A method, system and device for ai-assisted scoring of sedation state of icu patients

Abstract

The present application relates to the technical field of medical artificial intelligence, and provides an AI-assisted scoring method, system and device for the sedation state of an ICU patient, which comprises: collecting a patient image through a camera, recognizing a patient facial region and an eye region in real time by using an edge computing device; calculating a patient head movement state, recording and storing a head shaking indication value per minute; calculating a patient eye state, recording and storing an eye opening indication value per minute; calculating a head shaking indication ratio and an eye opening indication ratio of the patient within one hour; constructing and training a multilayer perceptron neural network, using the head shaking indication ratio and the eye opening indication ratio as input features, and obtaining and outputting a sedation state score of the patient. The patient sedation scoring system of the present application has better performance than a traditional manual scoring system, and has great advantages in objectivity, real-time performance, automation and reduction of labor costs.

Description

Technical Field

[0001] The embodiments of the present invention generally relate to the field of medical artificial intelligence technology, and in particular to an AI-assisted scoring method, system and device for sedation status of ICU patients. Background Technology

[0002] Sedation scores for ICU patients are crucial for patient treatment and care. From a clinical perspective, accurate sedation scores can prevent unforeseen events such as patient self-extubation, and by using precise medication to avoid over- or under-sedation, effectively prevent complications such as respiratory depression and delirium, and shorten the duration of mechanical ventilation. Appropriate sedation management can shorten patients' ICU stay, improve bed turnover, and significantly reduce treatment costs and improve the efficiency of medical resource utilization by optimizing sedative drug use and reducing the incidence of complications. Therefore, establishing an accurate and objective sedation scoring system is of great value in improving the quality of ICU treatment.

[0003] Traditional ICU sedation scoring methods primarily rely on manual assessment by healthcare professionals using tools such as the Richmond Agitation-Sedation Scale (RASS). This method has several significant limitations. First, the scoring process requires healthcare professionals to awaken the patient through verbal or physical stimulation, which can disrupt the patient's rest and even increase stress. Second, manual scoring is subjective; different assessors may give different scores to the same patient, leading to inconsistent results. Third, due to the heavy workload of healthcare professionals, assessments are typically conducted at fixed times (e.g., every 4-8 hours), making continuous monitoring impossible and potentially missing dynamic changes in the patient's condition. Furthermore, nighttime assessments increase the workload of healthcare professionals and may be affected by fatigue, reducing accuracy. Finally, traditional scoring methods lack objective data support; scoring records require manual entry, hindering data collection, analysis, and long-term follow-up. These limitations can affect the accuracy and timeliness of sedation management, thereby impacting treatment outcomes. Summary of the Invention

[0004] To address the above issues, this invention combines deep learning technology with medical monitoring, employing non-contact continuous video monitoring and edge computing devices to analyze patients' facial features and movement status in real time, thereby achieving automated and continuous assessment of the sedation level of ICU patients.

[0005] According to embodiments of the present invention, an AI-assisted scoring method, system, and device for sedation status of ICU patients are provided.

[0006] In a first aspect of the invention, an AI-assisted scoring method for the sedation status of ICU patients is provided. The method includes:

[0007] Step S01: Acquire patient images through a camera, use edge computing devices to identify the patient's facial and eye areas in real time, and upload the identification results to the cloud server in real time;

[0008] Step S02: Calculate the patient's head movement state based on the identified facial region: head shaking or non-head shaking state, and record and store the head shaking indicator value per minute.

[0009] Step S03: Calculate the patient's eye status based on the identified eye area: open or closed eyes, and record and store the eye-opening indicator value every minute;

[0010] Step S04: Calculate the patient's head-shaking index ratio within one hour based on the recorded head-shaking index values, and calculate the patient's eye-opening index ratio within one hour based on the recorded eye-opening index values.

[0011] Step S05: Construct and train a multilayer perceptron neural network, using the head-shaking sign ratio and the eye-opening sign ratio as input features, and optimize the network parameters using the backpropagation algorithm to obtain and output the patient's sedation status score.

