阅读说明：本技术 一种基于大数据的输液瓶换液顺序推荐方法及系统 (Infusion bottle liquid change sequence recommendation method and system based on big data ) 是由 张来娣 于 2021-07-29 设计创作，主要内容包括：本发明涉及一种基于大数据的输液瓶换液顺序推荐方法及系统,该方法采用神经网络中的DNN网络对实时采集的吊瓶图像进行语义分割,准确分割出吊瓶的原始图像,然后对吊瓶的原始图像中能表征输液进度的液位面分界线进行检测,利用设定时间间隔内液位面分界面的下降高度,计算出吊瓶药液输完所需的剩余时间,按照各吊瓶药液输完所需的剩余时间的长短进行排序,相当于根据各床病人的真实输液进度进行换液的紧急程度排序,使护士能够及时掌握各病床输液进度,从而有效提升换液效率。并且,无需病人持续关注输液进度,降低人力资源消耗和因呼叫不及时造成的危险情况的发生。(The invention relates to a big data-based infusion bottle liquid changing sequence recommendation method and a big data-based infusion bottle liquid changing sequence recommendation system, wherein a DNN network in a neural network is adopted to carry out semantic segmentation on infusion bottle images collected in real time, original images of infusion bottles are accurately segmented, then a liquid level surface boundary capable of representing the infusion progress in the original images of the infusion bottles is detected, the remaining time required by the infusion of infusion bottle liquid medicine is calculated by utilizing the descending height of the liquid level surface boundary in a set time interval, and the sorting is carried out according to the remaining time required by the infusion of each infusion bottle liquid medicine, namely the sorting of the liquid changing emergency degree is carried out according to the real infusion progress of each bed patient, so that a nurse can timely master the infusion progress of each sickbed, and the liquid changing efficiency is effectively improved. In addition, the patient does not need to continuously pay attention to the infusion progress, so that the human resource consumption and the dangerous condition caused by untimely calling are reduced.)

1. A big data-based infusion bottle liquid change sequence recommendation method is characterized by comprising the following steps:

step S1, acquiring images containing infusion bottles in each ward in real time, and processing the acquired images containing infusion bottles in each ward by using a neural network to obtain original images of each infusion bottle;

step S2, detecting the boundary of the liquid level surface in the original image of each hanging bottle; the original image of the transfusion bottle comprises a liquid area and an empty bottle area, and the liquid level boundary line is a liquid level boundary line for distinguishing the liquid area from the empty bottle area;

step S3, calculating the residual time required by the transfusion bottle after the transfusion of the liquid medicine by utilizing the descending height of the interface of the liquid level surface within the set time interval; after the remaining time required by the infusion of the liquid medicine in the infusion bottles of all the sickbeds is obtained by calculation according to the method, the sickbeds are sorted according to the length of the remaining time, and the sorting result is recommended as the priority processing sequence of liquid change.

2. The method for recommending a fluid replacement order of an infusion bottle based on big data as claimed in claim 1, wherein in step S2, the step of detecting the boundary of the fluid level surface in the original image of the infusion bottle is as follows:

1) converting the original image of the infusion bottle into a gray level image, and determining the gray level value of each pixel point;

2) traversing each pixel point in the gray level image, and determining a gray level adjustment value of the current pixel point according to a gray level difference value and an azimuth angle between the current pixel point and the adjacent n (n is more than or equal to 2) pixel points in the previous row; the calculation formula of the gray adjustment value is as follows:

wherein, Δ p_iTheta is the angle between the adjacent pixel and the current pixel, and deltag is the gray scale adjustment value_jIs a neighboring pixel andthe gray difference between the previous pixels, and n is the number of the adjacent pixels of the current pixel point;

3) superposing the gray adjustment value to the gray value of the current pixel point to update the gray value of the current pixel point;

4) obtaining a gray image after gray adjustment, calculating the gray value gradient of the pixel point and the adjacent pixel point of the previous line in the pixel points of which the gray values are updated, and determining a boundary line to be selected; and finally determining the liquid level boundary in the image by combining the known liquid level boundary width and the horizontal direction.

3. The big-data-based recommendation method for fluid change sequences of infusion bottles according to claim 2, wherein in the step 3), the specific step of updating the gray value of the current pixel point comprises the following steps:

when the gray difference value delta g between the current pixel point and the adjacent pixel point_jWhen the gray values are all larger than zero, the formula for updating the gray values is as follows:

p_i,new＝p_i-Δp_i

wherein p is_i,newIs the gray value, p, of the current pixel point updated after superposition_iIs the gray value, Δ p, of the current pixel point_iIs a gray scale adjustment value;

when the gray difference value delta g between the current pixel point and the adjacent pixel point_jWhen the gray values are all less than zero, the formula for updating the gray values is as follows:

p_i,new＝p_i+Δp_i

wherein p is_i,newIs the gray value, p, of the current pixel point updated after superposition_iIs the gray value, Δ p, of the current pixel point_iIs a gray scale adjustment value.

