TWI897244B - Intelligent recognition method and device of waste fluid dialysis - Google Patents

Abstract

An intelligent recognition method and device of waste fluid dialysis are disclosed. The method includes the steps of: storing a plurality of waste fluid dialysis image information in a database, each waste fluid dialysis image information including an indicator image information and being assigned a respective turbidity state; using a control unit to acquire an image color brightness of the indicator image information of each waste fluid dialysis image information; generating an artificial intelligence recognition module trained by machine learning or deep learning based on the aforesaid image color brightness of the indicator image information of each waste fluid dialysis image information and the respective turbidity state assigned to each waste fluid dialysis image information; inputting a waste fluid dialysis image message into the control unit; the artificial intelligence recognition module thereby obtaining an indicator image message of the waste fluid dialysis image message and generating a predictive turbidity state of the waste fluid dialysis image message.

Description

Dialysis waste fluid intelligent identification method and device

This invention relates to a method and device for intelligently identifying dialysis wastewater, and more particularly, to an invention capable of identifying the turbidity of dialysis wastewater.

The kidneys' basic physiological functions include urine production, regulating body water and osmotic pressure, regulating electrolyte concentrations, regulating acid-base balance, and endocrine function. Urine production is crucial for excreting waste products from the blood. When the kidneys lose their function and fail to function properly, toxins in the body cannot be excreted, causing harm to the body.

The prevalence of kidney disease in Taiwan is over 10%, and the density of dialysis patients is the highest in the world. Most patients with kidney disease use hemodialysis or peritoneal dialysis to replace the kidneys' original function. Hemodialysis involves withdrawing blood from the body, passing it through a hemodialysis machine, removing metabolic waste and impurities, and then returning the cleansed blood to the body. This is commonly known as "dialysis." To prevent the accumulation of excessive toxins in the body, patients must visit the hospital regularly for hemodialysis. For example, the recommended frequency is three times per week. However, this also imposes restrictions and inconveniences on patients' lives.

To improve patients' quality of life, peritoneal dialysis is a treatment option. It uses the membrane covering the internal organs within the abdominal cavity as a filter to remove water and waste from the body. Peritoneal dialysis can be performed at home, allowing for flexibility in daily life and work, without the pain of needle sticks, and reducing the risk of bloodstream infections. However, during peritoneal dialysis, in addition to injecting new dialysis fluid into the body, The dialysate remaining in the body must be drained. When performed in a medical facility, nurses can determine whether a patient has peritoneal inflammation by observing the clarity of the discharged wastewater. Peritoneal dialysis is a home treatment. Although nurses will objectively guide patients on how to identify the clarity of their dialysis wastewater, patients themselves lack the expertise to accurately determine whether they have peritoneal inflammation based on the dialysis wastewater.

Therefore, Republic of China Patent Publication No. I559254B discloses a "remote care system and method for peritoneal dialysis" that includes at least a storage container, a mounting bracket, a detachable detection device, and a mobile device. The storage container holds a recovered fluid from peritoneal dialysis. The mounting bracket houses the storage container. The detachable detection device includes a detection unit, a processor, and a communication interface. The processor uses a detection unit to detect the turbidity and color of the recovered liquid, performs a certain amount of analysis on the recovered liquid, and generates a turbidity prediction value based on the detection unit's detection results. The processor calculates first and second turbidity values corresponding to the first and second turbidities of the recovered liquid based on first and second algorithms, respectively. Based on a predetermined threshold, the processor selects one of the first and second turbidity values to generate the turbidity prediction value, which is then sent to the mobile device via a communication interface for further computational processing.