[0012] Furthermore, the specific steps for calculating the patient's head movement state based on the identified facial region in step S02 are as follows:

[0013] Step S021: Obtain the difference dx between the maximum and minimum X-axis coordinates of the patient's facial center point within one minute;

[0014] Step S022: Obtain the difference dy between the maximum and minimum Y-axis coordinates of the patient's facial center point within one minute;

[0015] Step S023: Calculate the maximum head movement distance within one minute, where

[0016] Step S024: Compare the distance with a preset threshold. When the distance is greater than the preset threshold, it is determined to be a head-shaking state, and the head-shaking indicator value is recorded as 1. When the distance is less than or equal to the preset threshold, it is determined to be a non-head-shaking state, and the head-shaking indicator value is recorded as 0.

[0017] Furthermore, the specific steps for calculating the patient's eye state based on the identified eye region in step S03 are as follows:

[0018] Step S031: Real-time detection of the patient's eye opening and closing status based on the improved YOLOv8 eye state detection algorithm;

[0019] Step S032: Count the number of times the patient opens their eyes within one minute;

[0020] Step S033: Compare the number of times the eyes open with a preset threshold. When the number of times the eyes open is greater than the preset threshold, it is determined to be an open-eye state, and the open-eye indicator value is recorded as 1. When the number of times the eyes open is less than or equal to the preset threshold, it is determined to be a closed-eye state, and the open-eye indicator value is recorded as 0.

[0021] Furthermore, the specific steps of the improved YOLOv8 eye state detection algorithm described in step S031 are as follows:

[0022] The patient's eyes were captured by a camera and labeled with software to obtain an eye state dataset, which was labeled into two categories: Open_eyes and Closed_eyes.

[0023] An improved YOLOv8 network was constructed, and a dual-branch attention enhancement module was designed, including a main branch and an auxiliary branch. The main branch maintains the original YOLOv8 feature extraction network structure, while the auxiliary branch is designed for eye feature enhancement and improved feature fusion strategies.

[0024] Furthermore, the auxiliary branches specifically include:

[0025] Spatial pyramid pooling module: Sets three pooling scales: {5×5, 9×9, 13×13}; After parallel pooling, dimensionality is reduced by 1×1 convolution; Feature maps of different scales are stitched and fused together;

[0026] Channel attention mechanism: global average pooling is used to obtain channel descriptors; two fully connected layers learn channel weights: FC1(C→C / r), FC2(C / r→C); weights are normalized using the sigmoid function; channel weighting is applied to the feature map;

[0027] Residual connection structure: Short-circuit connection between original features and enhanced features; additive fusion method; adjust the number of channels through 1×1 convolution.

[0028] Furthermore, the improved feature fusion strategy is specifically as follows:

[0029] Adaptive feature fusion: Calculate the main and auxiliary branch feature similarity matrix S: S[i,j]=F1[i]·F2[j] / (||F1[i]||×||F2[j]||); Generate fusion weights α and β based on similarity; Weighted fusion: Fout=α×F1+β×F2;

[0030] The feature pyramid network structure includes: top-down paths, bottom-up paths, and lateral connections;

[0031] Specifically, the top-down path is: P5(1 / 32)→P4(1 / 16)→P3(1 / 8); using 2x upsampling; 1×1 convolution to adjust the channels;

[0032] Bottom-up path: P3→P4→P5; using a 3×3 convolution with a stride of 2;

[0033] Lateral connection: feature maps are aligned and then added element by element; 3×3 convolution is used to fuse features.

[0034] Furthermore, the specific steps for calculating the patient's head-shaking indicator ratio within one hour based on the recorded head-shaking indicator values ​​in step S04 are as follows: count all head-shaking indicator values ​​within one hour, sum up all head-shaking indicator values ​​within one hour and divide by 60 to obtain the head-shaking indicator ratio.

[0035] The specific steps for calculating the patient's eye-opening indicator ratio within one hour based on the recorded eye-opening indicator values ​​in step S04 are as follows: count all eye-opening indicator values ​​within one hour, sum up all eye-opening indicator values ​​within one hour and divide by 60 to obtain the eye-opening indicator ratio.

