CN113570545A

CN113570545A - Visual identification pain grading assessment method

Info

Publication number: CN113570545A
Application number: CN202110609079.8A
Authority: CN
Inventors: 胡安民; 李惠萍; 海超; 钟雄雄
Original assignee: Shenzhen Peoples Hospital
Current assignee: Shenzhen Peoples Hospital
Priority date: 2021-06-01
Filing date: 2021-06-01
Publication date: 2021-10-29

Abstract

The invention discloses a method for grading evaluation of visual recognition pain, and belongs to the field of pain evaluation. The method comprises the following steps: acquiring image information of a patient; extracting posture information of the patient through a skeleton model based on the posture image, and extracting corresponding positions and relative distance information of the five sense organs of the patient through a five sense organ model based on the face image; and inputting the trained convolutional neural network and the long and short memory network model according to the extracted characteristic information to obtain a pain grading evaluation result of the patient. The technical scheme provided by the invention realizes the pain assessment method based on visual means, and the extracted data characteristics of the patient comprise the skeleton posture information of the patient and the facial information of five sense organs.

Description

Visual identification pain grading assessment method

Technical Field

The application relates to the field of artificial intelligence, in particular to a medical image recognition method.

Background

Pain is an unpleasant sensory and emotional experience associated with actual or potential tissue damage. Pain as a subjective and complex perception, the intensity of pain is usually assessed in a clinical setting by self-reported scales. Pain assessment provides assistance in the decision making of relevant medical interventions.

Such subjective methods are considered the "gold standard" for pain measurement. However, in some cases the reliability of self-reporting may be affected by a range of physiological, psychological and environmental factors. For example, many pain sufferers, when repeatedly evaluated, tend to exacerbate their severity and hold a negative attitude.

In view of the defects, the inventor extracts facial expression features and posture features of the patient through a deep convolutional neural network, so that the accuracy of classification recognition or prediction is finally improved.

Disclosure of Invention

The invention aims at predicting delirium in advance, and solves the problem of intelligently predicting delirium on the basis of historical data of delirium patients. The method specifically comprises the following steps:

firstly, screening out a human body image of a patient through a preset human recognition model based on image information.

And then, carrying out median filtering on the human body image through a preset skeleton recognition model to extract human body skeleton data.

Then, the facial features of the patient are obtained through a preset face detection algorithm, the feature information of the five sense organs is extracted, and the relative distance between the five sense organs is calculated. Based on facial expression data of a patient, feature information of the five sense organs is extracted, relative positions of the five sense organs on the face are calculated, and relative distances between the five sense organs are calculated.

And then, inputting the skeleton data and the data continuous frame image set of the five sense organs into a convolutional neural network and a long-term memory neural network, and constructing a deep neural network model.

And then, labeling the expression categories of each continuous frame image set as the expression categories expected to be output through the model training. The preset expression categories include: no pain, little pain, slight pain, obvious pain, severe pain, sharp pain; the pain label category of the pain label sequentially corresponds to 0 point, 1-2 points, 3-4 points, 5-6 points, 7-8 points and 9-10 points in the score of the patient autonomous numerical rating scale.

And finally, outputting the preprocessed image to the training model and calculating a damage function to obtain a loss function value. And adjusting parameters of the deep neural network model through a back propagation algorithm, so that the deviation between the output value of the input and output image processed by the training model and the mapping value of the expression category of the image is within a preset allowable range.

According to the technical scheme, the pain level identification method comprises the steps of firstly extracting a target image of a patient, then respectively extracting human skeleton feature data and position data of five sense organs of the patient from the image, and finally analyzing pain level features of the target image through a pre-constructed model.

Compared with the prior art, the method and the device have the advantages that the posture characteristic and the expression characteristic data of the patient are respectively extracted to construct the model, and the pain level of the patient can be evaluated more timely and comprehensively.

Drawings

In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below.

Fig. 1 is a block diagram of the present invention for visually recognizing pain level.

Detailed Description

The above and further features and advantages of the present invention are described in more detail below with reference to the accompanying drawings.

As shown in FIG. 1, the present invention is primarily directed to more accurately performing visual recognition pain level assessment, which incorporates real-time images of the patient's posture and facial expressions to dynamically assess the patient's pain.

Step S1: an image of a patient sample is acquired.

Specifically, an image containing a patient is input into a pre-constructed human image recognition model to acquire image information of the patient.

Step S2: posture information of the patient is acquired.

Specifically, posture information corresponding to the patient is extracted through a human posture model preset according to a human skeleton model.

Step S3: facial expression information of the patient is acquired.

Specifically, the image information of the patient is input into a preset facial expression extraction model, the position information of the five sense organs of the patient is extracted, and the distance corresponding to the five sense organs is calculated.

Step S4: the information was subjected to pain assessment.

Specifically, dynamic patient posture data and facial feature data are input into a preset model, and the pain state of the patient is output.

It should be noted that, for the video of the patient to be evaluated, the hospitalized patient is usually supine on the bed, the distance and position change is relatively small, and occlusion rarely occurs, and for this reason, a preset face tracking algorithm is adopted to combine with a preset face detection algorithm to accurately detect the posture feature and the facial features of the patient from the target image or the facial features and the facial features of the patient, so as to improve the detection performance.

The target image acquisition process is influenced by factors such as light, video acquisition equipment, patient angle and the like, which all interfere with the feature extraction of the patient. Therefore, the posture image and the face image are acquired from the target image and are preprocessed, and the bone posture and the five sense organs information obtained after the preprocessing are used as target processing data.

Wherein the characteristics of the bone pose may include: position information of the head, shoulders, upper limbs, palms, chests, abdomen, knees, and soles of the patient. Wherein the characteristics of the facial feature information may include: relative position and relative distance information of the five sense organs.

The above examples are only for illustrating the technical idea and features of the present invention, and the purpose thereof is to enable those skilled in the art to understand the content of the present invention and implement the present invention, and not to limit the protection scope of the present invention. All equivalent changes or modifications made according to the technical solutions and concepts described above in the spirit of the present invention should be covered within the protection scope of the present invention.

Claims

1. A method for visually identifying a graded assessment of pain, comprising:

acquiring a sample image of a patient, wherein the sample image comprises a posture image and a face image of the patient;

and outputting corresponding features according to the extracted image information, and generating an image recognition result corresponding to the medical image to be recognized through the pain evaluation model.

2. The method of claim 1, wherein the patient pose image is generated by a skeletal model to predict locations of corresponding human skeletal points in the image.

3. The method of claim 1, wherein the facial image of the patient is generated by a pre-defined model of the five sense organs to generate position and relative distance data of the corresponding five sense organs of the image.

4. The method of claim 1, wherein the pain grading model is constructed by constructing a deep neural network model from a convolutional neural network and a long-term memory neural network.

5. The method of claim 4, wherein the parameters of the deep neural network model are adjusted by back propagation algorithm.