP 2003-263502 A JP 2003-310558 A

  In Patent Document 2, an SOM is created by using data obtained by sampling a waveform from the maximum point to the minimum point in one cycle of the acceleration pulse wave at 12 points as data relating to the pulse wave. The waveform of the acceleration pulse wave actually has a complicated shape, and this method cannot cope with various shapes of the acceleration pulse wave. When the SOM is generated using only the pulse wave data and used for the health checkup Have some problems in terms of accuracy.

  The present invention has been made to solve the above-mentioned problems, and the object of the present invention is to provide a pulse that can obtain a more accurate diagnosis result in a health diagnosis system using SOM using pulse wave information. To provide a wave diagnostic system.

The pulse wave diagnosis system according to the present invention is a system using a self-organizing map learned by using pulse wave data as an input vector. The pulse wave diagnostic system has pulse wave measuring means for sampling and measuring pulse wave data of a subject, and SOM processing means for creating a self-organizing map using the measured pulse wave data as an input vector. In particular, the pulse wave measuring means sets the sampling frequency so that the number of sampling points from the maximum amplitude point to the minimum amplitude point in one cycle of the pulse wave data becomes the first predetermined number, and samples at that sampling frequency. Get sampling data. The SOM processing means uses a second predetermined number of sampling data obtained by sampling at the sampling frequency as an input vector. Furthermore, the SOM processing means simultaneously displays the waveform data of a plurality of periods included in the pulse wave data collected from the subject within a certain period on the self-organizing map for each period .

  By using the SOM (pulse wave map) relating to the pulse wave obtained by the present invention, it is possible to classify the pulse wave waveform with higher accuracy close to the diagnosis result of the doctor, and improve the accuracy of the diagnosis result.

Embodiments of the present invention will be described below with reference to the accompanying drawings.
The pulse wave diagnosis system described below is a system that enables diagnosis of a health condition using an SOM (Self-Organizing Map) relating to pulse wave data.

1. System Configuration FIG. 1 is a diagram showing an example of the configuration of a pulse wave diagnostic system (hereinafter referred to as “system”) of the present invention. In FIG. 1, the system includes an information processing apparatus 1 that executes a diagnostic process using SOM (hereinafter referred to as “pulse wave map”) related to pulse wave data, a sensor 31 that detects a pulse wave as an electrical signal, The amplifier 33 amplifies the output signal.

  The information processing apparatus 1 includes a control unit 11 that controls the overall operation, a display unit 17 that performs screen display, an operation unit 19 that is operated by a user, and a data storage unit that stores data and programs. For example, the display unit 17 is configured by a liquid crystal display, and the operation unit 19 is a keyboard, a mouse, or the like. Further, the information processing apparatus 1 includes an AD converter 23 that converts an analog signal into a digital signal at a set sampling frequency (sampling rate), and an interface 25 for connecting to an external device or a network.

  The control unit 11 includes a SOM processing unit 13 that performs creation of a pulse wave map and diagnostic processing based on the pulse wave map, and a pulse wave measurement unit 15 that processes detection information of the pulse wave detected by the sensor 31. The control unit 11 includes a CPU and an MPU, and realizes functions of an SOM processing unit 13 and a pulse wave measurement unit 15 described later by executing a program. The program executed by the control unit 11 can be provided through a communication line or a recording medium such as a CD-ROM.

  The data storage unit 21 is a means for storing data and programs, and is composed of, for example, a hard disk. The data storage unit 21 stores a program executed by the control unit 11, measured pulse wave data, information on the SOM, and the like.

  The sensor 31 detects a volume pulse wave obtained by measuring the volume change of the capillary at the fingertip caused by the change in the internal pressure of the artery. Specifically, the sensor 31 is a transmission type photosensor, and detects the hemoglobin concentration change at the fingertip as an absorbance change, and measures the volume pulse wave. An example of such a sensor is “TL-201T” manufactured by Nihon Kohden.

2. The operation of this system having a configuration higher than the system operation will be described in detail below.