4. The big-data-based recommendation method for fluid change orders of infusion bottles of claim 2, wherein the step 4) of determining the boundary of the fluid level in the image comprises the following steps:

(1) calculating the lengths of the mutually communicated adjusted pixel points according to a connected domain analysis method, judging a set in which the length of the connected domain is less than the width of the transfusion bottle as a noise point, and filtering the noise point, wherein the determination method of the width of the transfusion bottle comprises the following steps: performing connected domain analysis on semantic segmentation images of the infusion bottles, and selecting the longest side length communicated in the horizontal direction as the width of the infusion bottles;

(2) continuously screening the connected domain set subjected to the previous screening based on the horizontal direction, traversing each pixel point, calculating the slope direction between the pixel point and the adjacent pixel point, and filtering the pixel point if the pixel point is not in the horizontal direction;

(3) screening according to the gray value gradient information, screening pixel points with the same gray value gradient, and taking the straight line where the screened pixel points are located as the final liquid level boundary; the gray value gradient is the gray difference value of two lines of pixel points corresponding to the upper line and the lower line and is expressed as p_i,j-p_i+1,jI and i +1 denote rows and j denotes columns;

if more than one connected domain set meets the requirement, the mean square error of the gray value gradient between two adjacent upper and lower lines of pixel points is calculated, and the straight line where the pixel point corresponding to the minimum mean square error is located is taken as the final boundary of the liquid level surface.

5. The big data based recommendation method for fluid change order of infusion bottle as claimed in claim 1, wherein in step S3, the step of calculating the remaining time required for infusion of the infusion bottle medical fluid is as follows:

according to the period T of the set time interval and the volume of the transfusion bottle is V_iCalculating the current infusion rate of the infusion bottleThen combining the volume V of the non-liquid medicine area corresponding to the boundary line of the current liquid level surface₀Calculating the time required for the residual liquid medicine to be infused completely

6. A big data-based infusion bottle liquid change sequence recommendation method is characterized by comprising the following steps:

step S1, acquiring images containing infusion bottles in each ward in real time, and processing the acquired images containing infusion bottles in each ward by using a neural network to obtain original images of each infusion bottle;

step S2, detecting the boundary of the liquid level surface in the original image of each hanging bottle; the original image of the transfusion bottle comprises a liquid area and an empty bottle area, and the liquid level boundary line is a liquid level boundary line for distinguishing the liquid area from the empty bottle area;

step S3, according to the length of the boundary of the liquid level surface of the current transfusion bottle, judging whether the current transfusion progress enters the end stage,

the method for judging the entering of the end stage comprises the following steps:

comparing the length L1 of the boundary line of the current liquid level surface of the infusion bottle with the length L0 of the boundary line of the liquid level surface of the last period, and judging that the infusion process enters the end stage when the length of the boundary line of the current liquid level surface is lower than the length of the boundary line of the liquid level surface of the last period for n (n is more than or equal to 2) periods;

step S4, if the end stage is judged to be entered, the remaining time required for the transfusion bottle liquid medicine to be completely transfused is calculated by utilizing the descending height of the interface of the liquid level surface within the set time interval; after the remaining time required by the infusion of the liquid medicine in the infusion bottles of all the sickbeds is obtained by calculation according to the method, the sickbeds are sorted according to the length of the remaining time, and the sorting result is recommended as the priority processing sequence of liquid change.

7. The big-data-based recommendation method for fluid change sequences of infusion bottles of claim 6, wherein the priority ordering and the infusion rate adjustment are performed on each sickbed which is in the same ward and is at the end stage of the infusion progress, and the specific steps are as follows:

1) screening the object with the highest priority level to be changed, wherein the screening method comprises the following steps:

obtaining the shortest remaining infusion time according to the remaining infusion time of the infusion bottle corresponding to each sickbed, and taking the shortest infusion time as a reference, classifying the remaining infusion time in the same ward and the infusion device with the difference within the set time to the same highest priority level, and setting the infusion device as an object to be changed;

2) in the object to be changed, except the infusion bottle with the shortest remaining infusion time, a sound reminding mode and/or a text display reminding mode is adopted, and infusion rate adjustment suggestions of the infusion bottle are given to the sickbed corresponding to the object to be changed, and the infusion rate is slowed down.

8. The big-data-based recommendation method for fluid change orders of infusion bottles of claim 7, wherein the longest t in the remaining infusion time of the object to be changed is_lTaking the rate adjustment as a reference, wherein the adjusted infusion rate is as follows:

wherein v is_k,newFor adjusted infusion rate, V_kRepresenting the volume of a transfusion bottle corresponding to the kth object to be changed in the same ward, V₀Is the volume of the non-liquid medicine area.

9. The method for recommending a fluid change sequence of an infusion bottle based on big data as claimed in claim 7, wherein if the shortest infusion time of the fluid change objects in a plurality of wards is the same, calculating the priority characteristic value P of the fluid change sequence of each ward, the calculation expression is:

σ_kthe mean square error of the remaining time of the objects to be changed in the same ward is shown, N is the number N of sickbeds with infusion objects in the ward and the total number of the objects to be changed in the ward, and N is more than or equal to 1 and less than or equal to N;

sequencing according to the obtained priority characteristic value P in a liquid changing sequence with the same remaining shortest time and objects to be changed in each ward; except the ward corresponding to the highest priority characteristic value P, all other wards are sequentially slowed down in infusion rate according to the priority characteristic value P from high to low.

10. An infusion bottle liquid change sequence recommendation system based on big data is characterized by comprising a controller, a display and cameras arranged in each ward, wherein the controller is connected with each camera, the controller is connected with the display, and the controller is used for acquiring images shot by each camera, processing the images containing infusion bottles in each ward according to the method of any one of claims 1 to 9, generating a sequencing result of liquid change objects and displaying the sequencing recommendation result in the display.

