RU2819131C1 - Mobile detachable complex and method for automatic determination of hemolysis and/or lipaemia in blood samples - Google Patents

Abstract

FIELD: medicine.

SUBSTANCE: group of inventions refers to medicine, namely to laboratory diagnostics, and can be used for automatic determination of hemolysis and/or lipaemia in blood samples. Mobile detachable complex for automatic determination of the presence of hemolysis and/or lipaemia in centrifuged blood samples includes at least 2 digital short-focus cameras with a resolution of at least 1920x1080 pixels; annular LED white light source; tripod made with possibility of fixation to conveyor line of laboratory analyser of digital short-focus cameras and a circular LED white light source in the horizontal plane relative to the conveyor line of the laboratory analyser, a control and information processing system. Image is obtained by the cameras of the mobile detachable complex. Presence of a test tube on the image is detected using a first classification model. Plasma type is classified. If the first classification model detects the presence of a test tube with blood on the image, the operation of the second classification model is started, which simultaneously searches for image regions corresponding to plasma and classifies the type of plasma. Barcode placed on the test tube with blood is recognized using a second classification model. Results of determining presence of hemolysis and/or lipaemia in blood samples are displayed on the monitor.

EFFECT: group of inventions provides reliable determination of lipaemia and/or hemolysis in blood samples in laboratories without additional equipment due to use of a mobile detachable complex and classification models.

15 cl, 5 dwg, 4 tbl, 1 ex

Description

Field of technology to which the invention relates

The group of inventions relates to the field of medicine, namely to laboratory diagnostics, and can be used to automatically determine the presence of hemolysis and/or chylosis in blood samples.

State of the art

Automation is a priority direction for laboratory development, including in order to increase labor productivity by automating routine processes. In connection with the annually growing market for biochemical and immunological research in the Russian Federation associated with the spread of diseases of the cardiovascular system and cancer, as well as with the developing tendency of the population to monitor their health status and the continuing problem of pre-analytical processing of blood samples for the presence of chylosis and hemolysis, which impede biochemical analysis, it is relevant and necessary to develop a system for automatically analyzing the quality of incoming blood samples for the presence of chylosis and/or hemolysis with minimal participation of a doctor before performing the analysis.

Currently, medical samples are examined visually by qualified laboratory personnel. However, this method is labor-intensive, and the results are very subjective.

Also known is the method of spectrophotometry research based on measuring absorption spectra in the optical region of electromagnetic radiation (BUT K.M.S. et al. Optimization of hemolysis, icterus and lipemia interference thresholds for 35 clinical chemistry assays. Practical Laboratory Medicine. 2021, Volume 25 , e00232). However, devices that implement this method can be large (624 kg), as well as difficult to handle and maintain.

The use of computer vision is a much more accessible and at the same time no less effective method for determining the degree of hemolysis and lipemia (chylosis), since it does not require special equipment or personnel training. This approach is an alternative to such methods of analysis as spectrophotometry and less reliable methods of analysis based on the expert assessment of a laboratory employee. This approach can significantly speed up the analysis of blood samples and eliminate the possibility of human error in the analysis, which makes laboratory blood diagnostics more reliable and reliable.

The known device and method for determining the degree of hemolysis and lipemia are US 10746665 B2, and US10816538B2. The device consists of cameras, light sources, a conveyor and a computer, and the method includes obtaining an image of a blood sample at different exposure times and in several different spectra having different nominal wavelengths, creating image data with optimal exposure for each spectrum, classifying the data. In this case, the device can be a structural part of an automated laboratory system or analyzer.

A device and method for determining the degree of hemolysis and lipemia is known US 10928310 B2. The device consists of cameras, light sources and a computer, and the method includes obtaining an image of a blood sample in several exposures or with illumination in the 400-700 nm spectrum, consolidating images, obtaining statistical data for each of the pixels with optimal exposure in a set of transmittance data at different lengths waves (for example, R, G, B) and for each viewpoint, classification of the obtained data. In this case, the device can be a structural part of an automated laboratory system or analyzer, or stand-alone. This source of information is the closest analogue of the device and method we have created for automatically determining the presence of hemolysis and chylosis in blood samples.