2.1 Pulse wave First, the pulse wave handled by this system is explained. FIG. 2A shows the waveform of the volume pulse wave. A differential of this volume pulse wave is a velocity pulse wave shown in FIG. 2 (b), and a derivative of this velocity pulse wave is an acceleration pulse wave shown in FIG. 2 (c). In this system, this acceleration pulse wave is used as information on the pulse wave for health examination.

FIG. 2D is a diagram illustrating an example of a waveform of the acceleration pulse wave that is cut out from the maximum point to the maximum point of the acceleration pulse wave as one cycle (T). The acceleration pulse wave is a waveform during the systole of the heart and includes five components of a wave, b wave, c wave, d wave, and e wave. The a wave is a positive wave located above the base line l, and the b wave is a negative wave located below the base line l. The c, d, and e waves change to negative or positive depending on biological conditions. Conventionally, the “blood vessel age” is given by the following equation when the distances to the vertices of the component waves are a, b, c, d, and e with reference to the base line l.

  As a conventionally known classification method of acceleration pulse wave, there is the Sano-Oyamauchi method ("Evaluation of blood circulation by acceleration pulse wave and its application", Yuji Sano et al., Labor Science, 61 (3), pp. 129-143. 1985, "Evaluation of blood circulation by acceleration pulse wave and its application (2nd report)-Trial of waveform quantification", Yuji Sano et al., Physical Fitness Research, 68, pp. 17-25, 1988, etc. ). As shown in FIG. 3, the Sano / Oyamauchi method roughly classifies acceleration pulse wave waveforms into seven types A to G. According to this classification, with the aging, the ratio of the A and B type waveforms tends to decrease, and the ratio of the D, E, F, and G type waveforms tends to increase. Thus, it is known that the waveform of the acceleration pulse wave is closely related to the health condition.

  In the present invention, the waveform of the acceleration pulse wave for one cycle refers to a waveform cut out so as to start from the maximum point of the acceleration pulse wave. The maximum point is a vertex a, and thereafter, vertex b, vertex c, Let apex d and apex e.

2.2 Grouping of pulse wave data FIG. 4 shows two typical waveforms of acceleration pulse waves for one period. FIG. 4A shows a waveform when the minimum point of the waveform is the apex b, that is, A to D waveforms shown in FIG. FIG. 4B shows the waveform when the minimum point of the waveform is the vertex d, that is, the E to G type waveforms shown in FIG. As shown in FIG. 3, the waveforms of the five types of acceleration pulse waves have the minimum values at point b (second vertex) or point d (fourth vertex), and from A type to G type. As the value changes, the minimum value shifts from the point b to the point d. Thus, it is considered that the positions (amplitude values) of the points b and d play an important role in grasping the shape of the acceleration pulse wave.

Therefore, in the present invention, attention is paid to the time from the maximum point a to the minimum point b or d, and the acceleration pulse waves are classified into two groups based on this time. Specifically, the ratio (hereinafter referred to as “rate”) of the time ΔT from the maximum point (a point) of the waveform to the minimum point (b point or d point) of the waveform with respect to one period T of the waveform is obtained. The rate (ΔT / T) is compared with a predetermined threshold, and the waveform of the acceleration pulse wave is divided into a group in which the rate is equal to or less than the threshold (low rate group) and a group in which the rate is higher than the threshold (high rate group). Classify.
Rate = ΔT / T (2)

  FIG. 5 shows acceleration pulse wave data when a set of acceleration pulse wave data collected from a large number of subjects is classified by calculating a rate according to the above equation (2) with a threshold value of “0.2”. It is the figure which showed distribution (namely, the number of subjects). As shown in the figure, on the low rate side, the waveform with the minimum point b is concentrated, and on the high rate side, the waveform with the minimum point d is concentrated and distributed. It was seen.

  In the present invention, in consideration of the distribution result, a pulse wave map (SOM) is created for each of the high rate and the low rate. As a result, accurate diagnosis based on pulse wave data becomes possible.