Technical Field

The invention relates to the technical field of artificial intelligence and big data, in particular to a method and a system for recommending a liquid change sequence of an infusion bottle based on big data.

Background

At present, the patients are often required to be treated by infusion during the hospitalization process, and meanwhile, the patients are required to be equipped with nurses to carry out related auxiliary work. However, the number of nurses configured in each nurse station of the hospital department is limited, and often, only two to three nurses are provided, when there are more patients, especially when the infusion of a plurality of patients is dense, the workload of the nurses is multiplied, because there are more patients needing to change the infusion, the nurses can only process according to the calling condition of the patients, however, some patients do not call the nurses in time when the infusion bottles need to be changed due to negligence, and the nurses may be busy with the nursing work of other patients, so that the infusion bottles of some patients cannot be changed in time. In addition, some patients may call a nurse only when the infusion is about to be finished, and the nurse is busy at the moment, so that the dangerous situations such as needle leakage, blood backflow and the like of the patients are easily caused.

On the other hand, many patients are often called about five minutes in advance when the infusion is about to end in order to prevent the nurse from missing himself or having a slow response time in a busy period, which increases the urgency of the nurse and affects the response time of the patient in a more urgent situation to other beds. In addition, because the calling time of each sickbed is different and the positions of the sickbeds are different, the nurse can easily change the liquid back and forth, which not only causes the physical waste and overdraft of the nurse, but also increases the response time of other liquid changing work, and reduces the liquid changing efficiency.

Disclosure of Invention

The invention aims to provide a method and a system for recommending a liquid changing sequence of an infusion bottle based on big data, which are used for solving the problems that when a large number of patients to be changed have liquid, the infusion bottle is not changed timely due to missed calls and late calls of the patients, and the liquid changing efficiency is low due to unreasonable liquid changing of nurses when the number of patients to be changed has liquid.

Therefore, the adopted technical scheme is as follows:

in a first aspect, the invention provides a big data-based method for recommending a liquid change sequence of an infusion bottle, which comprises the following steps:

step S1, acquiring images containing infusion bottles in each ward in real time, and processing the acquired images containing infusion bottles in each ward by using a neural network to obtain original images of each infusion bottle;

step S2, detecting the boundary of the liquid level surface in the original image of each hanging bottle; the original image of the transfusion bottle comprises a liquid area and an empty bottle area, and the liquid level boundary line is a liquid level boundary line for distinguishing the liquid area from the empty bottle area;

step S3, calculating the residual time required by the transfusion bottle after the transfusion of the liquid medicine by utilizing the descending height of the interface of the liquid level surface within the set time interval; after the remaining time required by the infusion of the liquid medicine in the infusion bottles of all the sickbeds is obtained by calculation according to the method, the sickbeds are sorted according to the length of the remaining time, and the sorting result is recommended as the priority processing sequence of liquid change.

Preferably, the neural network is a DNN network, and the processing of acquiring images containing infusion bottles in each ward by using the neural network specifically includes:

1) the collected images containing the infusion device are used as a training data set, the data set is labeled, and the labeled labels are as follows: the infusion bottle in the infusion device is marked as 1, and the rest non-infusion bottle part is marked as 0; after the labeling is finished, randomly selecting a training set of the data set, and taking the rest data set as a verification set;

2) inputting image data and label data into a DNN network, wherein an encoder of the network is used for extracting image characteristics and converting the number of channels into the number of categories; the decoder of the network is used for transforming the height and width of the characteristic image into the size of the input image, thereby outputting the category of each pixel in the characteristic image;

3) training by using a loss function to obtain a transfusion bottle semantic segmentation graph;

4) the obtained semantic segmentation graph of the infusion bottle is compared with the original graphImage of a personAnd multiplying the corresponding pixel values to obtain an original image of the transfusion bottle.

Preferably, in step S2, the step of detecting the boundary of the liquid level surface in the original image of the hanging bottle is as follows:

1) and converting the original image of the infusion bottle into a gray level image, and determining the gray level value of each pixel point.

2) Traversing each pixel point in the gray level image, and determining a gray level adjustment value of the current pixel point according to a gray level difference value and an azimuth angle between the current pixel point and the adjacent n (n is more than or equal to 2) pixel points in the previous row; the calculation formula of the gray adjustment value is as follows:

wherein, Δ p_iTheta is the angle between the adjacent pixel and the current pixel, and deltag is the gray scale adjustment value_jThe gray difference value between the adjacent pixel and the current pixel is obtained, and n is the number of the adjacent pixels of the current pixel;

3) superposing the gray adjustment value to the gray value of the current pixel point to update the gray value of the current pixel point;

4) obtaining a gray image after gray adjustment, calculating the gray value gradient of the pixel point and the adjacent pixel point of the previous line in the pixel points of which the gray values are updated, and determining a boundary line to be selected; and finally determining the liquid level boundary in the image by combining the known liquid level boundary width and the horizontal direction.

Preferably, in step 3), the specific step of updating the gray value of the current pixel point includes:

when the gray difference value delta g between the current pixel point and the adjacent pixel point_jWhen the gray values are all larger than zero, the formula for updating the gray values is as follows:

p_i,new＝p_i-Δp_i

wherein p is_i,newIs the gray value, p, of the current pixel point updated after superposition_iIs the gray value, Δ p, of the current pixel point_iIs a gray scale adjustment value;

when the gray difference value delta g between the current pixel point and the adjacent pixel point_jWhen the gray values are all less than zero, the formula for updating the gray values is as follows:

p_i,new＝p_i+Δp_i

wherein p is_i,newIs the gray value, p, of the current pixel point updated after superposition_iIs the gray value, Δ p, of the current pixel point_iIs a gray scale adjustment value.

