CN111524122B - Method for constructing gauze blood-soaking amount estimation model based on characteristic engineering - Google Patents

本发明涉及一种基于特征工程的纱布浸血量估算模型构建方法，属于人工智能领域。该方法包括步骤：S1采集包含血量标注的浸血纱布图像，构建数据集；S2图像预处理，包括图像尺寸规整和色彩空间转换；S3图像浸血区域掩模提取，并获得图像浸血区域；S4提取浸血纱布图像特征，包括浸血区域血红蛋白量、HSV色彩空间各通道的均值和方差共14个特征，并进一步构建图像特征集；S5基于步骤S4中所构建的图像特征集，构建纱布浸血量估算的机器学习模型。本发明基于特征工程，通过所构建的模型能够较为快速准确的估计病人的术中失血量。

The invention relates to a method for constructing an estimation model for gauze soaked blood volume based on feature engineering, and belongs to the field of artificial intelligence. The method includes the steps of: S1 collecting a blood-soaked gauze image including blood volume annotation, and constructing a data set; S2 image preprocessing, including image size regularization and color space conversion; S3 image blood-soaked area mask extraction, and obtaining the image blood-soaked area ; S4 extracts the image features of the blood-soaked gauze, including 14 features including the amount of hemoglobin in the blood-soaked area, the mean and variance of each channel of the HSV color space, and further constructs an image feature set; S5 constructs an image feature set based on the image feature set constructed in step S4. A machine learning model for estimation of blood soaked in gauze. The present invention is based on feature engineering, and the constructed model can rapidly and accurately estimate the intraoperative blood loss of a patient.

Method for constructing gauze blood-soaking amount estimation model based on characteristic engineering

Technical Field

The invention belongs to the field of artificial intelligence, and relates to a method for constructing a gauze blood-soaking amount estimation model based on characteristic engineering.

Background

With the development of surgical techniques, there are about 3 hundred million surgical operations per year, with a post-operative complication rate of about 16.7%, and a mortality rate of about 0.5% of the work of anesthesiologists being critical in surgical operations, the most important and difficult task being to continuously monitor and assess blood loss during the operation, which can guide not only the first blood transfusion but also the second. If the intraoperative blood loss is underestimated, transfusion failure and postoperative complications may result. Blood loss during overestimation can lead to over-monitoring, transfusion-related complications and waste of blood products.

Intraoperative blood loss refers to a reduction in circulating blood volume, wherein the lost circulating blood volume includes the invisible components: blood (plasma) and visible components (mainly red blood cells). Intraoperative blood loss mainly includes bleeding or oozing from surgical platforms, blood content in gauze, aspirators and sterile towels. The biggest challenge in continuous intraoperative monitoring of blood loss is estimating the blood content in the gauze.

The main methods of estimating the blood absorbed by gauze include a visual evaluation method and a weighing method. Visual assessment method blood loss was estimated by visually measuring the area of blood on different sizes of surgical gauze. This method is the most commonly used method for clinical assessment of blood loss, but it is less accurate. When the blood loss is less than 150ml, it is easy to overestimate the blood loss. However, when the amount of blood lost is more than 300ml, the amount of blood lost is easily underestimated, and the more blood lost, the less apt to underestimate blood lost. The above method relates only to the amount of blood and does not take into account the different haemoglobin concentrations of the different patients and the dilution of the blood with saline during the operation, so that the capacity of this method to estimate the amount of visible components in the blood-soaked gauze is limited. The weighing method is relatively accurate, but the operation is complex, generally performed after the operation, and is inconvenient for quick intraoperative evaluation. Meanwhile, the weighing method needs the instrument nurse and the surgeon to accurately calculate the irrigation amount of the gauze, which has certain limitation in clinical application.

At present, artificial intelligence technology has been widely studied and applied in the medical field. In order to overcome the defects of the current method for evaluating the blood volume of blood soaked gauze to assist anaesthetists, a novel method based on characteristic engineering is provided. The method utilizes an image processing technology and hemoglobin data of a patient to extract key characteristics of the blood-soaked gauze image, and combines a characteristic engineering method to construct a model, so that the rapid and accurate estimation of the blood-soaked amount of the gauze can be realized.