2.3 Normalization of Pulse Wave Data In this system, waveform data of acceleration pulse waves within one cycle is used as pulse wave data input to the pulse wave map (SOM). Each amplitude value of the waveform data (sampling data) of the acceleration pulse wave obtained at each sampling period is normalized so that the maximum amplitude value of the pulse wave is 1 and the minimum amplitude value is 0. When the waveform data indicating the waveform of the acceleration pulse wave for one period is (x1, x2, x3, ..., xi), the input vector to the pulse wave map is (x1, x2, x3, ..., xj) (j ≦ i). Note that x1 is the amplitude value of the maximum point a. In particular, at that time, the pulse wave data is converted to time so that the total number of waveform data points (value of j) for the pulse wave of one cycle input to the pulse wave map is constant and the dimension of the minimum point is always the same. Normalize on the axis. By normalizing the waveform data in this way, the waveform shape can be distinguished with high accuracy.

  Specifically, in this embodiment, in the case of low-rate pulse wave data, the value of the minimum point b is the 15th dimension in the waveform data, and in the case of high-rate pulse wave data, the minimum point Sampling periods are set so that the value of d becomes the value of the 40th dimension in the waveform data, and 50 points (50 dimensions) of waveform data acquired in the sampling period are input to the pulse wave map.

  The number of dimensions of the minimum points b and d is not limited to 15 or 40. Further, the total number of pulse wave data points (value of j) input to the SOM is not limited to 50, and is preferably set to a value including at least four vertices from point a to point d.

2.4 Pulse Wave Data Acquisition Process The pulse wave data acquisition process input to the pulse wave map will be described. This process is mainly executed by the pulse wave measurement unit 15.

  Information on the volume pulse wave of the subject detected by the sensor 31 is amplified by the amplifier 33 and converted into a digital signal by the AD converter 23 for each sampling period. The pulse wave measurement unit 15 receives the pulse wave data of the digital signal from the AD converter 23 and secondarily differentiates it to obtain acceleration pulse wave data. Such acceleration pulse wave acquisition is performed while normalizing the above-described pulse wave data.

  Specifically, in the case of pulse wave data at a low rate (rate is equal to or less than a threshold), the sampling frequency of the pulse wave data is determined so that the number of sampling points from the maximum point a to the minimum point b is 15. For this reason, pulse wave data is first sampled at a predetermined sampling frequency, and the sampling frequency is reset based on the result. For example, as shown in FIG. 6A, when the number of sampling points from the maximum point a to the minimum point b is 25 when the pulse wave data is first sampled, as shown in FIG. In addition, the sampling frequency is reset so that the number of sampling points from the maximum point a to the minimum point b is 15, and the pulse wave data is sampled again at the reset sampling frequency. Then, as the waveform data for one period, data for 50 points from the maximum point a is input to the SOM.

  On the other hand, in the case of pulse wave data at a high rate (rate is greater than the threshold), the sampling frequency of the pulse wave data is determined so that the number of sampling points from the maximum point a to the minimum point d is 40 points. For example, as shown in FIG. 7A, when the number of sampling points from the maximum point a to the minimum point d when the pulse wave data is first sampled is 50, as shown in FIG. The sampling frequency is reset so that the number of sampling points from the maximum point a to the minimum point d is 40, and the pulse wave data is sampled again at the reset sampling frequency. Then, 50 points of data from the maximum point a are employed as waveform data for one cycle for applying to the SOM.

  As described above, in the present invention, it is first determined whether the pulse wave data is a low rate or a high rate, and the pulse wave data is set so that the number of sampling points from the maximum point to the minimum point becomes a value according to the type of the rate. Set the sampling frequency.

  FIG. 8 is a flowchart of the pulse wave data acquisition process executed by the pulse wave measurement unit 15. In FIG. 8, the pulse wave measurement unit 15 first samples pulse wave data at a predetermined sampling frequency (for example, 200 Hz) (S1). Next, a rate is calculated for the pulse wave data for one cycle according to the above equation (2) (S2). The calculated rate is compared with a predetermined threshold (for example, 0.2) (S3).