Preferably, in step 4), determining the boundary of the liquid level in the image includes:

(1) calculating the lengths of the mutually communicated adjusted pixel points according to a connected domain analysis method, judging a set in which the length of the connected domain is less than the width of the transfusion bottle as a noise point, and filtering the noise point, wherein the determination method of the width of the transfusion bottle comprises the following steps: performing connected domain analysis on semantic segmentation images of the infusion bottles, and selecting the longest side length communicated in the horizontal direction as the width of the infusion bottles;

(2) continuously screening the connected domain set subjected to the previous screening based on the horizontal direction, traversing each pixel point, calculating the slope direction between the pixel point and the adjacent pixel point, and filtering the pixel point if the pixel point is not in the horizontal direction;

(3) screening according to the gray value gradient information, screening pixel points with the same gray value gradient, and taking the straight line where the screened pixel points are located as the final liquid level boundary; the gray value gradient is the gray difference value of two lines of pixel points corresponding to the upper line and the lower line and is expressed as p_i,j-p_i+1,jI and i +1 denote rows and j denotes columns;

if more than one connected domain set meets the requirement, the mean square error of the gray value gradient between two adjacent upper and lower lines of pixel points is calculated, and the straight line where the pixel point corresponding to the minimum mean square error is located is taken as the final boundary of the liquid level surface.

Preferably, in step S3, the step of calculating the remaining time required for the infusion of the drug solution into the vial is as follows:

according to the period T of the set time interval and the volume of the transfusion bottle is V_iCalculating the current infusion rate of the infusion bottleThen combining the volume V of the non-liquid medicine area corresponding to the boundary line of the current liquid level surface₀Calculating the time required for the residual liquid medicine to be infused completely

The invention has the following beneficial effects:

the method comprises the steps of performing semantic segmentation on infusion bottle images acquired in real time by adopting a DNN (digital network) in a neural network, accurately segmenting original images of infusion bottles, detecting a liquid level boundary capable of representing infusion progress in the original images of the infusion bottles, calculating residual time required by infusion completion of infusion of the infusion bottle liquid medicine by utilizing the descending height of the liquid level boundary within a set time interval, sequencing according to the residual time required by infusion completion of the infusion bottle liquid medicine, equivalently sequencing according to the emergency degree of liquid change according to the real infusion progress of patients in each bed, and enabling nurses to timely master infusion progress of each sickbed, so that the liquid change efficiency is effectively improved. In addition, the patient does not need to continuously pay attention to the infusion progress, so that the human resource consumption and the dangerous condition caused by untimely calling are reduced.

In a second aspect, the invention provides another infusion bottle liquid change sequence recommendation method based on big data, which comprises the following steps:

step S1, acquiring images containing infusion bottles in each ward in real time, and processing the acquired images containing infusion bottles in each ward by using a neural network to obtain original images of each infusion bottle;

step S2, detecting the boundary of the liquid level surface in the original image of each hanging bottle; the original image of the transfusion bottle comprises a liquid area and an empty bottle area, and the liquid level boundary line is a liquid level boundary line for distinguishing the liquid area from the empty bottle area;

step S3, according to the length of the boundary of the liquid level surface of the current transfusion bottle, judging whether the current transfusion progress enters the end stage,

the method for judging the entering of the end stage comprises the following steps:

comparing the length L1 of the boundary line of the current liquid level surface of the infusion bottle with the length L0 of the boundary line of the liquid level surface of the last period, and judging that the infusion process enters the end stage when the length of the boundary line of the current liquid level surface is lower than the length of the boundary line of the liquid level surface of the last period for n (n is more than or equal to 2) periods;

step S4, if the end stage is judged to be entered, the remaining time required for the transfusion bottle liquid medicine to be completely transfused is calculated by utilizing the descending height of the interface of the liquid level surface within the set time interval; after the remaining time required by the infusion of the liquid medicine in the infusion bottles of all the sickbeds is obtained by calculation according to the method, the sickbeds are sorted according to the length of the remaining time, and the sorting result is recommended as the priority processing sequence of liquid change.

Preferably, the priority ordering and infusion rate adjustment are performed on each sickbed which is in the same ward and is at the end stage of the infusion progress, and the specific steps are as follows:

1) screening the object with the highest priority level to be changed, wherein the screening method comprises the following steps:

obtaining the shortest remaining infusion time according to the remaining infusion time of the infusion bottle corresponding to each sickbed, and taking the shortest infusion time as a reference, classifying the remaining infusion time in the same ward and the infusion device with the difference within the set time to the same highest priority level, and setting the infusion device as an object to be changed;

2) in the object to be changed, except the infusion bottle with the shortest remaining infusion time, a sound reminding mode and/or a text display reminding mode is adopted, and infusion rate adjustment suggestions of the infusion bottle are given to the sickbed corresponding to the object to be changed, and the infusion rate is slowed down.

Preferably, the longest t in the remaining transfusion time in the object to be replaced is_lTaking the rate adjustment as a reference, wherein the adjusted infusion rate is as follows:

wherein v is_k,newFor adjusted infusion rate, V_kRepresenting the volume of a transfusion bottle corresponding to the kth object to be changed in the same ward, V₀Is the volume of the non-liquid medicine area.