The main disadvantage of the above solutions is their stationarity. However, in the case of the autonomous version of the closest analogue, the disadvantage is the need for automatic supply of blood samples, which requires additional equipment for supplying samples.

The technical problem solved by the claimed group of inventions is the creation of a removable, mobile, easy-to-use and maintain device for automated detection of errors at the preanalytical stage, namely chyle and/or hemolysis, as well as the creation of a method for accurately determining the level of chyle and hemolysis.

Disclosure of the invention

The technical result of the created group of inventions lies in the reliable and reliable determination of chylosis and/or hemolysis in blood samples, implemented through a mobile removable complex for automatically determining the presence of hemolysis and/or chylosis, the simplicity and accessibility of which allows the use of the claimed solution in laboratories without additional equipment.

The technical result, in particular, is achieved through a mobile removable complex for automatically determining the presence of hemolysis and/or chylosis in centrifuged blood samples, including:

- at least 2 digital short-focus cameras with a resolution of at least 1920x1080 pixels;

- ring LED white light source;

- a tripod made with the possibility of fixing to the conveyor line of a laboratory analyzer of digital short-focus cameras and a ring LED source of white light in a horizontal plane relative to the conveyor line of the laboratory analyzer, consisting of a frame, holders made with the possibility of fixing and changing the vertical position of the cameras, and removable fastening elements , made with the ability to fix and remove the stand from the conveyor line of the laboratory analyzer,

in this case, the removable fastening elements are made either with the possibility of installing a tripod outside the conveyor line of the laboratory analyzer, so that short-focus cameras record samples through the side windows of the conveyor line of the laboratory analyzer,

or are designed with the ability to install short-focus cameras inside the conveyor line of a laboratory analyzer;

- a control and information processing system that analyzes images obtained from short-focus cameras using the first and second classification models, and is capable of recording, processing, storing and displaying on a monitor the results of determining the presence of hemolysis and/or chylosis in blood plasma samples with information about the patient to whom the blood sample belongs and the medical facility from which the blood sample came.

In a particular case, the ring LED white light source is designed with an illumination of 1400 Lux.

In a particular case, the ring LED white light source is equipped with a diffuser.

In a particular case, the ring LED white light source has a power of no more than 20 W.

In a particular case, the ring LED white light source is made with a ring diameter of 20 to 40 cm.

In a particular case, the control and information processing system is either a desktop computer, a laptop, or a smartphone.

In a particular case, the first and second classification models are recorded in the permanent memory of the control and information processing system.

In a particular case, the first and second classification models are recorded on a remote server, in a cloud storage, and the information management and processing system is configured to temporarily load classification model data into its memory.

The technical result is also achieved by a method for automatically determining the presence of hemolysis and/or chylosis in centrifuged blood samples, including the steps:

1. image acquisition by cameras of the mobile removable complex described above;

2. detecting the presence of a test tube in the image using the first classification model;

3. classification of plasma type: if the first classification model detects the presence of a test tube with blood in the image, the work of the second classification model is launched, which simultaneously searches for image areas corresponding to plasma and classifies the plasma type;

4. recognition of a barcode located on a test tube with blood using a second classification model;

5. displaying on the device monitor the results of determining the presence of hemolysis and/or chylosis in blood samples.

If there is no tube in the image or if the image quality does not allow the tube to be detected, a message may be displayed on the monitor stating that the tube was not detected.

If there is no barcode or if it cannot be determined, a message may be displayed on the monitor stating that the barcode is not recognized.

In a particular case, the device monitor displays the results of determining the presence of hemolysis and/or chylosis in blood samples with information about the patient to whom the blood sample belongs and the medical institution from which the blood sample was received.