Or there is a patent publication number CN113476677A of the People's Republic of China that discloses a "peritoneal dialysis fluid exchange detection device" belonging to the field of medical equipment technology. The peritoneal dialysis fluid exchange detection device includes a base and a column. The bottom of the column is fixedly connected to the base. The top of the column is evenly distributed with hooks. A manifold is provided below the hooks. A dialysis solution tray is provided below the manifold. A dialysis solution bag is placed on the top of the dialysis solution tray. A second container is provided below the dialysis solution tray. The second support is equipped with a glucose concentration meter, a visual monitor, and a laser emitter. A drainage solution tray containing a drainage solution bag is located below the second support. This device monitors fecal status during peritoneal dialysis treatment, providing patients with a basic understanding of their treatment status and helping to stabilize their mood. Medical staff can also directly assess the patient's treatment effectiveness based on the test information. It is convenient, easy to use, and provides accurate test information.

While the aforementioned method can use the emission and absorption of light to determine the turbidity of wastewater, the changes in light are easily affected by other external light sources, leading to misjudgment of the wastewater's turbidity.

And there is U.S. Patent 20130345622A1 "EUSABLE EFFLUENT DRAIN CONTAINER WITH KEY FEATURE FOR DIALYSIS AND OTHER MEDICAL FLUID THERAPIES", which discloses a dialysis system including an outflow dialysate source, a discharge container configured to receive the outflow dialysate via a discharge tube, and a force sensor. The discharge container includes an at least semi-rigid body defining a first key feature. The weighing sensor includes a second mating key feature positioned and arranged to mate with the first key feature; in addition, the description specifically mentions in paragraph [0018] that the container also includes features for observing and sampling the outflow or used dialysate. For example, the front or top of the container may have one or more windows 166 for observing the liquid discharged from the container. Alternatively, one or more of the waste fluid inlet cap or the waste fluid outlet cap may be transparent or see-through to facilitate observation of the effluent within the container. Furthermore, the container may be sealed together from separate components, one or more of which are transparent or see-through to facilitate observation of the color and consistency of the discharged fluid, desirably providing printed text and/or a colored surface to assist the patient in determining whether the effluent is turbid.

Although the aforementioned case allows patients to actively observe whether their wastewater is turbid by using auxiliary printed text and/or colored surfaces as a reference, each patient's subjective judgment is different, so the results of the judgment will also vary. Moreover, patients lack professional knowledge and may make misjudgments.

Therefore, the present inventors propose a dialysis wastewater intelligent identification method to enable patients to understand the condition of their wastewater after dialysis. The method comprises the following steps: Storing a plurality of dialysis wastewater image information in a database, each dialysis wastewater image information including an indicator portion image information, and each dialysis wastewater image information is assigned a turbidity condition; A control unit obtains the aforementioned dialysis wastewater image information; The image color brightness of the indicator image information of the dialysis waste image information is generated by machine learning or deep learning training according to the image color brightness of the indicator image information of the dialysis waste image information and the turbidity state assigned to each of the dialysis waste image information; the dialysis waste image information is input into the control unit, and the artificial intelligence recognition module is generated. The group can obtain an indicator image information of the dialysis waste liquid image information and predict a predicted turbidity state of the dialysis waste liquid image information; wherein a closed accommodating space is provided in a housing, a camera unit and a backlight unit are provided in the housing, and adjacent to the accommodating space, and the camera unit and the backlight unit are arranged facing each other, the backlight unit has a side, and the side is provided with an indicator The side of the display unit faces the camera unit, and the display unit includes a plurality of blocks arranged according to color brightness. A transparent bag containing dialysis wastewater is placed in the storage space and positioned between the camera unit and the backlight unit, with the dialysis wastewater in the transparent bag covering the display unit. The camera unit is operated to photograph the transparent bag and the display unit to obtain image information of the dialysis wastewater.

Furthermore, the control unit transmits the predicted turbidity status corresponding to the dialysis wastewater image information to a cloud management unit.

Furthermore, the aforementioned turbidity conditions include normal, slightly turbid, and turbid. The predicted turbidity conditions correspond to the three conditions: normal, slightly turbid, and turbid.

Furthermore, the algorithm used to train the artificial intelligence recognition module using machine learning or deep learning includes one of the following: Logistic Regression, Random Forest, Support Vector Machines (SVM), K-Nearest Neighbor (KNN), eXtreme Gradient Boosting (XGBoost), or Multilayer Perceptron (MLP).