[0036] In a second aspect of the invention, an AI-assisted scoring system for the sedation status of ICU patients is provided. The system includes:

[0037] Image recognition module: used to capture patient images through a camera, use edge computing devices to identify the patient's facial and eye areas in real time, and upload the recognition results to the cloud server in real time;

[0038] Head shaking indicator value recording module: used to calculate the patient's head movement status based on the identified facial area: head shaking state or non-head shaking state, and record and store the head shaking indicator value every minute;

[0039] Eye-opening indicator value recording module: used to calculate the patient's eye status based on the identified eye area: eye-opening or eye-closed, and record and store the eye-opening indicator value every minute;

[0040] Indication ratio calculation module: used to calculate the patient's head-shaking indication ratio within one hour based on the recorded head-shaking indication value, and to calculate the patient's eye-opening indication ratio within one hour based on the recorded eye-opening indication value.

[0041] Sedation status scoring module: used to construct and train multilayer perceptron neural network, using head shaking sign ratio and eye opening sign ratio as input features, and employing backpropagation algorithm to optimize network parameters, obtain and output the patient's sedation status score.

[0042] In a third aspect of the invention, an electronic device is provided. The electronic device includes a memory and a processor, the memory storing a computer program, the processor executing the program to implement the method according to a first aspect of the invention.

[0043] In a fourth aspect of the invention, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the method according to a first aspect of the invention.

[0044] This invention combines deep learning technology with medical monitoring, employing non-contact continuous video monitoring and edge computing devices to analyze patients' facial features and movement status in real time, thereby achieving automated and continuous assessment of the sedation level of ICU patients.

[0045] It should be understood that the description in the Summary of the Invention is not intended to limit the key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description.

[0046] The beneficial effects of this invention are:

[0047] 1. The system adopts non-contact continuous video monitoring, and analyzes the patient's facial features and movement status in real time through edge computing devices, avoiding physical interference with the patient and ensuring that the patient gets sufficient rest;

[0048] 2. The system can monitor continuously 24 hours a day, capturing subtle changes in the patient in real time, such as objective indicators like eye opening and closing status and head movement frequency, providing continuous and accurate scoring data;

[0049] 3. The system is based on a standardized evaluation algorithm, which eliminates errors caused by subjective human judgment and ensures the consistency and reliability of the scoring results;

[0050] 4. AI-assisted scoring significantly reduces the workload of medical staff. Attached Figure Description

[0051] The above and other features, advantages, and aspects of the various embodiments of the present invention will become more apparent from the accompanying drawings and the following detailed description. Wherein:

[0052] Figure 1 A flowchart of an AI-assisted scoring method for sedation status of ICU patients according to an embodiment of the present invention is shown;

[0053] Figure 2 A flowchart of an improved YOLOv8 eye state detection algorithm based on an embodiment of the present invention is shown;

[0054] Figure 3 A schematic diagram of the patient's head-shaking rate over the past 24 hours is shown according to an embodiment of the present invention;

[0055] Figure 4 A schematic diagram showing the 24-hour eye-opening rate of patients according to an embodiment of the present invention is shown;

[0056] Figure 5 A schematic diagram of a patient's AI-assisted sedation score over the past 24 hours, according to an embodiment of the present invention, is shown.

[0057] Figure 6 A comparison between AI-assisted scoring and traditional human RASS scoring according to an embodiment of the present invention is shown;

[0058] Figure 7 A system block diagram of AI-assisted scoring of sedation status in ICU patients according to an embodiment of the present invention is shown;

[0059] Figure 8 A schematic diagram of an AI-assisted scoring device for sedation status of ICU patients according to an embodiment of the present invention is shown. Detailed Implementation

[0060] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0061] According to embodiments of the present invention, an AI-assisted scoring method, system, and device for sedation status of ICU patients are proposed. By combining deep learning technology with medical monitoring, non-contact continuous video monitoring is adopted, and edge computing devices are used to analyze the patient's facial features and movement status in real time, thereby realizing automated and continuous assessment of the sedation level of ICU patients.

[0062] The principles and spirit of the present invention will be explained in detail below with reference to several representative embodiments.