In a particular case, to determine the degree of chylosis and/or hemolysis, video recordings obtained from the device’s cameras are used, with tubes with blood samples present in the frames.

In a particular case, the first classification model is a convolutional neural network with MobileNetV3 architecture.

In a particular case, the second classification model is a convolutional neural network with the YOLOvS architecture.

The above aspects of the technical result are achieved: firstly, thanks to the design features of the created complex, which provide the ability to fix and remove the device for determining the presence of hemolysis and/or chylosis in blood samples and the compatibility of the device with any automated laboratory system; secondly, thanks to a method for determining the presence of hemolysis and/or chylosis in blood samples with at least 2 cameras with a resolution of 1920x1080, which ensures the most informative images of blood samples.

Brief description of the drawings.

The invention is illustrated by illustrative material.

Fig. 1 A) - schematic representation of the conveyor line and the location of the device chambers relative to the test tubes with samples, where 1 - conveyor line (top view), 2 - test tubes with samples, 3 - device chambers.

Fig. 1 B) - an example of an image of a conveyor line as seen from the camera of the device, located on top, above the test tubes, where 2 is a test tube with a sample, 4 is a sample holder.

Fig. 1 B) is an example of an image of a conveyor line obtained from a device camera located on the side of the sample, where 5 is a side window for shooting, 2 is a test tube with a sample, 4 is a sample holder.

Fig. 2 A, B) - examples of images with a clearly visible (A) and indistinguishable barcode (B) for cameras located opposite the side observation windows.

Fig. 3 A) is an example of images from a Logitech HD Webcam C270 camera with a resolution of 1280×720. Fig. 3 B) - example of an image from a Logitech C922 Pro 1920×1080 camera. Exposure 30ms, additional lighting used

Fig. 4 A) - a schematic representation of the arrangement of the chambers of a device with four chambers relative to the test tubes from the conveyor line (top view), where 1 is the conveyor line, 2 is test tubes with samples, 3 is the device chambers.

Fig. 4 B) - photograph of a conveyor line with a passing blood tube, obtained from the device’s cameras. The numbers on the images reflect the camera numbering (K1-K4) presented in panel A.

Fig. 4 B) - image of the installation in the configuration of four cameras on a conveyor, where 1 - conveyor line, 3 - device cameras (additionally marked with yellow frames), 6 - control computer, 7 - source of additional lighting for the device.

Fig. 5 - the result of the device in predicting the classes of the test tube: (A) - “bar code is visible”, (B) - “chylosis”, (C) - “hemolysis”, (D) - “norm”.

Carrying out the invention

1. Testing the design of device prototypes.

As a result of the experiments, several experimental samples of the device (prototypes) were constructed that allow photographing test tubes with centrifuged blood moving along a conveyor line from several viewing angles in different lighting conditions with different exposure times and number of frames per second:

- prototype with two cameras;

- prototype with four cameras.

A compact design for off-line applications is also possible, implementing the single-chamber method. In this case, it will be necessary to first read the barcode by turning the tube with the barcode towards the camera, and then turn the tube to the area where the blood plasma section is visible. Thus, when using a single-chamber design, automatic detection of the presence of hemolysis and/or chylosis is not realized. In addition, when feeding tubes manually, sufficient image clarity is not achieved.

Prototypes with two and four cameras were tested on a conveyor line at the State Clinical Hospital No. 67 named after. L.A. Vorokhobova - Beckman Coulter Power Express, in order to identify their shortcomings and select the optimal configuration for further work.

First of all, the solution was tested using two CMOS RGB cameras, with a maximum digital resolution of 1280x720, a minimum exposure of 30 ms, a viewing angle of 55°, and a frame rate per second of 30 Hz (Logitech C270 HD model). The cameras were located on the sides of the conveyor and were focused on the tubes through observation windows. In FIG. 1A shows a diagram of the location of the cameras relative to the conveyor line.