Preferably, if the shortest infusion time of the objects to be changed in a plurality of wards is the same, calculating the priority characteristic value P of the change order of each ward, wherein the calculation expression is as follows:

σ_kthe mean square error of the remaining time of the objects to be changed in the same ward is shown, N is the number N of sickbeds with infusion objects in the ward and the total number of the objects to be changed in the ward, and N is more than or equal to 1 and less than or equal to N;

sequencing according to the obtained priority characteristic value P in a liquid changing sequence with the same remaining shortest time and objects to be changed in each ward; except the ward corresponding to the highest priority characteristic value P, all other wards are sequentially slowed down in infusion rate according to the priority characteristic value P from high to low.

The invention has the following beneficial effects:

the infusion bottle image collected in real time in the neural network is processed to obtain an original image of the infusion bottle, then a liquid level boundary capable of representing the infusion progress in the original image of the infusion bottle is detected, and when the liquid level boundary enters the end stage of infusion, the emergency degree of liquid change is sequenced according to the real infusion progress of each bed patient, so that a nurse can master the infusion progress of each bed in time, and the liquid change efficiency is effectively improved; but also can ensure that the patient can be fully transfused before the nurse changes the liquid, thereby improving the transfusion quantity. In addition, the patient does not need to continuously pay attention to the infusion progress, so that the human resource consumption and the dangerous condition caused by untimely calling are reduced.

In a third aspect, the invention provides another infusion bottle liquid change sequence recommendation system based on big data, which comprises:

the controller is connected with the cameras and the display, and is used for acquiring images photographed by the cameras, processing the images containing the infusion bottles in the wards according to the method mentioned in the first aspect or the second aspect, generating a sorting result of the liquid changing objects, and displaying the sorting recommendation result in the display.

The invention has the following beneficial effects:

the method of the first and second aspects above can be implemented by hardware devices of a controller, a display, and a camera disposed in each ward, and the liquid exchange efficiency is improved.

Drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions and advantages of the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and other drawings can be obtained by those skilled in the art without creative efforts.

Fig. 1 is a flowchart of a method for recommending a liquid change sequence of an infusion bottle according to embodiment 1 of the present invention;

FIG. 2-1 shows p in example 1 of the present invention_iAnd the gray difference value deltag between three adjacent pixels_jSchematic diagrams each being greater than zero;

FIG. 2-2 shows p in example 1 of the present invention_iAnd the gray difference value deltag between three adjacent pixels_jAll are less than zero.

Detailed Description

The embodiments provided by the invention are specifically described below with reference to the accompanying drawings.

Example 1:

the invention discloses a big data-based infusion bottle liquid changing sequence recommendation method, which mainly aims to realize effective sequencing of liquid changing operations of infusion bottles of hospital beds, improve liquid changing efficiency and reduce dangerous accidents, and the invention has the following conception: acquiring images containing infusion bottles (infusion bottles) in each ward in real time, and processing the images containing the infusion bottles in each ward by using a neural network to obtain original images of the infusion bottles; detecting the boundary of the liquid level surface in the original image of each hanging bottle, and calculating the residual time required by the liquid medicine of the hanging bottle after the liquid medicine is completely infused by utilizing the descending height of the boundary of the liquid level surface within a set time interval; after the remaining time required by the infusion of the liquid medicine in the infusion bottles of a plurality of sickbeds is obtained by calculation according to the method, the sickbeds are sorted according to the length of the remaining time, and the sorting result is recommended as the priority treatment sequence of liquid change.

Specifically, as shown in fig. 1, the method includes the following specific steps:

and step S1, acquiring images containing transfusion bottles (infusion bottles) in each ward in real time, and processing the acquired images containing the transfusion bottles in each ward by using a neural network to acquire original images of the transfusion bottles.

In the step, when the image is collected, the RGB camera is used to be placed above the ward for taking a picture under the condition of normal and sufficient illumination, wherein the picture taking range of the camera is required to cover all sickbed areas and corresponding infusion devices in the ward.

In this step, the adopted neural network is a DNN network, semantic segmentation is adopted for sensing the infusion bottle in the photographed image, the DNN network is an Encoder-Decoder structure (Encoder-Decoder), and the specific training content is as follows:

1) the collected images containing the infusion device are used as a training data set, the data set is labeled, and the labeled labels are as follows: the infusion bottle in the infusion set is marked with 1 and the other (non-infusion part) is marked with 0. After the labeling is completed, 80% of the data set is randomly selected as a training set, and the remaining 20% is selected as a validation set.

2) Inputting image data and label data into a DNN network, wherein an Encoder of the network is used for extracting image characteristics and converting the number of channels into the number of categories; a Decoder of the network is used to transform the height and width of the feature map into the size of the input image, thereby outputting the class of each pixel in the feature image.

3) And (4) training by using a Loss function (cross entropy Loss function), and obtaining a transfusion bottle semantic segmentation graph by using the Loss function.

4) And multiplying the obtained transfusion bottle semantic segmentation graph with the pixel value corresponding to the original image so as to obtain the original image of the transfusion bottle.

In step S2, the boundary of the liquid level in the original image of each hanging bottle is detected.

In the obtained original image of the infusion bottle, the infusion bottle area comprises a liquid area and an empty bottle area, and although the attributes of the two areas are different, the brightness difference of the two areas is small under the same illumination intensity, and the liquid level boundary line for distinguishing the liquid area from the empty bottle area is not obvious.