[0063] Figure 1 This is a schematic flowchart of an AI-assisted scoring method for the sedation status of ICU patients according to an embodiment of the present invention. The method includes:

[0064] Step S01: Acquire patient images through a camera, use edge computing devices to identify the patient's facial and eye areas in real time, and upload the identification results to the cloud server in real time;

[0065] Step S02: Calculate the patient's head movement state based on the identified facial region: head shaking or non-head shaking state, and record and store the head shaking indicator value per minute.

[0066] Step S03: Calculate the patient's eye status based on the identified eye area: open or closed eyes, and record and store the eye-opening indicator value every minute;

[0067] Step S04: Calculate the patient's head-shaking index ratio within one hour based on the recorded head-shaking index values, and calculate the patient's eye-opening index ratio within one hour based on the recorded eye-opening index values.

[0068] Step S05: Construct and train a multilayer perceptron neural network, using the head-shaking sign ratio and the eye-opening sign ratio as input features, and optimize the network parameters using the backpropagation algorithm to obtain and output the patient's sedation status score.

[0069] It should be noted that although the operation of the method of the present invention has been described in a specific order in the above embodiments and figures, this does not require or imply that the operations must be performed in that specific order, or that all the operations shown must be performed to achieve the desired result. Additionally or alternatively, certain steps may be omitted, multiple steps may be combined into one step, and / or one step may be broken down into multiple steps.

[0070] To provide a clearer explanation of the AI-assisted scoring method for the sedation status of ICU patients, a specific embodiment is described below. However, it is worth noting that this embodiment is only for better illustrating the present invention and does not constitute an undue limitation on the present invention.

[0071] The following specific example will further illustrate the method of AI-assisted scoring of sedation status in ICU patients:

[0072] like Figure 2 As shown, the improved YOLOv8 eye state detection algorithm includes the following steps:

[0073] Step 1: Use a camera to capture images of the patient's eyes and use software to label the eye images to obtain an eye state dataset; the dataset contains 2000 images, with a training set to validation set ratio of 8:2, consisting of 1600 images in the training set and 400 images in the validation set; the labels are divided into two categories: open eyes (Open_eyes) and closed eyes (Closed_eyes).

[0074] Step 2: Construct an improved YOLOv8 network and design a dual-branch attention enhancement module (DAEM). Specifically, the main branch maintains the original YOLOv8 feature extraction network structure, while the auxiliary branch is designed for eye feature enhancement and improved feature fusion strategies.

[0075] The auxiliary branch introduces a spatial pyramid pooling module to obtain multi-scale receptive field information; a channel attention mechanism is adopted to adaptively adjust the feature channel weights; and a residual connection structure is designed to fuse the original features and the enhanced features.

[0076] The auxiliary branches are as follows:

[0077] (1) Spatial Pyramid Pooling Module (SPP):

[0078] Three pooling scales are set: {5×5, 9×9, 13×13};

[0079] Dimensionality reduction is achieved through parallel pooling followed by 1×1 convolution.

[0080] Feature maps of different scales are stitched together and fused.

[0081] (2) Channel attention mechanism (SE module):

[0082] Global average pooling is used to obtain channel descriptors;

[0083] The learning channel weights for the two fully connected layers are: FC1(C→C / r), FC2(C / r→C);

[0084] Weights are normalized using the sigmoid function;

[0085] Channel weighting is applied to the feature map.

[0086] (1) Residual connection structure:

[0087] Short-circuit connections are made between the original features and the enhanced features;

[0088] An additive fusion method is adopted;

[0089] Adjust the number of channels using 1×1 convolution.

[0090] The feature fusion strategy is as follows:

[0091] (2) Adaptive feature fusion:

[0092] Calculate the feature similarity matrix S between the main and auxiliary branches: S[i,j]=F1[i]·F2[j] / (||F1[i]||×||F2[j]||);

[0093] The fusion weights α and β are generated based on similarity.

[0094] Weighted fusion: Fout = α × F1 + β × F2.

[0095] (3) Feature Pyramid Network (FPN) structure:

[0096] 1) Top-down path:

[0097] P5(1 / 32)→P4(1 / 16)→P3(1 / 8);

[0098] Use 2x upsampling;

[0099] 1×1 convolution adjustment channel.