Next, using an additional lighting source, statistics were collected on ~300 tubes using two cameras located on the sides of the conveyor, on the basis of which the proportion of tubes with a barcode located outside the camera’s visibility range was manually assessed (Fig. 2A, B ). It was found that the proportion of tubes with an indistinguishable barcode in the two-chamber design was ~42±3%.

Additionally, the sufficiency of the camera resolution was checked: Logitech C270 HD cameras (camera model 1, resolution 1280×720, exposure value 30 ms, 30 frames per second) and Logitech C922 HD Pro (camera model 2, digital resolution 1920×1080) were tested , exposure duration 30 ms, number of frames per second - 30, viewing angle 78°). Using an algorithm from the pyzbar library, the ability to read barcodes from images was assessed; it was determined that using a camera with a resolution of less than 1920×1080 pixels it is impossible to obtain images of barcodes (Fig. 3A, B shows representative frames obtained using the specified camera models).

To eliminate the shortcomings of the two-camera scheme (lack of barcodes on some of the images of the test tubes, low resolution, which does not allow capturing some of the barcode lines), it is proposed to use cameras with a resolution of at least 1920×1080 pixels. Also, installing two additional cameras with a resolution of at least 1920×1080 pixels allows you to further minimize the number of tubes with unrecognized barcodes (see Fig. 4).

To fix the position of the cameras relative to each other and the illuminator, a design based on optomechanical holders and a tripod was developed for the first prototype of the device, which allows fixing and changing the vertical position of the cameras (a general view of the installation is shown in Fig. 4B).

This design provides for the installation of at least 2 digital short-throw cameras with a resolution of at least 1920x1080 pixels and a ring LED white light source on a conveyor line using a tripod, a computer, a Wi-Fi transmitter, a display and a cable for connecting the network.

In this case, the tripod elements ensure fixation of the light source and cameras in a horizontal plane relative to the conveyor line of the laboratory analyzer. Tripod holders fix the cameras and provide the ability to change the vertical position of the cameras. Removable tripod mounting elements allow you to easily attach and remove the tripod with cameras and light source from the laboratory analyzer conveyor line. Also, if the conveyor line does not have a window analyzer, the removable tripod mounting elements allow you to fix the tripod inside the conveyor line.

Since heating blood tubes can render the biomaterial unusable, a light source that does not heat the sample is required. Therefore, the power of the light source should be no more than 20 W.

Images obtained under fixed lighting conditions, achieved through the use of an additional light source, as well as the use of a fixed camera geometry provided by a tripod, make it possible to achieve high repeatability of the color characteristics of blood plasma samples recorded using digital cameras. High repeatability of color characteristics makes it possible to determine the presence of hemolysis and/or chylosis in samples with high reliability and reliability.

2. Selecting computer vision models for classifying blood samples.

Tubes with samples of centrifuged blood, detected by cameras located in the design used, can be located in different parts of the image, and in some cases the area of the tube with blood plasma can be hidden by a barcode used to identify the sample. Also, the image may simultaneously contain several test tubes with different classes of suitability of blood plasma for research. In this regard, to correctly identify samples in images, a computer vision model can be used, which allows one to simultaneously detect - determine the position of an object in the image - and classify objects located in the detected areas. One of the most common models with high classification accuracy are models based on the use of convolutional neural networks with the YOLO architecture (Diwan, T., Anirudh, G., & Tembhurne, J.V. (2022). Object detection using YOLO: challenges, architectural successors , datasets and applications. Multimedia Tools and Applications, 1-33). The reasons for choosing the YOLOv5 model also include deployment flexibility, high speed and high accuracy of object detection using this architecture.

To detect and classify blood samples in test tubes, five modifications of the YOLO v5 model were trained and tested: YOLOv5n, YOLOv5s, YOLOv5m, YOLOv51 and YOLOv5x. The main fundamental architectural features of these models are identical, and the differences lie in the depth of the models, which affect the final accuracy of data processing. Five models were trained to identify the option whose characteristics would best match the planned performance parameters.