  As a result of comparison, if the rate is equal to or lower than the threshold value, that is, if the rate is low, sampling of the pulse wave data is performed on the AD converter 23 so that the number of sampling points from the maximum point a to the minimum point b is 15. The frequency is reset (S4). On the other hand, if the rate is higher than the threshold, that is, if the rate is high, the sampling frequency is reset for the AD converter 23 so that the number of sampling points from the maximum point a to the minimum point d is 40 ( S5).

  Thereafter, the pulse wave data is sampled again at the reset sampling frequency (S6). And the data for 50 points from the point a among the data for one cycle is acquired as data to be applied to the SOM. The data for 50 points from the point a obtained in this way is stored as pulse wave data in the data storage unit 21 in association with information such as the subject name and data acquisition date.

［表２］
Table 2 高レートのときの学習条件

2.5 Pulse Wave Map Generation The SOM processing unit 13 creates a pulse wave map by inputting the acceleration pulse wave data obtained as described above into the SOM and learning. That is, pulse wave data is collected for each subject by the above method, the rate is calculated by the above equation (2) for each collected pulse wave data, and a plurality of collected pulse wave data groups are calculated based on the rate. Classify into low rate groups and high rate groups. For each group, a pulse wave map is created for each of the low rate group and the high rate group by learning the pulse wave map using only the pulse wave data included in each group. The two created pulse wave maps are stored in the data storage unit 21. The input data to the pulse wave map is 50-dimensional vector data indicating the waveforms of 50 acceleration pulse waves obtained by the above method. The learning conditions are as follows.
[Table 1]
Table 1 Learning conditions at low rate

[Table 2]
Table 2 Learning conditions at high rate

  An example of a pulse wave map created as described above is shown in FIG. In the figure, a map Ml is a pulse wave map created only from data included in the low rate group, and a map Mh is a pulse wave map created only from data included in the high rate group. In the low-rate map Ml, the waveform is continuously arranged from the A type to the D type, and in the high-rate map Mh, the waveform is continuously arranged from the D type to the G type. . Therefore, by combining these two maps Ml and Mh, a map in which waveforms are continuously arranged from A type to G type as a whole can be obtained. By using the pulse wave map generated in this way, diagnosis based on a continuous waveform shape is possible.

2.6 Diagnosis Processing Using Pulse Wave Map Hereinafter, diagnosis processing using a pulse wave map will be described. This process is mainly executed by the SOM processing unit 13. Buttons 51 and 53 as shown in FIG. 10 are displayed on the display unit 17 regarding the diagnostic processing based on the pulse wave map. When the pulse wave measurement start button 51 is pressed, collection of pulse wave data of a patient who receives a diagnosis is started.

  When the pulse wave measurement start button 51 is pressed, pulse wave data is acquired by the pulse wave measurement unit 15 via the sensor 31, and the acquired data is held in the data storage unit 21. The pulse wave data acquisition process is as described above. In particular, during the diagnosis process, measurement is performed for a predetermined time (for example, 20 seconds) so that pulse wave data of a plurality of cycles can be acquired.

  When the diagnosis start button 53 is pressed following the measurement of the pulse wave, the diagnosis is started based on the data acquired immediately before. The SOM processing unit 13 reads data of a subject to be diagnosed from the data storage unit 21 and calculates a rate. If the rate is low after the processing of the flowchart shown in FIG. 8, the low-rate pulse wave map Ml is read from the data storage unit 21. Alternatively, if the rate is high after the processing of the flowchart shown in FIG. 8, the high-rate pulse wave map Mh is read from the data storage unit 21. With reference to the read pulse wave map, a node closest to the subject's data is searched, and the searched node is displayed on the map so as to be visible. Thus, the subject or doctor can visually recognize the state of the pulse wave by looking at the map, and diagnosis based on the waveform becomes possible.