Disclosure of Invention

In view of this, the present invention aims to provide a method for constructing a gauze blood-soaking amount estimation model based on feature engineering. The method utilizes the characteristic engineering to extract the key characteristics in the blood soaked gauze image, constructs a gauze blood soaking amount estimation model through a machine learning algorithm, and can realize the rapid and accurate estimation of the gauze blood volume in the operation process.

In order to achieve the purpose, the invention provides the following technical scheme:

a method for constructing a gauze blood-soaking amount estimation model based on feature engineering comprises the following steps:

s1: collecting a blood soaking gauze image containing blood volume labels, and constructing a data set;

s2: image preprocessing, including image size normalization and color space conversion;

s3: extracting a mask of the image blood soaking area, and obtaining the image blood soaking area;

s4: extracting blood-soaked gauze image characteristics, including 14 characteristics of hemoglobin amount in a blood-soaked area and mean value and variance of each channel in HSV color space, and constructing an image characteristic set;

s5: based on the set of image features constructed in step S4, a machine learning model of the gauze bleeding amount estimation is constructed.

Optionally, in step S1, when the blood-soaked gauze image is finished, the gauze is spread flatly, and the blood-soaked gauzes are photographed one by one.

Optionally, in the step S2, the size of the captured image is adjusted to 480 × 480 pixels, and the adjusted image is converted from the RGB color space to the HSV color space.

Optionally, in step S3, an image blood-soaking area mask is extracted according to the H value in the HSV color space, and two masks are specifically defined as:

where (i, j) is the pixel position of the blood-soaked image in the RGB color space.

Optionally, in the step S3, two blood-soaked areas of the blood-soaked gauze image are obtained by using a mask, which is defined as:

wherein (i, j) is the pixel position of the blood-soaked image in RGB color space, B_i,jAnd (3) obtaining HSV pixel vector values of the positions of the original blood-soaked gauze pictures (i, j).

Optionally, in step S4, the hemoglobin amount is obtained by multiplying a ratio of hemoglobin concentration to blood-immersed region area, and the specific steps include:

(1) calculating the number of pixels under the two blood soaking area masks, and respectively recording as PR1_numAnd PR2_num；

(2) The ratio of the blood-soaked area under the two masks is calculated respectively, namely the ratio of the whole image is calculated:

(3) normalization treatment of the patient hemoglobin concentration:

wherein Hbc represents the hemoglobin concentration of a current single patient, and Max and Min represent the maximum value and the minimum value of the hemoglobin concentrations of all patients respectively;

(4) the hemoglobin amount of the blood-soaked area under the two masks is respectively calculated and is defined as the product of the area ratio of the blood-soaked area and the normalized hemoglobin concentration:

Hgb1＝Hbc×AR1，

Hgb2＝Hbc×AR2。

optionally, in the step S4, for each blood-soaked gauze image, a mean and a variance of each channel of the blood-soaked area in the HSV color space are calculated, and since there are two blood-soaked area masks, mean and variance features are calculated for the two generated blood-soaked areas respectively; these features are noted as: h1_ mean, H1_ std, S1_ mean, S1_ std, V1_ mean, V1_ std, H2_ mean, H2_ std, S2_ mean, S2_ std, V2_ mean, V2_ std, 12 features in total.

Optionally, in the step 4, in each blood-soaked gauze image, 14 features are extracted in total, including the features Hgb1 and Hgb2, and the 12 features; extracting features from all pictures in the data set constructed in the step S1 to form an image feature set; and serially constructing the image feature set or parallelly constructing the image feature set.

Optionally, in the step 5, the gauze blood-soaking amount estimation machine learning model is a multiple linear regression model constructed on the constructed image feature set.

The invention has the beneficial effects that: according to the method for constructing the gauze blood immersion amount estimation model, the blood immersion gauze image is collected and marked, the gauze blood immersion amount is constructed, a training data set is constructed, the image size is regulated through image preprocessing, color space conversion is carried out, image characteristics are extracted based on characteristic engineering and combined with hemoglobin concentration information of a patient, a multiple linear regression gauze blood immersion amount estimation model is constructed by taking a quadratic error function as a loss function, the gauze blood immersion amount can be quickly and accurately estimated, accurate reference is provided for anesthesiologists, and the safety of the patient is improved.