The development and initial training of this model for classifying blood samples included data ordering (data marking), which was performed by the team of City Clinical Hospital No. 67 named after. L.A. Vorokhobov for a set of training samples: the final dataset consisted of 14,100 images, which were divided into training (13,403 images) and validation (697 images) samples.

To run a YOLO v5-based model that would detect a tube in a frame and classify the type of blood plasma in it, it was necessary to train a version of the YOLO v5 algorithm, pre-tuned on a dataset of the COCO open library with predefined parameters: source data size, batch, learning rate (see Table 1).

In accordance with the characteristics of the image and the GPU resources available at this stage, the best effect from data training was achieved with a batch of 32 elements. To select the main base model, 5 YOLO v5 models were trained and tested, the descriptions of which are presented in Table 2.

Correct training of models on a properly balanced dataset, as well as the choice of model hyperparameters (number of training epochs), architecture (including the number of model parameters), learning rate parameter values, input data size, batch size, etc. is a key stage in model training , which recognizes the class of the test tube. The collected dataset included a large number of frames of various test tubes of samples suitable for analysis (“normal” samples), samples with severe chylosis, hemolysis, and chylosis and hemolysis. The training parameters were chosen so that the classification accuracy was at least 0.85.

3. Preparation of data for training computer vision models to classify samples into hemolysis, chylosis and normal types.

Using a device created by a team of applicants, working using 4 cameras with the following shooting parameters: digital resolution 1920×1080, the necessary data set (dataset) of 14,100 images of blood tubes was collected. For the model to detect and classify blood samples to work correctly, standardization of the content of image annotations and their formats was required. This algorithm for standardizing input data included the following steps:

- distribution of frames with test tubes into folders depending on the type of plasma;

- class balancing;

- transfer of initial data to the program for marking photographs Supervise.ly, selecting a detection area for test tubes and assigning samples to one of 5 classes;

- downloading data from Supervise.ly;

- converting data into the format necessary for training the model.

To train the models, samples were pre-marked by a qualified specialist to detect chylosis and hemolysis based on visual inspection. Images from 4 camera viewing angles were structured and divided into 5 classes - “chylosis”, “hemolysis”, “normal”, “plasma color not visible” and “barcodes”.

In order for the machine learning model used for the project to generalize well to the classes involved in training, it was necessary to equalize the sample sizes for each class. In cases of unbalanced samples with insufficient numbers of rarer samples, the model will ignore the smaller class and misclassify it, which will reduce the accuracy and quality of the training results. To prevent a decrease in the quality of training, before loading the data into Supervise.ly, quantitative balancing of the samples of each class was carried out.

Of the entire prepared sample, 95% were used to train the object detection algorithm, the remaining 5% were used to validate the operation of the algorithm. The presence in the sample of images in which the contents of the sample are covered with a barcode increases the accuracy of detection and classification of tubes: this is due to the fact that without this class, such tubes would be classified as chylous, hemolytic or normal, which would reduce the overall efficiency of work models. Background images are photographs without test tubes and without a class and are added to the training set to reduce the number of false positives (False Positive), which leads to an increase in the overall accuracy (precision) of the model.

The data obtained as a result of the work was used by the team to train computer vision models of the YOLO v5 generation.

4. Evaluation of research results.

Models for classifying blood samples were verified and the sensitivity and specificity of the model working in conjunction with the device’s digital cameras was determined. The sensitivity and specificity values of the basic YOLO v5 models were tested on the validation dataset in order to identify the most effective of this series of models for further work to improve and maximize its effectiveness. In this regard, to test the developed basic YOLO v5 models, five indicators were adopted in this study:

1. Precision. The target value is at least 85%, where: Precision = True positive / (True positive+False positive). Calculation formula 1:

where TP is the number of true positive cases, FP is the number of false positive cases.