[0100] 2) Bottom-up path:

[0101] P3→P4→P5;

[0102] A 3×3 convolution with a stride of 2 is used.

[0103] 3) Horizontal connection:

[0104] After aligning the feature maps, add them element by element;

[0105] 3×3 convolution fusion features.

[0106] The improved feature fusion strategy achieves bidirectional information flow of features at different scales by designing a feature pyramid network structure, thereby introducing cross-layer feature aggregation.

[0107] Experimental Procedure: The improved YOLOv8 network was configured with a batch size of 16, epochs of 100, an initial learning rate of 0.01, and an input image size of 640×640. The optimizer used was SGD. Under the same experimental conditions, a comparative experiment was conducted using the YOLOv8 network and the network model of this invention. The P, R, and mAP values ​​of each model on the eye state dataset were obtained. The detection results are shown in Table 1.

[0108] Table 1

[0109] Model P R mAP@0.5 mAP@0.5:0.95 YOLOv8 0.956 0.962 0.975 0.658 This invention 0.991 0.977 0.983 0.690

[0110] As can be seen from the table, the model of this invention has a significant performance improvement on the eye state dataset. It is 3.5% higher than YOLOv8 in Precision, 1.5% higher than YOLOv8 in Recall, 0.8% higher than YOLOv8 in mAP@0.5, and 3.2% higher than YOLOv8 in mAP@0.5:0.95. This indicates that the model of this invention effectively improves the performance of the eye state detection algorithm.

[0111] like Figure 1 As shown, the AI-assisted scoring method for sedation status in ICU patients includes the following steps:

[0112] Step S01: Acquire patient images through a camera, use edge computing devices to identify the patient's facial and eye areas in real time, and upload the identification results to the cloud server in real time;

[0113] Step S02: Calculate the patient's head movement state based on the identified facial region, specifically including:

[0114] Step S021: Obtain the difference dx between the maximum and minimum X-axis coordinates of the patient's facial center point within one minute;

[0115] Step S022: Obtain the difference dy between the maximum and minimum Y-axis coordinates of the patient's facial center point within one minute;

[0116] Step S023: Calculate the maximum head movement distance within one minute, where

[0117] Step S024: Compare the distance with a preset threshold. When the distance is greater than the preset threshold, it is determined to be a head-shaking state, and the head-shaking indicator value is recorded as 1. When the distance is less than or equal to the preset threshold, it is determined to be a non-head-shaking state, and the head-shaking indicator value is recorded as 0.

[0118] Step S03: Calculate the patient's eye condition based on the identified eye region, specifically including:

[0119] Step S031: Real-time detection of the patient's eye opening and closing status based on the improved YOLOv8 eye state detection algorithm;

[0120] Step S032: Count the number of times the patient opens their eyes within one minute;

[0121] Step S033: Compare the number of times the eyes open with a preset threshold. When the number of times the eyes open is greater than the preset threshold, it is determined to be an open-eye state, and the open-eye indicator value is recorded as 1. When the number of times the eyes open is less than or equal to the preset threshold, it is determined to be a closed-eye state, and the open-eye indicator value is recorded as 0.

[0122] Step S04: Calculate the patient's head-shaking indicator ratio within one hour based on the recorded head-shaking indicator values. Specifically: record and store the head-shaking indicator values ​​every minute; calculate all head-shaking indicator values ​​within one hour; sum all head-shaking indicator values ​​within one hour and divide by 60 to obtain the head-shaking indicator ratio, such as... Figure 3 The diagram shows the patient's head-shaking rate over the past 24 hours.

[0123] The eye-opening indicator ratio for a patient within one hour is calculated based on the recorded eye-opening indicator values. Specifically: record and store the eye-opening indicator values ​​every minute; calculate all eye-opening indicator values ​​within one hour; sum all eye-opening indicator values ​​within one hour and divide by 60 to obtain the eye-opening indicator ratio. Figure 4 The diagram shows the percentage of patients with open eyes over the past 24 hours.