A dialysis waste liquid intelligent identification device is used in conjunction with a transparent bag containing dialysis waste liquid. The dialysis waste liquid intelligent identification device comprises: a housing having a storage space; a camera unit disposed in the housing and adjacent to the storage space; a backlight unit disposed in the housing and adjacent to the storage space, and the backlight unit and the camera unit are disposed facing each other, the backlight unit having a side surface provided with an indicator light. The display unit includes a plurality of blocks arranged according to color brightness, and the side is arranged facing the photographic unit; a control unit is electrically connected to the photographic unit, and the control unit has a database that stores a plurality of dialysis waste liquid image information, each dialysis waste liquid image information includes an indicator portion image information, and each dialysis waste liquid image information is assigned a turbidity state; the control unit obtains The image color brightness of the indicator image information of the dialysis waste image information is generated by machine learning or deep learning training based on the image color brightness of the indicator image information of the dialysis waste image information and the turbidity state of the dialysis waste image information; the transparent bag containing the dialysis waste is placed in the storage space and the camera unit is positioned between the bag and the dialysis waste image information. and the backlight unit, and the dialysis waste fluid in the transparent bag covers the indicator portion; the camera unit photographs the transparent bag and the indicator portion to obtain a dialysis waste fluid image message; the dialysis waste fluid image message is input into the control unit, so that the artificial intelligence recognition module can obtain an indicator portion image message of the dialysis waste fluid image message and predict a predicted turbidity state of the dialysis waste fluid image message.

Furthermore, the housing includes an upper housing and a lower housing, and the upper housing covers the lower housing to form the accommodation space.

Furthermore, a cloud management unit is electrically connected to the control unit, and the control unit transmits the predicted turbidity status corresponding to the dialysis waste fluid image information to the cloud management unit.

Furthermore, the control unit has a display screen disposed on the housing, and the predicted turbidity condition can be selectively displayed on the display screen.

Furthermore, the control unit is electrically connected to the backlight unit.

Based on the above technical features, the following effects can be achieved:

1. Place the transparent bag containing dialysis waste in a closed space, with the backlight unit as the sole illumination source. This results in more stable images of the dialysis waste in the transparent bag, less susceptible to interference from other external light sources.

2. An indicator is installed on one side of the backlight unit near the camera unit. When the camera unit captures an image of the dialysis wastewater in the transparent bag, the artificial intelligence recognition module can identify the color and brightness of the image and predict the turbidity of the dialysis wastewater in the transparent bag.

1: Shell

11: Upper housing

12: Lower housing

13: Storage Space

2: Photography unit

3: Backlight unit

31: Instruction Department

4: Control unit

41: Artificial Intelligence Recognition Module

42: Display screen

43:Database

5: Cloud Management Unit

6: Transparent bag

7A, 7B, 7C: Dialysis waste image information

71A, 71B, 71C: Indicator video information

[Figure 1] is a three-dimensional external view of the dialysis wastewater intelligent identification device of the present invention.

[Figure 2] is an exploded view of the dialysis wastewater intelligent identification device of the present invention.

[Figure 3] is an external view of the backlight unit of the present invention.

[Figure 4] is a block diagram of the dialysis wastewater intelligent identification device of the present invention.

[Figure 5] is a flow chart of the dialysis wastewater intelligent identification method of the present invention.

Figure 6 shows the placement of a transparent bag into the dialysis wastewater intelligent identification device of the present invention.

[Figure 7] is a schematic diagram showing the transparent bag being placed in the storage space of the present invention.

[Figure 8] is a cross-sectional view of Figure 7.

Figure 9A is a schematic diagram of dialysis waste image information, which is predicted to be "normal" in turbidity, as identified by the dialysis waste intelligent identification method of the present invention.

Figure 9B is a schematic diagram of dialysis wastewater image information whose turbidity is predicted to be "slightly turbid" using the dialysis wastewater intelligent identification method of the present invention.

