Machine learning based external standard-free sample quantitative detection method

Abstract

The present application provides a machine learning-based external standard-free sample quantitative detection method, characterized in that it comprises the following steps: (a) adding the sample on a medium carrier or into a detection kit, and imaging the sample; (b) automatically analyzing the image signal of the sample image and predicting the sample concentration result by using a trained machine learning model; and (c) outputting the sample quantitative result. By combining the concept of traditional colorimetry with popular machine learning methods, an intelligent device algorithm is developed, making it possible to have a more convenient and rapid sample detection scheme compared with traditional colorimetry.

Description

TECHNICAL FIELD

[0001] The present application relates to a machine learning-based external standard-free sample quantitative detection method, belonging to the technical field of artificial intelligence colorimetric sample analysis. BACKGROUND

[0002] Machine learning is a multi-disciplinary subject, involving probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory and other disciplines, and is the core of artificial intelligence and the fundamental approach to making computers intelligent. It is a specialized study of how computers simulate or implement human learning behavior to acquire new knowledge or skills, reorganize existing knowledge structure to continuously improve their performance.

[0003] The existing traditional laboratory detection method has defects such as large detection equipment volume, high cost, and complex operation, and can only be used in laboratories and hospitals, which limits their application in resource-limited environments. Compared with traditional laboratory methods, intelligent devices have good portability, low cost, reliability, flexibility and other advantages, and are very suitable for on-site rapid sample detection.

[0004] Intelligent devices have excellent imaging capabilities and excellent computing efficiency and are easy to carry. This helps researchers solve many difficult problems. In recent years, intelligent device colorimetry has become an instant and rapid on-site sample detection technology, which is widely used in biological, chemical, pharmaceutical, clinical and other fields. However, the current intelligent colorimetric method still needs external standard quantitative solution, external detection hardware and other complicated procedures, so we propose a more convenient solution, that is, to realize more convenient and rapid sample detection through machine learning. SUMMARY

[0005] Technical scheme: In view of the limitations of the current colorimetric detection method, in order to achieve the above purpose, the present application adopts the following technical scheme:

[0006] A machine learning-based external standard-free sample quantitative detection method, comprising the following steps:

[0007] (a) adding the sample to the medium carrier or into the detection kit, and imaging the sample;

[0008] (b) inputting the sample image, automatically analyzing the image signal and loading the trained machine learning model to predict the sample concentration result;

[0009] (c) outputting the sample quantitative result.

[0010] In one embodiment of the present application, the sample volume added in step (a) is 50 μL

[0011] In an embodiment of the present application, the sample imaging method in step (b) comprises a picture taken by a device with a camera function.

[0012] In an embodiment of the present application, the program in step (c) automatically analyzes the image signal, and the automatic analysis of the image signal comprises extracting color signal mean value with a user-defined ROI as the region size; the image signal comprises RGB and HSV color space color signal value.

[0013] In an embodiment of the present application, the image signal in step (b) comprises RGB and HSV color space color signal value.

[0014] In an embodiment of the present application,

[0015] The machine learning model in (b) is:

[0016]

[0017] y = w1Q + w2S1 + … + w n S n + β,

[0018] wherein Q, S_1…S_m and y ∈ R n ; Q is a quantitative parameter vector composed of one or more of the color parameters RGBHSV six-parameter color values most related to the concentration of the sample to be detected; S_1…S_m respectively represent an influence factor vector, including illumination, shooting distance, camera parameters; y is a standard concentration result vector composed of standard sample concentration, and β is a constant term; all training set sample images are analyzed and color parameter values are read to obtain influence factor values; finally, the corresponding concentration of the sample is taken as a label vector, and the sample color parameter values and influence factor values are combined to form a training set matrix, and an optimized gradient descent loss function is constructed and trained.

[0019] In an embodiment of the present application, the loss function of the machine learning model is:

[0020]

[0021] wherein X is the feature matrix of the training sample, and w represents the weight matrix; α is 0.1, that is, the L1 regularization term limits the degree of the model, the greater the α, the lower the degree of freedom of the model, and the less likely to overfit.

[0022] In an embodiment of the present application, the process of predicting the sample concentration result in step (b) is as follows: first, a machine learning model is trained according to a large number of sample data sets taken in various experimental environments, and the model with the optimal prediction performance is selected and placed in the smartphone program; second, when detecting, the mean values of the RGB and HSV signals of the loaded sample area are automatically read to obtain sample characteristics, then the program predicts the sample quantitative result according to the machine learning model and prints the output to the screen; thus, the sample rapid automatic detection is completed.