[0124] Step S05: Construct and train a multilayer perceptron neural network, specifically including: designing a neural network structure containing an input layer, multiple hidden layers, and an output layer; using the head-shaking sign ratio and the eye-opening sign ratio as input features; optimizing network parameters using the backpropagation algorithm; and outputting the patient's sedation status score, such as... Figure 5 The image shows the patient's AI-assisted score over the past 24 hours.

[0125] like Figure 6 As shown, the AI-assisted scoring system of the present invention has good consistency with the traditional human scoring system, and at the same time, it has better objectivity than the traditional human scoring system, while providing real-time, continuous and accurate scoring data.

[0126] Based on the same inventive concept, this invention also proposes an AI-assisted scoring system for the sedation status of ICU patients. The implementation of this system is similar to the implementation of the methods described above; repeated details will not be repeated. Figure 7 As shown, the system 100 includes:

[0127] Image recognition module 101: used to acquire patient images through a camera, use edge computing devices to identify the patient's facial and eye areas in real time, and upload the recognition results to the cloud server in real time;

[0128] Head shaking indicator value recording module 102: used to calculate the patient's head movement state based on the identified facial area: head shaking state or non-head shaking state, and record and store the head shaking indicator value every minute.

[0129] Eye-opening indicator value recording module 103: used to calculate the patient's eye status based on the identified eye area: eye-opening state or eye-closed state, and record and store the eye-opening indicator value every minute;

[0130] Indication ratio calculation module 104: used to calculate the patient's head-shaking index ratio within one hour based on the recorded head-shaking index value, and to calculate the patient's eye-opening index ratio within one hour based on the recorded eye-opening index value.

[0131] Sedation status scoring module 105: Used to construct and train a multilayer perceptron neural network, using the head shaking sign ratio and eye opening sign ratio as input features, and employing the backpropagation algorithm to optimize network parameters, obtain and output the patient's sedation status score.

[0132] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the described module can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0133] like Figure 8As shown, the device includes a central processing unit (CPU), which can perform various appropriate actions and processes based on computer program instructions stored in read-only memory (ROM) or loaded from storage units into random access memory (RAM). The RAM can also store various programs and data required for device operation. The CPU, ROM, and RAM are interconnected via a bus. Input / output (I / O) interfaces are also connected to the bus.

[0134] Multiple components in the device are connected to the I / O interface, including: input units such as keyboards and mice; output units such as various types of displays and speakers; storage units such as disks and optical discs; and communication units such as network interface cards (NICs), modems, and wireless transceivers. The communication unit allows the device to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0135] The processing unit executes the various methods and processes described above, such as method steps S01 to S05. For example, in some embodiments, method steps S01 to S05 may be implemented as a computer software program tangibly contained in a machine-readable medium, such as a storage unit. In some embodiments, part or all of the computer program may be loaded and / or installed on the device via ROM and / or a communication unit. When the computer program is loaded into RAM and executed by the CPU, one or more steps of method steps S01 to S05 described above may be performed. Alternatively, in other embodiments, the CPU may be configured to execute method steps S01 to S05 by any other suitable means (e.g., by means of firmware).

[0136] The functions described above in this document can be performed at least in part by one or more hardware logic components. For example, exemplary types of hardware logic components that can be used, without limitation, include: field programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload programmable logic devices (CPLDs), and so on.

[0137] The program code used to implement the methods of the present invention can be written in any combination of one or more programming languages. This program code can be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing system, such that when executed by the processor or controller, the functions / operations specified in the flowcharts and / or block diagrams are implemented. The program code can be executed entirely on the machine, partially on the machine, as a standalone software package partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0138] In the context of this invention, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, system, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. Machine-readable media can include, but are not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, systems, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0139] Furthermore, although the operations are described in a specific order, this should be understood as requiring that such operations be performed in the specific order shown or in sequential order, or requiring that all illustrated operations be performed to achieve the desired result. In certain environments, multitasking and parallel processing may be advantageous. Similarly, although several specific implementation details are included in the above discussion, these should not be construed as limiting the scope of the invention. Certain features described in the context of individual embodiments may also be implemented in combination in a single implementation. Conversely, various features described in the context of a single implementation may also be implemented individually or in any suitable sub-combination in multiple implementations.