Additional advantages, objects, and features of the invention will be set forth in part in the description which follows and in part will become apparent to those having ordinary skill in the art upon examination of the following or may be learned from practice of the invention. The objectives and other advantages of the invention may be realized and attained by the means of the instrumentalities and combinations particularly pointed out hereinafter.

Drawings

For the purposes of promoting a better understanding of the objects, aspects and advantages of the invention, reference will now be made to the following detailed description taken in conjunction with the accompanying drawings in which:

FIG. 1 is a schematic flow chart of a method for constructing a gauze blood-soaking amount estimation model based on feature engineering;

FIG. 2 is a schematic diagram of an image of blood-soaked gauze according to an embodiment;

FIG. 3 is a schematic diagram showing an image of a blood-soaked area in the embodiment;

FIG. 4 is a flowchart of an embodiment.

Detailed Description

The embodiments of the present invention are described below with reference to specific embodiments, and other advantages and effects of the present invention will be easily understood by those skilled in the art from the disclosure of the present specification. The invention is capable of other and different embodiments and of being practiced or of being carried out in various ways, and its several details are capable of modification in various respects, all without departing from the spirit and scope of the present invention. It should be noted that the drawings provided in the following embodiments are only for illustrating the basic idea of the present invention in a schematic way, and the features in the following embodiments and examples may be combined with each other without conflict.

Wherein the showings are for the purpose of illustrating the invention only and not for the purpose of limiting the same, and in which there is shown by way of illustration only and not in the drawings in which there is no intention to limit the invention thereto; to better illustrate the embodiments of the present invention, some parts of the drawings may be omitted, enlarged or reduced, and do not represent the size of an actual product; it will be understood by those skilled in the art that certain well-known structures in the drawings and descriptions thereof may be omitted.

The same or similar reference numerals in the drawings of the embodiments of the present invention correspond to the same or similar components; in the description of the present invention, it should be understood that if there is an orientation or positional relationship indicated by terms such as "upper", "lower", "left", "right", "front", "rear", etc., based on the orientation or positional relationship shown in the drawings, it is only for convenience of description and simplification of description, but it is not an indication or suggestion that the referred device or element must have a specific orientation, be constructed in a specific orientation, and be operated, and therefore, the terms describing the positional relationship in the drawings are only used for illustrative purposes, and are not to be construed as limiting the present invention, and the specific meaning of the terms may be understood by those skilled in the art according to specific situations.

As shown in fig. 1, a method for constructing a gauze blood-soaking amount estimation model based on feature engineering includes the following steps:

s1, collecting blood soaking gauze images containing blood volume labels to construct a data set;

s2, preprocessing the image, including image size regulation and color space conversion;

s3, performing mask extraction on the image blood soaking area, and obtaining the image blood soaking area;

s4, extracting blood-soaked gauze image features, including 14 features of hemoglobin amount in a blood-soaked area and mean value and variance of each channel in HSV color space, and further constructing an image feature set;

s5 constructs a machine learning model of the gauze bleeding amount estimation based on the set of image features constructed in step S4.

Further, in step S1, when the operation is finished, the blood-soaked gauze image is obtained by spreading the gauze flatly and shooting the blood-soaked gauze one by one.

Further, in step S2, the size of the captured image is adjusted to 480 × 480 pixels, and the adjusted image is converted from the RGB color space to the HSV color space.

Further, in step S3, extracting an image blood-soaked area mask according to the H value in the HSV color space, and the two masks are specifically defined as:

where (i, j) is the pixel position of the blood-soaked image in the RGB color space.

Further, in the step S3, two blood-soaked areas of the blood-soaked gauze image are obtained by using the mask, which are defined as:

wherein (i, j) is the pixel position of the blood-soaked image in RGB color space, B_i,jAnd (3) obtaining HSV pixel vector values of the positions of the original blood-soaked gauze pictures (i, j). In practice, BR1 and BR2 are two blood-soaked images obtained from the original blood-soaked gauze image, each image having only red and black, red indicating the blood-soaked area and black indicating the non-blood-soaked area.