[0023] In an embodiment of the present application, the smart device includes a smartphone or a tablet computer and other portable devices with good computing capability.

[0024] In an embodiment of the present application, the detection kit includes a three-dimensional structured kit providing pretreatment and color development detection functions, which is composed of a photosensitive bottom layer and a liquid storage tank, the liquid storage tank is square with a side length of 5 mm, the groove depth of the liquid storage tank is 2 mm, and the inner core structure is a detachable strip-shaped chip, each chip can add 6 samples, and can be used separately, up to 4 chips can be loaded in each kit at the same time, and the kits can also be used side by side, so as to achieve high-throughput detection.

[0025] In an embodiment of the present application, an electronic device is also provided, which includes a memory and a processor, wherein the memory stores a smart device program, and the processor is configured to run the smart device program to perform the method of any one of claims 1 to 8.

[0026] The present application also provides an integrated detection device, which includes a sampling barrel, a sample holder arranged in the sampling barrel, and a kit arranged on the sample holder, wherein the sampling barrel further includes an imaging device and a detection unit, the detection unit takes a sample in the kit through the imaging device, and the sample concentration value is quickly predicted according to a machine learning model after the sample signal characteristics are automatically extracted by the method.

[0027] Advantages

[0028] The present application has the advantages that the detection process is further simplified, the machine learning method is more flexible, fast and accurate compared with the traditional colorimetric detection and laboratory methods such as large detection device volume, high cost and complex operation, and the like, the staff can use the method to quickly and accurately complete sample detection in daily and on-site environments, and the time for determining the sample quantitative whole process is not more than 2 minutes.

[0029] The present application provides a machine learning-based external standard-free sample quantitative detection method. By combining the concept of traditional colorimetry with popular machine learning methods, an intelligent device algorithm is developed, making it possible to detect samples more conveniently and quickly compared to traditional colorimetry. Using this method, the entire sample quantitative determination process takes no more than 2 minutes, and the accuracy of the results has been verified by standard laboratory methods BRIEF DESCRIPTION OF DRAWINGS

[0030] Figure 1 Machine learning method detection process

[0031] Figure 2 Model training flow

[0032] Figure 3 Actual sample example 1 for machine learning prediction

[0033] Figure 4 Actual sample example 2 for machine learning prediction

[0034] Figure 5 Prediction of actual samples in the machine learning model training process, a-d are 38, 110, 240, and 310 training data, respectively. The horizontal axis represents the standard result of the sample, and the vertical axis represents the model prediction result.

[0035] Figure 6 Kit structure diagram DETAILED DESCRIPTION

[0036] The present application will be further illustrated below with specific examples, which should be understood as only for illustrating the present application and not for limiting the scope of the present application. After reading the present application, those skilled in the art can modify various equivalent forms of the present application, which fall within the scope defined by the appended claims.

[0037] The present application provides a machine learning-based external standard-free sample quantitative detection method, which includes the following steps:

[0038] (d) Drop the sample on the medium carrier or into the detection kit, and image the sample;

[0039] (e) Import the sample image, automatically analyze the image signal, and load the trained machine learning model to predict the sample concentration result;

[0040] (f) Output the sample quantitative result.

[0041] EMBODIMENT

[0042] Herein, the emodin sample is taken as an example. Emodin is an anthraquinone compound, has a special reaction, is almost insoluble in water, soluble in ethanol and alkali solution, and shows a red-purple reaction (Borntrager reaction) in an alkaline solution. This reaction is one of the most commonly used methods for identifying the presence of hydroxyanthraquinone components in Chinese herbal medicines. Therefore, in this example, mixed alkali is used as a color developing reagent (10% NaOH and 4% NH4OH mixed in equal volume), and with the increase of the concentration of emodin, the red color of the solution will change from light to dark. Therefore, the machine learning method is used to learn the solution characteristics (including colorimetric values and some external characteristics) in the overall linear concentration range, and an intelligent device detection method capable of rapid and simple quantification is established.