[0140] Although the subject matter has been described using language specific to structural features and / or methodological logic, it should be understood that the subject matter defined in the appended claims is not necessarily limited to the specific features or actions described above. Rather, the specific features and actions described above are merely illustrative examples of implementing the claims.

Claims

1. A method for AI-assisted scoring of sedation status in ICU patients, characterized in that, The method includes: Step S01: Acquire patient images through a camera, use edge computing devices to identify the patient's facial and eye areas in real time, and upload the identification results to the cloud server in real time; Step S02: Calculate the patient's head movement state based on the identified facial region: head shaking or non-head shaking state, and record and store the head shaking indicator value per minute. Step S03: Calculate the patient's eye status (open or closed) based on the identified eye region, record and store the eye-opening indicator value every minute. Specific steps are as follows: Step S031: Real-time detection of the patient's eye opening and closing state based on the improved YOLOv8 eye state detection algorithm; the improved YOLOv8 eye state detection algorithm constructs a dual-branch attention enhancement module including a main branch and an auxiliary branch, the auxiliary branch is designed for eye feature enhancement and improved feature fusion strategy; Step S032: Count the number of times the patient opens their eyes within one minute; Step S033: Compare the number of times the eyes open with a preset threshold. When the number of times the eyes open is greater than the preset threshold, it is determined to be an open-eye state and the open-eye indicator value is recorded as 1. When the number of times the eyes open is less than or equal to the preset threshold, it is determined to be a closed-eye state and the open-eye indicator value is recorded as 0. Step S04: Calculate the patient's head-shaking index ratio within one hour based on the recorded head-shaking index values, and calculate the patient's eye-opening index ratio within one hour based on the recorded eye-opening index values. Step S05: Construct and train a multilayer perceptron neural network, using the head-shaking sign ratio and the eye-opening sign ratio as input features, and use the backpropagation algorithm to optimize the network parameters to obtain and output the patient's sedation status score. The auxiliary branches specifically include: Spatial pyramid pooling module: Sets three pooling scales: {5×5, 9×9, 13×13}; After parallel pooling, dimensionality is reduced by 1×1 convolution; Feature maps of different scales are stitched and fused. Channel attention mechanism: global average pooling is used to obtain channel descriptors; two fully connected layers learn channel weights: FC1: C→C / r, FC2: C / r→C; weights are normalized using the sigmoid function; channel weighting is applied to the feature map; Residual connection structure: Short-circuit connections are made between the original features and the enhanced features; an additive fusion method is used; the number of channels is adjusted through 1×1 convolution; The improved feature fusion strategy is as follows: Adaptive feature fusion: Calculate the similarity matrix S of the main and auxiliary branches: S[i,j] = F1[i]•F2[j] / (||F1[i]|| × ||F2[j]||); Generate fusion weights α and β based on similarity; Weighted fusion: Fout = α×F1 + β×F2; The feature pyramid network structure includes: top-down paths, bottom-up paths, and lateral connections; Specifically, the top-down path propagates from P5 to P4, and then to P3; it uses 2x upsampling; and 1×1 convolution adjusts the channels. The top-up path propagates from P3 to P4, and then to P5; a 3×3 convolution with a stride of 2 is used. Lateral connection: feature maps are aligned and then added element by element; 3×3 convolution is used to fuse features.

2. The method for AI-assisted scoring of sedation status in ICU patients according to claim 1, characterized in that, The specific steps for calculating the patient's head movement state based on the identified facial region in step S02 are as follows: Step S021: Obtain the difference dx between the maximum and minimum X-axis coordinates of the patient's facial center point within one minute; Step S022: Obtain the difference dy between the maximum and minimum Y-axis coordinates of the patient's facial center point within one minute; Step S023: Calculate the maximum head movement distance within one minute, where ； Step S024: Compare the distance with a preset threshold. When the distance is greater than the preset threshold, it is determined to be a head-shaking state, and the head-shaking indicator value is recorded as 1. When the distance is less than or equal to the preset threshold, it is determined to be a non-head-shaking state, and the head-shaking indicator value is recorded as 0.