2. IoU (Intersection over Union score) - metric for segmentation (detection) of a test tube in the image. The target value is at least 60%. Calculation formula 2:

where R is the area of the object boundary determined by the model; R' - true area of the object boundary

3. Response (Recall). The target value is at least 95%, where: Recall = True positive / (True positive + False negative). Calculation formula:

where TP is the number of true positive cases, FT is the number of false positive cases, FN is the number of false negative cases.

4. Classification speed for one blood sample in milliseconds per 1 video frame.

5. mAP (mean Average Precision). Average AP value (Average Precision) when detecting a blood tube (the higher the mAP, the better the test tube detection result). Calculation formula:

where C is the number of classes.

The effectiveness of the models for classifying blood samples in test tubes was tested on 697 images of the validation set, the results are presented in Table 4.

Thus, as a result of the experiments, it was found that of the five versions of the models, the YOLOv51 version of the model demonstrated the best results in accuracy, response and tAP. The average accuracy of this model was 98.95%: the model correctly classified images with chylosis in 100% of cases, with hemolysis - in 98% of cases, with normal blood plasma - in 100% of cases, with barcodes - in 100% of cases, with images where plasma is not visible - in 98% of cases.

Examples of implementation of the invention.

Example 1.

The performance of the invention was verified on the basis of a conveyor system for biochemical analysis of blood plasma samples at the clinical diagnostic laboratory of City Clinical Hospital No. 67 named after. L.A. Vorokhobova.

To process the results obtained using the device, a client computer was prepared and used to use and test a prototype ASUS UX430UAR PAK with the main characteristics:

- processor: Intel Core i7-8550U, CPU @ 1.80 GHz 1.99 GHz;

- RAM: 16 GB;

- operating system type: 64-bit Windows 11 Pro operating system;

- monitor: 1920×1080 pixels;

- pen and touch input: keyboard, touch input support (touch points: 2);

- Internet browser: Google Chrome 98.0.4758.102.

To process images received from the device’s cameras, a cloud server with the following characteristics was used:

server with the following characteristics:

- number of processor cores: 4;

- RAM: 8 GB;

- data storage: 60 GB SSD;

- operating system type: Ubuntu 20.04.

To test the prototype device, blood plasma samples belonging to various classes - “normal”, “hemolysis” and “chylosis”, as well as samples in which the blood plasma did not fall into the field of view of one or more chambers of the device were run along the conveyor line. At the same time, the “true value” of the class of belonging of the test tube was assessed visually by a laboratory assistant at the clinical diagnostic laboratory of City Clinical Hospital No. 67.

To assess the accuracy of the device and method, the following indicators were used:

1. Accuracy (Precision), where: Precision = True positive / (True positive + False positive).

2. IoU (Intersection over Union score) - test tube segmentation metric in the image. The target value is at least 60%.

3. Response (Recall), where: Recall = True positive / (True positive + False negative).

4. Classification speed for one blood sample in milliseconds per 1 frame of video: no more than 10 seconds.

5. mAP (mean Average Precision). Average AP value (Average Precision) when detecting a blood tube (the higher the mAP, the better the test tube detection result).

The final verification of the effectiveness of the device and the classification models of blood samples used in the device for biochemical analyzes (laboratory and clinical diagnostics) took place on 2000 images of samples, balanced in accuracy. Testing the classification of blood tubes using the trained YOLOv51 model demonstrated the following results:

- accuracy: 98.85%;

- IoU: 97.99%;

- response: 97.94%;

- classification speed: 25-30 milliseconds;

- mAP 0.5:0.95: 66%.

In FIG. Figure 5 shows examples of images obtained using the device, which show different classes of tubes. Fig. 5 (A) demonstrates cases when blood plasma is not visible in the images and the model correctly reflects this fact. In FIG. Figure 5 (B-D) shows cases of blood plasma tubes with severe chylosis (Fig. 5B), with severe hemolysis (Fig. 5C) and normal (Fig. 5D). For these classes, the sensitivity and specificity of the classification ranged from 95 to 98%, which demonstrates the high degree of reliability of the proposed device and method.