Further, in step S4, the hemoglobin amount is obtained by multiplying the ratio of the hemoglobin concentration to the blood-immersed region area, and the specific steps include:

(1) calculating the number of pixels under the two blood soaking area masks, and respectively recording as PR1_numAnd PR2_num；

(2) The ratio of the blood-soaked area under the two masks is calculated respectively, namely the ratio of the whole image is calculated:

(3) normalization treatment of the patient hemoglobin concentration:

wherein Hbc represents the hemoglobin concentration of a current single patient, and Max and Min represent the maximum value and the minimum value of the hemoglobin concentrations of all patients respectively;

(4) the hemoglobin amount of the blood-soaked area under the two masks is respectively calculated and is defined as the product of the area ratio of the blood-soaked area and the normalized hemoglobin concentration:

Hgb1＝Hbc×AR1，

Hgb2＝Hbc×AR2。

further, in the step S4, for each blood-soaked gauze image, the mean and variance of each channel in the HSV color space of the blood-soaked area are calculated, and since there are two blood-soaked area masks, the mean and variance features are calculated for the two generated blood-soaked areas respectively. These features are noted as: h1_ mean, H1_ std, S1_ mean, S1_ std, V1_ mean, V1_ std, H2_ mean, H2_ std, S2_ mean, S2_ std, V2_ mean, V2_ std, 12 features in total.

Further, in the step 4, a total of 14 features are extracted from each blood-soaked gauze image, including the features Hgb1 and Hgb2 in claim 6 and 12 features in claim 7. Features are extracted from all pictures in the data set constructed in step S1 to form an image feature set. The image feature set may be constructed serially or in parallel.

Further, in the step S5, the gauze blood immersion amount estimation machine learning model is a multiple linear regression model constructed on the image feature set constructed in the step S4. The multiple linear regression model is defined as:

y_est＝β₀+β₁×f₁+β₂×f₂+…+β₁₄×f₁₄，

wherein, y_estFor blood volume estimation, beta₀,β₁,…,β₁₄Is a parameter, f₁,f₂,…,f₁₄Is the characteristic of the blood-soaked gauze image. The model solution uses a least square method to optimize the following loss functions:

wherein n is the number of images in the training set.

The preferred embodiment of the present invention will be described in detail with reference to fig. 1.

Example (b):

a method for constructing a gauze blood-soaking amount estimation model based on characteristic engineering is adopted to construct the gauze blood-soaking amount estimation model, and comprises the following steps:

step one, with reference to fig. 2, firstly resizing a 3968 × 2976 × 3 original image matrix to 480 × 480 × 3, and then converting the resized RGB image into an HSV space.

Step two, firstly dividing a blood soaking area and a non-blood soaking area according to H, S, V values in an HSV space, wherein the blood soaking area is divided into two parts:

two masks (the blood-immersed region is assigned as "255" and the non-blood-immersed region is assigned as "0") are obtained by the above two divisions, and the ratio of the blood-immersed region is calculated using the value of the H space.

And step three, obtaining two blood soaking area images by using the mask obtained in the HSV space, as shown in figure 3.

And step four, respectively calculating the mean value and the variance of the H, S, V channels of the two blood soaking area images in the HSV space, wherein the total number of the features is 12.

And step six, executing the step one to the step five to all the images in the training set, wherein each image obtains 14 features to form an image feature set. And the processing can be carried out in series or in parallel.

And seventhly, training a multiple linear regression model by using the image feature set obtained in the sixth step aiming at the following quadratic loss function:

the trained model can be used for estimating the blood soaking amount of the gauze.

The flow chart of the embodiment is shown in fig. 4.

Finally, the above embodiments are only intended to illustrate the technical solutions of the present invention and not to limit the present invention, and although the present invention has been described in detail with reference to the preferred embodiments, it will be understood by those skilled in the art that modifications or equivalent substitutions may be made on the technical solutions of the present invention without departing from the spirit and scope of the technical solutions, and all of them should be covered by the claims of the present invention.