Figure 9C is a schematic diagram of dialysis waste image information whose turbidity is predicted to be consistent with "turbidity" using the dialysis waste intelligent identification method of the present invention.

Taking into account the above technical features, the main functions of the dialysis wastewater intelligent identification method and device of the present invention will be clearly demonstrated in the following embodiments.

Please refer to Figures 1, 2, 3, and 4. The dialysis wastewater intelligent identification device of the present invention comprises a housing 1, a camera unit 2, a backlight unit 3, a control unit 4, and a cloud management unit 5. It is used in conjunction with a transparent bag 6 containing dialysis wastewater.

The housing 1 has a storage space 13. Specifically, the housing 1 includes an upper housing 11 and a lower housing 12. The upper housing 11 covers the lower housing 12 to form the storage space 13. The housing 1 is in the shape of a rectangular column. The upper housing 11 and the lower housing 12 are pivotally connected at one of the long sides and can be rotated relative to each other to open or close.

The camera unit 2 and the backlight unit 3 are both mounted within the housing 1, adjacent to the housing space 13. The backlight unit 3 and the camera unit 2 are positioned facing each other. The backlight unit 3 has an indicator 31 located on a side of the backlight unit 3 facing the camera unit 2. Specifically, the backlight unit 3 is a rectangular sheet mounted on the inner wall of the lower housing 12, while the camera unit 2 is mounted on the inner wall of the upper housing 11. The camera unit 2's lens is oriented toward the backlight unit 3. The indicator 31 includes a plurality of image blocks arranged according to color brightness.

The control unit 4 is electrically connected to the camera unit 2 and the backlight unit 3. The control unit 4 includes an artificial intelligence recognition module 41, a display screen 42, and a database 43. The display screen 42 is disposed on the housing 1. The cloud management unit 5 is electrically connected to the control unit 4. The database stores a plurality of dialysis waste image information. Each dialysis waste image information includes an indicator image information, and each dialysis waste image information is assigned a turbidity status.

Referring to Figures 4, 5, 6, 7, 8, and 9A, the dialysis waste intelligent identification method of the present invention includes the following steps: the control unit obtains the color and brightness of the image information of the indicator portion of the dialysis waste image information, and generates the artificial intelligence identification module through machine learning or deep learning training based on the image color and brightness of the image information of the indicator portion of the dialysis waste image information and the turbidity status assigned to the dialysis waste image information.

Next, the transparent bag 6 is placed in the accommodating space 13 so that the transparent bag 6 is between the photographic unit 2 and the backlight unit 3 and is placed flat so that the dialysis waste fluid in the transparent bag 6 can completely cover the indicator portion 31; the backlight unit 3 is activated to emit light to illuminate the transparent bag 6, and the photographic unit 2 photographs the transparent bag 6 to obtain a dialysis waste fluid image message 7A, which includes an indicator portion 31. The indicator image information 71A includes the image color brightness. Specifically, when the transparent bag 6 containing dialysis wastewater is placed between the camera unit 2 and the backlight unit 3, the color brightness of the indicator 31 captured by the camera unit 2 is altered by the dialysis wastewater and becomes different from the actual color brightness of the indicator 31, resulting in the image color brightness. The image color brightness is then recorded in the control unit 4.

The control unit 4 receives the dialysis waste image signal 7A and uses the artificial intelligence recognition module 41 to recognize the indicator image signal 71A, thereby further predicting a predicted turbidity state of the dialysis waste image signal 7A. The control unit 4 then transmits the predicted turbidity state corresponding to the dialysis waste image signal 7A to the cloud management unit 5, or directly displays the result on the display screen 42. Nursing staff can also view the user's dialysis waste status in real time through the cloud management unit 5 and promptly remind the user of the predicted dialysis waste result.