In order to obtain a clear and effective boundary of the liquid level surface, the original image of the infusion bottle obtained in step S1 needs to be processed, and the specific processing procedures are as follows:

1) and converting the original image of the infusion bottle into a gray level image, and determining the gray level value of each pixel point.

2) And traversing each pixel point in the gray level image, and determining the gray level adjustment value of the current pixel point according to the gray level difference value and the azimuth angle between the current pixel point and the three adjacent pixel points in the previous row.

By any pixel point p_iFor the example of the center pixel point, at p_iWithin the eight neighborhood pixels, p is counted respectively_iThe gray level difference Δ g between the three adjacent pixels above it_jWherein j is 1,2,3, which respectively represent three adjacent pixel points and p from left to right in turn_iThe gray scale difference of (a); for the determination of the azimuth angle, the invention passes each adjacent pixel and p_iThe included angle between the two points is used as an azimuth angle theta to be determined, and the azimuth angle is used as a weight reference to perform the calibration on the current pixel point p_iIs adjusted, wherein the adjustment value Δ p is adjusted_iThe specific calculation formula is as follows:

wherein, Δ p_iTheta is the angle between the adjacent pixel and the current pixel, and deltag is the gray scale adjustment value_jThe gray difference between the adjacent pixel and the current pixel is obtained, and n is the number of the adjacent pixels of the current pixel.

In this step, since the liquid level is horizontal, the gray scale of each pixel is adjusted according to the gradient in the vertical direction in the eight-neighborhood of the pixel. Thus, if m (m >8) neighborhood pixels are used as the basic pixel relationship, then n is an integer greater than 3, for example, n is 4 or 5.

In this step, the sine value of the azimuth angle θ is used as a weight to represent the pixel adjustment value Δ p of the gray scale difference between a certain adjacent pixel and the current pixel_iFor example, in three adjacent pixel points from left to right, the azimuth angle between the pixel point on the right and the current pixel point is 45 degrees, and the weight value is 0.5; the azimuth angle between the left pixel point and the current pixel point is 135 degrees, and the weight value is-0.5; the azimuth angle between the middle pixel point and the current pixel point is 90 degrees, and the weight value is 1. Adjusting the value of Δ p to the pixel due to consideration of the weights_iThe contribution degree of the azimuth angle is only considered, and the positive and negative of the numerical value are not considered, so that the absolute value is adopted in the formula to solve the problem of removing the sign of the sine value of the azimuth angle and the problem of removing the positive and negative signs of the gray difference value.

In this step, when obtaining the pixel adjustment value, the gray scale difference between the pixel and the adjacent pixel is considered, that is, only the pixel with the gray scale difference is subjected to gray scale adjustment to increase the gray scale difference between the pixels, if the three gray scale differences Δ g are calculated_jAnd all the pixels are zero, which indicates that the gray level of the pixel point is not different from that of the adjacent pixel points, and the current pixel does not need to be adjusted.

3) Superposing the gray adjustment value to the gray value of the current pixel point to update the gray value of the current pixel point, when p is_iAnd the gray difference value deltag between three adjacent pixels_jWhen the gray values of the three adjacent pixel points are all larger than the gray value of the current pixel point, and therefore the superposition formula is as follows:

p_i,new＝p_i-Δp_i

wherein p is_i,newAnd the gray value of the current pixel point is updated after superposition.

As shown in fig. 2-1, the gray value of the current pixel point is 100, and the gray values of the three adjacent pixel points in the previous row are 120, and 120, respectively, then the adjustment value obtained by calculation according to the formula in step 2) is 40, since the pixel point p is_iThe gray values of the pixel points are all smaller than the gray values of three adjacent pixel points, so as to increase the gray difference between the current pixel point and the adjacent pixel points, therebyWhen the gray adjustment value is superimposed, the gray adjustment value needs to be subtracted from the current gray value, and the updated gray value is 60.

When p is_iAnd the gray difference value deltag between three adjacent pixels_jWhen the gray values of the three adjacent pixel points are all less than zero, the gray values of the three adjacent pixel points are all less than the gray value of the current pixel point, so that the superposition formula is as follows:

p_i,new＝p_i+Δp_i

wherein p is_i,newAnd the gray value of the current pixel point is updated after superposition.

Through the steps, errors caused by noise pixel points can be effectively avoided, and due to the fact that the difference between pixel values in the liquid area and pixel values in the non-liquid area are only larger near the boundary, the more obvious liquid level boundary can be effectively obtained through the adjustment.

As shown in fig. 2-2, the gray value of the current pixel point is 100, and the gray values of the three adjacent pixel points in the previous row are respectively 80, and 80, then the adjustment value obtained by calculation according to the formula in step 2) is 40, since the pixel point p is_iThe gray values of the pixels are all larger than the gray values of the three adjacent pixels, in order to increase the gray difference between the current pixel and the adjacent pixels, therefore, when the gray adjustment value is superimposed, the gray adjustment value needs to be added on the basis of the current gray value, and the updated gray value is 140.

4) Obtaining a gray image after gray adjustment, calculating gray value gradient (namely gray difference) of the pixel point and the adjacent pixel point of the previous line in the pixel points of which the gray values are updated, and determining a boundary line to be selected; then, combining the known liquid level boundary width and the horizontal direction, finally determining the liquid level boundary in the image.