  When the diagnosis start button 53 is pressed, arbitrary pulse wave data may be designated from the pulse wave data acquired in the past and stored in the data storage unit 21. In addition to displaying the diagnosis results for one patient on one map, the diagnosis results of a plurality of patients can be displayed together on one map. In this case, a doctor or the like designates pulse wave data of a plurality of patients, and the SOM processing unit 13 reads the designated plurality of pulse wave data from the data storage unit 21 and performs the above-described processing.

  In the present embodiment, the diagnosis results of waveform data of a plurality of periods included in acceleration pulse wave data measured for a single patient within a predetermined time (for example, 20 seconds) are divided into waveforms for each period, and the same. Display on the map. In general, when a person is in a healthy state, the data measured within a certain period of time includes a pulse wave with a substantially constant number of cycles, but the number changes when suffering from a disease such as influenza. (For example, heart rate increases and increases.) As in this embodiment, by displaying pulse wave data collected at a certain time on a single map for each patient for each period, the number of pulse waves for one period included in the certain time can be visualized. Can be easily recognized, and changes in health status can be easily recognized. In general, when a person's physical condition is good, the temporal change, that is, variation in the shape of the pulse wave waveform is small, while when the physical condition is poor, the variation in the pulse waveform is large. Therefore, as shown in this embodiment, by displaying the diagnosis result for the pulse wave for each cycle on one map, when the physical condition is good, the diagnosis result of each pulse wave is displayed at a relatively close position on the map. When the physical condition is poor, the diagnosis result of each pulse wave is displayed at relatively dispersed positions, so that the diagnosis of good / bad physical condition is facilitated.

  11 and 12 show pulse wave maps on which diagnosis results are displayed. FIG. 11 shows a pulse wave map on which a diagnosis result with respect to pulse wave data of a patient diagnosed as being relatively healthy by a doctor is displayed. In the figure, the diagnosis result is displayed in a portion surrounded by a broken line of the low-rate pulse wave map Ml. From this pulse wave map Ml, it can be read that the patient's pulse wave waveform is of type B. The result obtained from this map Ml also matches the doctor's diagnosis result.

  FIG. 12 shows a pulse wave map on which a diagnosis result for the pulse wave data of a patient diagnosed as arteriosclerosis by a doctor is displayed. In the figure, the diagnosis result is displayed in a portion surrounded by a broken line of the high-rate pulse wave map Mh. It can be read from this pulse wave map Mh that the pulse wave waveform of the patient is G type. The result obtained from this map Mh matches the doctor's diagnosis result.

3. Summary As described above, by using a pulse wave map obtained by the system of the present invention, it is possible to classify a pulse wave waveform with high accuracy close to a doctor's diagnosis result. In particular, in the present invention, it is possible to classify pulse wave waveforms with high accuracy by normalizing sampling data input to the pulse wave map. At the time of generating the pulse wave map, the pulse wave is classified into a plurality of groups by paying attention to the minimum point of the acceleration pulse wave, and a pulse wave map is created for each classified group. This makes it possible to create an SOM that continuously changes from A type to G type. Here, the continuous change from the A type to the G type means that, for example, the B type is always arranged next to the A type, and the F type and the G type are never arranged. In addition, by standardizing the waveform data input to the SOM so that the minimum point of the waveform always has the same dimension, the waveform can be classified with high accuracy.

  The specific values such as the number of sampling points, the sampling frequency, and the learning conditions at the time of SOM creation used in the above description are merely examples, and the scope of the present invention is not limited to these values.

  In the above description, only data relating to the waveform of the pulse wave is used as input data for the pulse wave map. However, in addition to the waveform data, various data relating to the health condition of the subject (blood pressure, cholesterol, neutrality) are used as input data. Information other than the waveform of the pulse wave, such as fat and γ-GTP, may be included.

  In the above example, the pulse waves are classified based on the ratio (rate) of the time (ΔT) from the maximum point a to the minimum points b and d with respect to one period (T). Other values may be calculated in place of the rate as long as the value can be distinguished between the case where the point is and the point d.

  In the above example, pulse waves are classified into two groups, a low rate group and a high rate group. However, it is also possible to classify the pulse waves into three or more groups.