Please refer to Figures 4, 9A, 9B and 9C. Specifically, when the artificial intelligence recognition module 41 is generated by machine learning or deep learning training, multiple images of dialysis waste fluid with different turbidity conditions are used as the aforementioned dialysis waste fluid image information. In this embodiment, since it is difficult to obtain clinical samples, when constructing the aforementioned dialysis waste fluid image information, red, yellow and white pigments are used to simulate the appearance of a natural drainage bag, and 27 liquid samples with different colors and turbidity conditions are mixed. Approximately 1,000 images of dialysis waste fluid are taken and then analyzed by professional nursing staff. Assisting classification; for example, 1,000 images of dialysis wastewater are divided into three categories according to the turbidity state, namely normal, slightly turbid and turbid, for training generation. The 1,000 images are divided into a training data set, a validation data set and a test data set. The training data and the validation data contain 90%, that is, about 900 images, and the test data is the remaining 10%, that is, about 100 images. MobileNet_v1_0.5_224 is selected as the core training model of deep learning image classification technology, or other models such as Logistic Regression are used. Different algorithms, including Random Forest, Support Vector Machines (SVM), K-Nearest Neighbor (KNN), eXtreme Gradient Boosting (XGBoost), and Multilayer Perceptron (MLP), are used to train an artificial intelligence recognition module 41 capable of distinguishing between three types of turbidity: normal, slightly turbid, and turbid. More specifically, the artificial intelligence recognition module 41 distinguishes different turbidity conditions based on the relationship between the image color brightness and the turbidity conditions assigned to the dialysis waste image information.

As shown in Table 1, approximately 100 dialysis waste images from the test data were used to evaluate accuracy, recall, and precision. In terms of accuracy, which represents the ratio of correctly identified images to the total number of images tested, the AI recognition module 41 achieved approximately 91% accuracy in predicting the turbidity condition as normal, approximately 100% accuracy in predicting the turbidity condition as slightly turbid, and approximately 99% accuracy in predicting the turbidity condition as turbid. Overall, the accuracy reached 97%. Regarding recall, which measures the proportion of images with the same recognition result as the expert labeling, the recall rates were 87% for normal, 100% for slightly turbid, and 95% for turbid, respectively, for an overall recall of approximately 90%. In terms of accuracy, which represents the proportion of images identified as the correct level of haze, the accuracy of images identified as normal and hazy is 100%, while the accuracy of images identified as slightly hazy is only 55%, for an overall accuracy of 90%.

When the image captured by the camera unit 2 is similar to the dialysis waste image signal 7A shown in FIG. 9A , the artificial intelligence recognition module 41 determines the image color and brightness of the indicator image signal 71A of the dialysis waste image signal 7A and can further predict that the predicted turbidity of the dialysis waste represented by the dialysis waste image signal 7A is normal.

Similarly, when the image captured by the camera unit 2 is similar to the dialysis waste image signal 7B shown in FIG. 9B , the artificial intelligence recognition module 41 determines the image color brightness of the indicator image signal 71B of the dialysis waste image signal 7B and can further predict that the turbidity of the dialysis waste represented by the dialysis waste image signal 7B is slightly turbid.

Similarly, when the image captured by the camera unit 2 is similar to the dialysis waste image signal 7C shown in FIG. 9C , the artificial intelligence recognition module 41 determines the image color brightness of the indicator image signal 71C within the dialysis waste image signal 7C and can further predict that the turbidity of the dialysis waste represented by the dialysis waste image signal 7C is turbid.

The description of the above embodiments should provide a comprehensive understanding of the operation, use, and efficacy of the present invention. However, the above embodiments are merely preferred embodiments of the present invention and should not be construed as limiting the scope of the present invention. Simply equivalent variations and modifications made within the scope of the patent application and the description of the invention are also covered by the present invention.