Step S3, calculating the remaining time required for the transfusion bottle to finish the transfusion of the liquid medicine by utilizing the descending height of the interface of the liquid level surface within the set time interval; after the remaining time required by the infusion of the liquid medicine in the infusion bottles of a plurality of sickbeds is obtained by calculation according to the method, the sickbeds are sorted according to the length of the remaining time, and the sorting result is recommended as the priority treatment sequence of liquid change.

Specifically, after the boundary of the liquid level surface is obtained, the infusion rate of the infusion bottle is calculated in a period T with one minute as a set time interval. Because the volume of the transfusion bottle has various specifications and the falling height of the boundary of the liquid level surface is only relied on, the falling volume of the liquid medicine in the period can not be judged, therefore, the volume of the transfusion bottle is set as V in the invention_iCalculating h according to the falling height of the boundary of the liquid level surface in the period T_TCalculate the corresponding drop volume as V_TThe volume of the non-liquid medicine region corresponding to the boundary of the current liquid level surface is V₀Then the infusion rate of the current infusion bottle can be calculatedAnd calculating the time required for the residual liquid medicine to be completely infused

In this embodiment, two ways may be adopted to obtain:

the first method is as follows: the formula for calculating the drop volume of the boundary of the liquid level surface is:

V_T＝*h_i

in the formula, V_TThe falling volume of the boundary of the liquid level surface, s is the sectional area of the transfusion bottle, h_iThe falling height of the boundary of the liquid level surface in the period T.

The second method comprises the following steps: the formula for calculating the drop volume of the boundary of the liquid level surface is:

in the formula, V_TIs the falling volume of the boundary line of the liquid level, V_iIs the volume of the transfusion bottle, h_TIs the height of the transfusion bottle, h_iThe falling height of the boundary of the liquid level surface in the period T.

Example 2:

the embodiment provides a method for recommending a liquid change sequence of an infusion bottle based on big data, which is different from the method in the embodiment 1 in that in step S2, after the gray value of the pixel point is updated, there are more noise points in the gray image, and therefore before determining the liquid level boundary, the gray image needs to be continuously processed to filter the noise points to obtain a more effective liquid level boundary.

Based on the above consideration, the specific process of continuing to process the grayscale image (in sub-step 4 in step S2) is:

(1) firstly, calculating the lengths of the mutually communicated adjusted pixel points according to a connected domain analysis method (which is the prior art), judging a set of which the length of the connected domain is less than the width of the infusion bottle as a noise point which cannot be a liquid level boundary, and filtering the noise point, wherein the width of the infusion bottle can be subjected to the connected domain analysis by performing the same semantic segmentation image on the infusion bottle, and the longest side length communicated in the horizontal direction is selected.

(2) Because the liquid level surface is basically horizontal, the connected domain set which is screened in the previous step is continuously screened based on the horizontal direction, each pixel point is traversed, the slope direction between each pixel point and the adjacent pixel point is calculated, and filtering is carried out if the slope direction is not the horizontal direction.

(3) After the influence of illumination is removed, the gray value gradient of each pixel point at the boundary of the liquid level surface (i.e. the gray difference value corresponding to two upper and lower rows of pixel points, which is expressed as p)_i,j-p_i+1,j) And finally, screening according to the gray value gradient information, screening pixel points with the same gray value gradient, namely, after the screening is carried out again through the process, if more than one connected domain set meets the requirement at the moment, calculating the mean square error of the gray value gradient between the upper and lower lines of adjacent pixel points, and taking the straight line of the pixel point corresponding to the minimum mean square error as the final liquid level boundary.

Through the above steps, in the embodiment 1, when there are many noise points in the original image of the hanging bottle in step S2, the noise can be well removed, and the false detection of the liquid level boundary caused by the fact that the noise is not filtered out is avoided.

Example 3:

the embodiment provides a method for recommending a liquid change sequence of an infusion bottle based on big data, which is different from the method in the embodiment 1 in that in step S3, in order to input more liquid in the infusion bottle into a patient as much as possible and avoid changing the infusion bottle for the patient prematurely, the priority processing sequence of recommended liquid change needs to be controlled, and the recommended sequence is optimized.

Before sorting according to the length of the remaining time, the method further comprises the following steps:

judging whether the current infusion progress enters a tail stage or not according to the length of a boundary of a liquid level surface of the current infusion bottle, if so, calculating the remaining duration of the infusion progress of all the infusion bottles entering the tail stage, and only sequencing the infusion progress of all the infusion bottles entering the tail stage, wherein the shorter the remaining infusion duration is, the higher the ranking is, and the sequencing result is used as a priority processing sequence for liquid change to be recommended.

In this step, the method for judging whether the current infusion progress enters the end stage is as follows:

since the diameter of the transfusion bottle at the bottle opening position is gradually shortened, namely the flat body structure is gradually narrowed at the last progress part of the transfusion, when the boundary line of the liquid level surface enters the area, the liquid in the transfusion bottle is about to be input, and therefore, the stage is set as the end stage of the transfusion progress in the invention. The judgment process of the end stage is as follows: and when the current liquid level boundary line of the transfusion bottle is lower than the liquid level boundary line of the previous period for two periods according to the comparison between the current liquid level boundary line L1 of the transfusion bottle and the length L0 of the liquid level boundary line of the previous period, judging that the transfusion progress is at the end stage.