Claims (26)

1. Mobile removable complex for automatic determination of the presence of hemolysis and/or chylosis in centrifuged blood samples, including : - at least 2 digital short-focus cameras with a resolution of at least 1920x1080 pixels; - ring LED white light source; - a tripod made with the possibility of fixing to the conveyor line of a laboratory analyzer of digital short-focus cameras and a ring LED source of white light in a horizontal plane relative to the conveyor line of the laboratory analyzer, consisting of a frame, holders made with the possibility of fixing and changing the vertical position of the cameras, and removable fastening elements , made with the ability to fix and remove the stand from the conveyor line of the laboratory analyzer, in this case, the removable fastening elements are made either with the possibility of installing a tripod outside the conveyor line of the laboratory analyzer, so that short-focus cameras record samples through the side windows of the conveyor line of the laboratory analyzer, or are designed with the ability to install short-focus cameras inside the conveyor line of a laboratory analyzer; - a control and information processing system capable of analyzing images obtained from short-focus cameras using the first and second classification models, and configured to record, process, store and display on a monitor the results of determining the presence of hemolysis and/or chylosis in plasma samples blood with information about the patient to whom the blood sample belongs and the health care facility from which the blood sample came. 2. The mobile removable complex according to claim 1, characterized in that the ring LED white light source is made with an illumination of 1400 Lux. 3. Mobile removable complex according to claim 1, characterized in that the ring LED white light source is equipped with a diffuser. 4. Mobile removable complex according to claim 1, characterized in that the ring LED white light source is made with a power of no more than 20 W. 5. Mobile removable complex according to claim 1, characterized in that the ring LED white light source is made with a ring diameter of 20 to 40 cm. 6. Mobile removable complex according to claim 1, characterized in that the control and information processing system is either a desktop computer, a laptop, or a smartphone. 7. Mobile removable complex according to claim 1, characterized in that the first and second classification models are recorded in the permanent memory of the control and information processing system. 8. Mobile removable complex according to claim 1, characterized in that the first and second classification models are recorded on a remote server, in a cloud storage, and the control and information processing system is designed with the ability to temporarily load classification model data into its memory. 9. A method for automatically determining the presence of hemolysis and/or chylosis in centrifuged blood samples, including the steps: - image acquisition by cameras of the mobile removable complex according to claims. 1-8; - detecting the presence of a test tube in the image using the first classification model; - classification of plasma type: if the first classification model detects the presence of a test tube with blood in the image, the work of the second classification model is launched, which simultaneously searches for image areas corresponding to plasma and classifies the plasma type; - recognition of a barcode located on a test tube with blood using a second classification model; - displaying on the device monitor the results of determining the presence of hemolysis and/or chylosis in blood samples. 10. The method according to claim 9, characterized in that if there is no test tube in the image or if the image quality does not allow the test tube to be detected, a message is displayed on the monitor stating that the test tube was not detected. 11. The method according to claim 9, characterized in that if there is no barcode or if it cannot be determined, a message is displayed on the monitor stating that the barcode is not recognized. 12. The method according to claim 9, characterized in that the device monitor displays the results of determining the presence of hemolysis and/or chylosis in blood samples with information about the patient to whom the blood sample belongs and the medical institution from which the blood sample was received. 13. The method according to claim 9, characterized in that to determine the degree of chyle and/or hemolysis, video recordings obtained from cameras of the removable complex according to claims are used. 1-8, with tubes with blood samples present in the frames. 14. The method according to claim 9, characterized in that the first classification model is a convolutional neural network with MobileNetV3 architecture. 15. The method according to claim 9, characterized in that the second classification model is a convolutional neural network with YOLOv5 architecture.