[0043] It should be noted that, Figure 3 , 4 The chip part inside the detection kit proposed by the present application can take samples, specifically, after the sample is added dropwise, the sampling device is placed on the sample or the kit for taking pictures; if it is an integrated detection device, the sample is placed in the sample holder, and then the imaging device is used for taking pictures; after the sampling is completed, the sample signal feature extraction (including sample color value) is automatically completed by the algorithm of the program of the present application, then the program rapidly predicts the sample concentration value according to the machine learning model, in addition, in order to further verify the accuracy of the result, the standard result is also used for comparison, see Figure 5 .

[0044] 50 μL of emodin solution samples with different concentrations are accurately taken and added dropwise into the chip, and the sample concentration covers the entire linear range, so as to observe the quantitative accuracy of the system.

[0045] The above experiment is carried out in the kit shown in the following Figure 6 The kit is designed for machine learning sample detection, and is composed of a photosensitive bottom layer and a liquid storage tank. A detachable strip-shaped chip design is adopted, each chip can add 6 samples, and at most 4 chips can be loaded at the same time, and 24 samples can be added dropwise at the same time. In addition, the kits can also be combined, and a plurality of kits can be placed side by side for the sampling area of the image acquisition device, so as to greatly improve the detection throughput.

[0046] Table 1: Comparison of machine learning prediction and standard actual sample results

[0047]

[0048]

[0049] The above is only a preferred embodiment of the present application, and it should be noted that for those skilled in the art, without departing from the principles of the present application, a number of improvements and refinements can also be made, which should be considered as the protection scope of the present application.

Claims

1. A method for machine learning based external standard-free sample quantitative detection, characterized in that, The method comprises the following steps: (a) dropping the sample on a medium carrier or into a detection kit, and imaging the sample; (b) inputting the sample image, automatically analyzing the image signal, and loading a trained machine learning model to predict the sample concentration result; (c) outputting the sample quantitative result; The machine learning model in step (b) is: ， wherein, Q , S_1…S_m and y ∈ R n ; Q is a quantitative parameter vector composed of one or more of the six parameters of the color value RGBHSV which are most relevant to the concentration of the sample to be detected; S_1…S_m respectively represent the influencing factor vectors, including illumination, shooting distance, camera parameters; y is a standard concentration result vector composed of the concentration of the standard sample, β is a constant term; The color parameter values of all the training set sample images are analyzed and read out to obtain the influence factor values; Finally, the sample corresponding concentration is taken as a label vector, and the sample color parameter values and the influence factor values are spliced to form a training set matrix, and an optimized gradient descent loss function is constructed and trained. The loss function of the machine learning model is: ， where X is the feature matrix of the training sample, w represents the weight matrix, and α is 0.

1.

2. The method of claim 1, wherein, The sample imaging method in step (a) includes taking pictures using a smart device with a photographing function.

3. The method of claim 1, wherein, The automatic analysis of the image signal in step (b) includes extracting the color signal mean value with a user-defined ROI as the region size; the image signal includes RGB and HSV color space color signal values.

4. The method of claim 1, wherein, The process of predicting the sample concentration result in step (b) is: First, train the machine learning model according to the sample data set taken under various experimental environments, and select the model with the best prediction performance; Second, during detection, automatically read the RGB and HSV signal mean values of the loaded sample region to obtain the sample features, predict the sample quantitative result according to the machine learning model, and print the output to the screen.

5. The method of claim 2, wherein, The smart device includes a smartphone or a tablet computer and other portable devices with good computing power.

6. The method of claim 1, wherein, The detection kit includes a three-dimensional structured kit that provides pretreatment and color development detection functions, which is composed of a photosensitive bottom layer and a liquid storage tank. The liquid storage tank is square with a side length of 5 mm, and the groove depth of the liquid storage tank is 2 mm. The inner core structure is a detachable strip-shaped chip. Each chip can add 6 samples, and can be used separately. Each kit can load up to 4 chips at the same time, and the kits can also be used side by side to achieve high-throughput detection. 7.An electronic device comprising a memory and a processor, the electronic device characterized by, The memory stores a smart device program, and the processor is configured to run the smart device program to perform the method of any one of claims 1-6.

8. An integrated detection apparatus comprising a sampling cartridge, a sample holder provided in the sampling cartridge, and a reagent cartridge provided on the sample holder, wherein The imaging device and the detection unit are also included in the sampling barrel. The detection unit takes samples by imaging the samples in the kit through the imaging device. After automatically extracting the sample signal features through the method of any one of claims 1-6, the sample concentration value is quickly predicted according to the machine learning model.

Images