Example 4:

the embodiment provides a big data-based infusion bottle liquid changing sequence recommendation method, which is different from the method in embodiment 1 in that, in order to improve the processing efficiency of nurses and reduce the labor intensity of nurses, in this embodiment, each hospital bed which is in the same ward and is at the end stage of the infusion progress needs to be subjected to priority sequencing and infusion rate adjustment, so that nurses can uniformly process each hospital bed which needs to be subjected to liquid changing operation in the same ward at one time when changing liquid, and the specific adjustment process is as follows:

1) screening the object with the highest priority level to be changed, wherein the screening method comprises the following steps:

the shortest remaining infusion time is obtained according to the remaining infusion time of the infusion bottle corresponding to each sickbed, and the infusion device with the remaining infusion time in the same ward and the difference between the remaining infusion time and the remaining infusion time in a set time (such as three minutes) is classified into the same highest priority level and is set as an object to be changed.

2) In the object to be changed, except the infusion bottle with the shortest remaining infusion time (the infusion device for the first fluid change treatment of the nurse), a voice reminding mode and/or a text display reminding mode are adopted, and infusion rate adjustment suggestions of the infusion bottle are given to patients corresponding to other objects to be changed, and the infusion rate is slowed.

The specific method comprises the following steps: the longest t in the remaining transfusion time in the object to be changed_lThe rate adjustment is carried out as reference, and the infusion rate v in the same ward is adjusted_kAdjusted velocity v_k,newThe mathematical expression of (a) is:

wherein, V_kRepresenting the volume of a transfusion bottle corresponding to the kth object to be changed in the same ward.

In the step, because the patient can not accurately adjust the infusion rate to meet the formula, when the suggestion of adjusting the infusion rate is given, the adjustment scales can be marked on the infusion rate adjusting pulley of the infusion bottle, and each unit scale corresponds to one infusion rate and can be referred by the patient for adjustment.

Therefore, after the infusion rates of other infusion bottles are slowed down, when a nurse treats the liquid changing conditions of the infusion bottles, the rest infusion bottles are still not infused completely, and the phenomenon that the infusion is completed when the nurse does not treat the infusion bottles, so that the blood return of an infusion port of a patient is caused is avoided.

On the basis, if the shortest infusion time of the objects to be changed in a plurality of wards is the same, the objects to be changed in the shortest infusion time are prioritized according to the actual situation in the ward where the shortest infusion time is located, and the objects to be changed in the shortest infusion time are prioritized to prevent a nurse from having time to change the objects to be changed in the same time. The method also comprises the following processing steps:

setting the mean square error of the remaining time of the object to be changed in the same ward as sigma_kThe total number of the objects to be transfused in each ward is N, N is more than or equal to 1 and less than or equal to N, and N is the number of sickbeds with transfused objects in the sickrooms. A priority characteristic value P reflecting the order of change of fluid for each ward can thus be established, which is calculated by the expression:

then the liquid changing sequence with the same remaining shortest time and containing the objects to be changed in each ward can be sorted according to the priority characteristic value P according to the formula.

After the sequencing sequence is obtained, except the ward corresponding to the highest priority characteristic value P, all the other wards are sequentially slowed down in infusion rate according to the priority characteristic value P from high to low. For example, the priority characteristic value P of the ward 1 is the highest, and the fluid-changing infusion bottles (corresponding to the patient bed numbers) with the highest priority level are sorted according to the steps 1) and 2) above, for example, the patient bed number is 102>103; the priority characteristic value of the No. 2 ward is P times, if only the sickbed 201 in the ward is the object to be changed; the priority feature value P of the ward No. 3 is next, if only the sickbed 302 in the ward is the object to be changed, the final sequence of the object to be changed is: 102>103>201>302, according to the residual transfusion time t of the object to be changed in the previous rank₁₀₃Calculating the adjusted speed v of the infusion rate of the infusion bottle corresponding to the sickbed 201₂₀₁The calculation formula is as follows:

where Δ t is a set one hour difference, t₂₀₁＝t₁₀₃+Δt。

Then, the rest of the objects to be changed are ranked according to the previous rankTime t of infusion₂₀₁Calculating the adjusted speed v of the infusion rate of the infusion bottle corresponding to the patient bed 302₃₀₂The calculation formula is as follows:

wherein, t₃₀₂＝t₂₀₁+Δt。

Finally, the liquid changing sequence of each ward can be obtained according to the preliminary ranking sequence of the infusion progress in each ward and the further sequencing of the priority characteristic value P when the preliminary ranking is the same, and a nurse can timely carry out the liquid changing operation according to the obtained ranking sequence.

Example 5:

the embodiment provides a big data-based infusion bottle liquid changing sequence recommendation system which comprises a controller, a display and cameras arranged in each ward, wherein the controller is connected with each camera, the controller is connected with the display, and is used for acquiring images photographed by each camera, processing the images containing infusion bottles (infusion bottles) in each ward according to any one of the methods in the embodiments 1 to 4, generating a sequencing result of liquid changing objects, and displaying the sequencing recommendation result on the display.

It should be noted that: the precedence order of the above embodiments of the present invention is only for description, and does not represent the merits of the embodiments. And specific embodiments thereof have been described above. Other embodiments are within the scope of the following claims. In some cases, the actions or steps recited in the claims may be performed in a different order than in the embodiments and still achieve desirable results. In addition, the processes depicted in the accompanying figures do not necessarily require the particular order shown, or sequential order, to achieve desirable results. In some embodiments, multitasking and parallel processing may also be possible or may be advantageous.

The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents, improvements and the like that fall within the spirit and principle of the present invention are intended to be included therein.