1.一种基于特征工程的纱布浸血量估算模型构建方法，其特征在于：该方法包括以下步骤：1. a method for constructing a gauze soaked blood volume estimation model based on feature engineering, is characterized in that: the method comprises the following steps: S1：采集包含血量标注的浸血纱布图像，构建数据集；S1: Collect blood-soaked gauze images containing blood volume annotations to construct a data set; S2：图像预处理，包括图像尺寸规整和色彩空间转换；S2: Image preprocessing, including image size regularization and color space conversion; S3：图像浸血区域掩模提取，并获得图像浸血区域；具体步骤为：根据HSV色彩空间中的H值提取图像浸血区域掩模，共两个掩模；利用掩模得到浸血纱布图像的两个浸血区域；S3: Extract the blood-soaked area mask of the image, and obtain the blood-soaked area of the image; the specific steps are: extracting the blood-soaked area mask of the image according to the H value in the HSV color space, a total of two masks; using the mask to obtain the blood-soaked gauze two blood-soaked areas of the image; S4：提取浸血纱布图像特征，针对每一个浸血区域，提取图像特征：包括浸血区域血红蛋白量、HSV色彩空间各通道的均值和方差，共14个特征，S4: Extract the image features of the blood-soaked gauze. For each blood-soaked area, extract the image features: including the amount of hemoglobin in the blood-soaked area, the mean and variance of each channel in the HSV color space, a total of 14 features, S5：基于步骤S4中所构建的图像特征集，构建纱布浸血量估算的机器学习模型。S5: Based on the image feature set constructed in step S4, construct a machine learning model for estimating the amount of blood soaked in gauze. 2.根据权利要求1所述的一种基于特征工程的纱布浸血量估算模型构建方法，其特征在于：在所述步骤S1中，浸血纱布图像在手术结束时，将纱布平整展开，逐张浸血纱布统一拍摄。2. A method for constructing a blood-soaked gauze volume estimation model based on feature engineering according to claim 1, characterized in that: in the step S1, when the blood-soaked gauze image is at the end of the operation, the gauze is flattened and unfolded one by one. Zhang blood soaked gauze uniform shooting. 3.根据权利要求1所述的一种基于特征工程的纱布浸血量估算模型构建方法，其特征在于：在所述步骤S2中，将所拍摄图像尺寸规整到480x480像素大小，并将尺寸规整后的图像从RGB色彩空间转换到HSV色彩空间。3. a kind of gauze blood soaking amount estimation model construction method based on feature engineering according to claim 1, is characterized in that: in described step S2, the size of the photographed image is regulated to 480x480 pixel size, and the size is regulated The resulting image is converted from RGB color space to HSV color space. 4.根据权利要求1所述的一种基于特征工程的纱布浸血量估算模型构建方法，其特征在于：在所述步骤S3中，根据HSV色彩空间中的H值提取图像浸血区域掩模，共两个掩模，具体定义为：4. A method for constructing a model for estimating blood soaking in gauze based on feature engineering according to claim 1, wherein in the step S3, an image blood soaking area mask is extracted according to the H value in the HSV color space , a total of two masks, specifically defined as: 第一个掩模

first mask

第二个掩模

second mask

其中，(i,j)为浸血图像在RGB色彩空间中的像素位置。where (i,j) is the pixel position of the blood-soaked image in the RGB color space. 5.根据权利要求4所述的一种基于特征工程的纱布浸血量估算模型构建方法，其特征在于：在所述步骤S3中，利用掩模得到浸血纱布图像的两个浸血区域，定义为：5. A method for constructing a blood-soaked gauze volume estimation model based on feature engineering according to claim 4, wherein in the step S3, two blood-soaked areas of the blood-soaked gauze image are obtained by using a mask, defined as: 第一个浸血区域