1: Housing 11: Upper housing 12: Lower housing 42: Display

Claims (9)

A method for intelligent dialysis wastewater identification comprises the following steps: Storing a plurality of dialysis wastewater image information in a database, each dialysis wastewater image information including an indicator portion image information, and each dialysis wastewater image information being assigned a turbidity status; A control unit obtains the image color and brightness of the indicator portion image information of the dialysis wastewater image information, and generates an artificial intelligence identification module using machine learning or deep learning training based on the image color and brightness of the indicator portion image information of the dialysis wastewater image information and the turbidity status assigned to each dialysis wastewater image information; Inputting a dialysis waste fluid image signal into the control unit enables the artificial intelligence recognition module to obtain an indicator portion image signal of the dialysis waste fluid image signal and predict a predicted turbidity condition of the dialysis waste fluid image signal; A housing is provided with a closed storage space. A camera unit and a backlight unit are disposed within the housing, adjacent to the storage space and facing each other. The backlight unit has a side surface provided with an indicator portion, which faces the camera unit. The indicator portion includes a plurality of image blocks arranged according to color brightness. A transparent bag containing dialysis wastewater is placed within the storage space, between the camera unit and the backlight unit, with the dialysis wastewater within the transparent bag covering the indicator portion. The camera unit captures images of the transparent bag and the indicator portion to obtain image information of the dialysis wastewater. The dialysis wastewater intelligent identification method as described in claim 1, wherein the control unit transmits the predicted turbidity state corresponding to the dialysis wastewater image information to a cloud management unit. The dialysis wastewater intelligent identification method as described in claim 1, wherein the aforementioned turbidity conditions include three conditions: normal, slightly turbid, and turbid, and the predicted turbidity conditions correspond to the three conditions: normal, slightly turbid, and turbid. The method for intelligent identification of dialysis wastewater as described in claim 1, wherein the algorithm used to train the artificial intelligence identification module using machine learning or deep learning includes one of the following: Logistic Regression, Random Forest, Support Vector Machines (SVM), K-Nearest Neighbor (KNN), eXtreme Gradient Boosting (XGBoost), or Multilayer Perceptron (MLP). A dialysis wastewater intelligent identification device is used in conjunction with a transparent bag containing dialysis wastewater. The dialysis wastewater intelligent identification device comprises: a housing having a storage space; a camera unit disposed within the housing and adjacent to the storage space; a backlight unit disposed within the housing and adjacent to the storage space, with the backlight unit and the camera unit facing each other. The backlight unit has a side surface provided with an indicator portion comprising a plurality of image blocks arranged according to color brightness, and the side surface is disposed facing the camera unit. A control unit is electrically connected to the imaging unit and includes a database storing a plurality of dialysis waste image information. Each dialysis waste image information includes an indicator image information, and each dialysis waste image information is assigned a turbidity state. The control unit obtains image color and brightness of the indicator image information of the dialysis waste image information, and generates an artificial intelligence recognition module using machine learning or deep learning training based on the image color and brightness of the indicator image information of the dialysis waste image information and the turbidity state assigned to each dialysis waste image information. The transparent bag containing dialysis wastewater is placed in the storage space between the camera unit and the backlight unit, with the dialysis wastewater in the transparent bag covering the indicator portion. The camera unit is operated to capture the transparent bag and the indicator portion to obtain an image of the dialysis wastewater. The dialysis wastewater image information is input into the control unit, allowing the artificial intelligence recognition module to obtain an image of the indicator portion of the dialysis wastewater image information and predict a predicted turbidity condition of the dialysis wastewater image information. As described in claim 5, the dialysis waste liquid intelligent identification device, wherein the housing includes an upper housing and a lower housing, and the upper housing covers the lower housing to form the accommodating space. The dialysis waste fluid intelligent identification device as described in claim 5, wherein a cloud management unit is electrically connected to the control unit, and the control unit transmits the predicted turbidity state corresponding to the dialysis waste fluid image information to the cloud management unit. The dialysis waste fluid intelligent identification device as described in claim 5, wherein the control unit has a display screen, the display screen is arranged on the housing, and the predicted turbidity condition can be selectively displayed on the display screen. The dialysis waste fluid intelligent identification device as described in claim 5, wherein the control unit is electrically connected to the backlight unit.