  In the above example, the sampling frequency of the AD converter 23 is reset so that the number of sampling points from the maximum amplitude point a to the minimum amplitude points b and d in the pulse wave data for one cycle becomes a predetermined number. Instead of resetting the sampling frequency, the number of sampling points from the maximum amplitude point a to the minimum amplitude points b and d can be adjusted by arithmetic processing such as thinning out or complementing the sampled data.

  In the above example, the diagnosis results for the waveforms of a plurality of cycles collected within a certain period for one patient are displayed on one map, but only the diagnosis results for the waveform data for one cycle are displayed. You may do it.

The figure which showed an example of the structure of the pulse-wave diagnostic system of this invention Diagram for explaining pulse wave Illustration explaining the classification of acceleration pulse waves by the Sano-Oyamauchi method Diagram showing typical waveform of acceleration pulse wave for one cycle The figure which shows distribution of acceleration pulse wave data when accelerating pulse wave is classified based on the rate of the present invention Diagram for explaining how to set the sampling frequency when collecting pulse wave data Diagram for explaining how to set the sampling frequency when collecting pulse wave data Flow chart of pulse wave data acquisition processing The figure which shows an example of the pulse-wave map created by this invention The figure which shows the button displayed on a display part regarding the diagnostic process by a pulse wave map in a pulse wave diagnostic system. The figure which shows the pulse-wave map by which the diagnostic result with respect to the pulse-wave data of the patient diagnosed as comparatively healthy by the doctor is displayed The figure which shows the pulse wave map by which the diagnostic result with respect to the pulse wave data of the patient diagnosed as arteriosclerosis by a doctor was displayed

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Information processing apparatus 11 Control part 13 SOM process part 15 Pulse wave measurement part 17 Display part 19 Operation part 21 Data storage part 23 AD converter 25 Interface 31 Sensor 33 Amplifier Mh Pulse wave map (for high rate)
ML pulse wave map (for low rate)

Claims (7)

A pulse wave diagnosis system using a self-organizing map learned by using pulse wave data as an input vector,
Pulse wave measuring means for sampling and measuring the pulse wave data of the subject;
SOM processing means for creating a self-organizing map using the measured pulse wave data as an input vector,
The pulse wave measuring means sets the sampling frequency so that the number of sampling points from the maximum amplitude point to the minimum amplitude point in one cycle of the pulse wave data becomes the first predetermined number, and samples at the set sampling frequency. Sampling data,
The SOM processing means, using the second predetermined number of sampling data obtained by sampling at the sampling frequency as the input vector,
The SOM processing means simultaneously displays waveform data of a plurality of periods included in pulse wave data collected from a subject within a certain period on a self-organizing map for each period.
A pulse wave diagnostic system characterized by that.   2. The SOM processing means classifies pulse wave data into a plurality of groups based on a position of a minimum point in one cycle of a pulse wave, and creates a self-organizing map for each group. The described pulse wave diagnostic system.   The SOM processing means includes a low-rate group in which the ratio of the time (ΔT) from the maximum point to the minimum point of the amplitude in one period with respect to one period (T) of the pulse wave data is equal to or less than a threshold value. 3. The pulse wave diagnosis system according to claim 2, wherein the rate is classified into a high rate group larger than a threshold, and a self-organizing map is created separately for each of the low rate group and the high rate group.   The pulse wave diagnosis system according to claim 2, wherein the first predetermined number is set for each group.   The pulse wave diagnosis system according to claim 1, wherein the SOM processing means displays a diagnosis result for the pulse wave data of the subject to be diagnosed on the generated self-organizing map. The pulse wave diagnosis system according to any one of claims 1 to 5 , wherein the pulse wave data is acceleration pulse wave data obtained by differentiating a volume pulse wave twice. The second predetermined number, according to claim 6, wherein said input vector is the set from the maximum amplitude point of the data and the maximum amplitude point to include data of the first to third vertices, it is characterized by Pulse wave diagnostic system.

Images