first blood-soaked area

第二个浸血区域

Second blood-soaked area

其中，(i,j)为浸血图像在RGB色彩空间中的像素位置，B_i,j为原始浸血纱布图片(i,j)位置的HSV像素矢量值。Among them, (i,j) is the pixel position of the blood-soaked image in the RGB color space, and B _i,j is the HSV pixel vector value of the original blood-soaked gauze image at (i,j) position. 6.根据权利要求5所述的一种基于特征工程的纱布浸血量估算模型构建方法，其特征在于：在所述步骤S4中，血红蛋白量由血红蛋白浓度与浸血区域面积占比相乘得到，具体步骤包括：6. A method for constructing a model for estimating blood soaking volume in gauze based on feature engineering according to claim 5, wherein in the step S4, the amount of hemoglobin is obtained by multiplying the hemoglobin concentration and the area ratio of the blood soaking area , the specific steps include: (1)计算两个浸血区域掩模下的像素个数，分别记为PR1_num和PR2_num；(1) calculate the number of pixels under the two blood-soaked area masks, which are respectively denoted as PR1 _num and PR2 _num ; (2)分别计算两个掩模下的浸血区域面积占比，即占整张图像的比例：(2) Calculate the proportion of the area of the blood soaked area under the two masks, that is, the proportion of the entire image:

(3)对患者血红蛋白浓度归一化处理：(3) Normalize the patient's hemoglobin concentration:

其中，Hbc表示当前单个患者的血红蛋白浓度，Max和Min分别代表所有患者血红蛋白浓度的最大值和最小值；Among them, Hbc represents the current hemoglobin concentration of a single patient, and Max and Min represent the maximum and minimum hemoglobin concentrations of all patients, respectively; (4)分别计算两个掩模下浸血区域的血红蛋白量，定义为浸血区域面积占比与归一化血红蛋白浓度的乘积：(4) Calculate the amount of hemoglobin in the blood-soaked area under the two masks respectively, which is defined as the product of the area proportion of the blood-soaked area and the normalized hemoglobin concentration: Hgb1＝Hbc×AR1，Hgb1=Hbc×AR1, Hgb2＝Hbc×AR2。Hgb2=Hbc×AR2. 7.根据权利要求6所述的一种基于特征工程的纱布浸血量估算模型构建方法，其特征在于：在所述步骤S4中，对每一幅浸血纱布图像，计算浸血区域在HSV色彩空间中各通道的均值和方差，由于存在两个浸血区域掩模，对所产生的两个浸血区域分别计算均值和方差特征；将这些特征记为：H1_mean、H1_std、S1_mean、S1_std、V1_mean、V1_std、H2_mean、H2_std、S2_mean、S2_std、V2_mean、V2_std，共12个特征。7. The method for constructing a blood-soaked gauze volume estimation model based on feature engineering according to claim 6, wherein in the step S4, for each blood-soaked gauze image, calculate the blood-soaked area in the HSV The mean and variance of each channel in the color space, since there are two blood-soaked area masks, the mean and variance features are calculated for the two generated blood-soaked areas respectively; these features are recorded as: H1_mean, H1_std, S1_mean, S1_std, V1_mean, V1_std, H2_mean, H2_std, S2_mean, S2_std, V2_mean, V2_std, a total of 12 features. 8.根据权利要求7所述的一种基于特征工程的纱布浸血量估算模型构建方法，其特征在于：在所述步骤S4中，每一幅浸血纱布图像，共提取14个特征，包括特征Hgb1和Hgb2，以及所述12个特征；对步骤S1中所构建数据集中的所有图片提取特征，形成图像特征集；图像特征集串行构建获并行构建。8. A method for constructing a blood-soaked gauze volume estimation model based on feature engineering according to claim 7, characterized in that: in the step S4, for each blood-soaked gauze image, a total of 14 features are extracted, including Features Hgb1 and Hgb2, and the 12 features; features are extracted from all the pictures in the data set constructed in step S1 to form an image feature set; the image feature set is constructed serially and constructed in parallel. 9.根据权利要求8所述的一种基于特征工程的纱布浸血量估算模型构建方法，其特征在于：在所述步骤S5中，纱布浸血量估算机器学习模型为所构建的图像特征集上构建的多元线性回归模型。9 . A method for constructing a model for estimating blood soaking in gauze based on feature engineering according to claim 8 , wherein in the step S5 , the machine learning model for estimating blood soaking in gauze is the constructed image feature set. 10 . The multiple linear regression model built on.