3. The method for AI-assisted scoring of sedation status in ICU patients according to claim 1, characterized in that, The specific steps of the improved YOLOv8 eye state detection algorithm described in step S031 are as follows: The patient's eyes were captured by a camera and labeled with software to obtain an eye state dataset, which was labeled into two categories: Open_eyes and Closed_eyes. An improved YOLOv8 network was constructed, and a dual-branch attention enhancement module was designed, including a main branch and an auxiliary branch. The main branch maintains the original YOLOv8 feature extraction network structure, while the auxiliary branch is designed for eye feature enhancement and improved feature fusion strategies.

4. The method for AI-assisted scoring of sedation status in ICU patients according to claim 1, characterized in that, The specific steps for calculating the patient's head-shaking index ratio within one hour based on the recorded head-shaking index values ​​in step S04 are as follows: count all head-shaking index values ​​within one hour, sum up all head-shaking index values ​​within one hour and divide by 60 to obtain the head-shaking index ratio. The specific steps for calculating the patient's eye-opening indicator ratio within one hour based on the recorded eye-opening indicator values ​​in step S04 are as follows: count all eye-opening indicator values ​​within one hour, sum up all eye-opening indicator values ​​within one hour and divide by 60 to obtain the eye-opening indicator ratio.

5. A system for AI-assisted scoring of sedation status in ICU patients, characterized in that, The system implements the method as described in any one of claims 1 to 4, comprising: Image recognition module: used to capture patient images through a camera, use edge computing devices to identify the patient's facial and eye areas in real time, and upload the recognition results to the cloud server in real time; Head shaking indicator value recording module: used to calculate the patient's head movement status based on the identified facial area: head shaking state or non-head shaking state, and record and store the head shaking indicator value every minute; Eye-opening indicator recording module: Used to calculate the patient's eye status (open or closed) based on the identified eye region, and to record and store the eye-opening indicator value every minute. The specific steps are as follows: Step S031: Real-time detection of the patient's eye opening and closing state based on the improved YOLOv8 eye state detection algorithm; the improved YOLOv8 eye state detection algorithm constructs a dual-branch attention enhancement module including a main branch and an auxiliary branch, the auxiliary branch is designed for eye feature enhancement and improved feature fusion strategy; Step S032: Count the number of times the patient opens their eyes within one minute; Step S033: Compare the number of times the eyes open with a preset threshold. When the number of times the eyes open is greater than the preset threshold, it is determined to be an open-eye state and the open-eye indicator value is recorded as 1. When the number of times the eyes open is less than or equal to the preset threshold, it is determined to be a closed-eye state and the open-eye indicator value is recorded as 0. Indication ratio calculation module: used to calculate the patient's head-shaking indication ratio within one hour based on the recorded head-shaking indication value, and to calculate the patient's eye-opening indication ratio within one hour based on the recorded eye-opening indication value. Sedation status scoring module: used to construct and train a multilayer perceptron neural network, using the head shaking sign ratio and eye opening sign ratio as input features, and using the backpropagation algorithm to optimize the network parameters, obtain and output the patient's sedation status score; The auxiliary branches specifically include: Spatial pyramid pooling module: Sets three pooling scales: {5×5, 9×9, 13×13}; After parallel pooling, dimensionality is reduced by 1×1 convolution; Feature maps of different scales are stitched and fused. Channel attention mechanism: global average pooling is used to obtain channel descriptors; two fully connected layers learn channel weights: FC1: C→C / r, FC2: C / r→C; weights are normalized using the sigmoid function; channel weighting is applied to the feature map; Residual connection structure: Short-circuit connections are made between the original features and the enhanced features; an additive fusion method is used; the number of channels is adjusted through 1×1 convolution; The improved feature fusion strategy is as follows: Adaptive feature fusion: Calculate the similarity matrix S of the main and auxiliary branches: S[i,j] = F1[i]•F2[j] / (||F1[i]|| × ||F2[j]||); Generate fusion weights α and β based on similarity; Weighted fusion: Fout = α×F1 + β×F2; The feature pyramid network structure includes: top-down paths, bottom-up paths, and lateral connections; Specifically, the top-down path propagates from P5 to P4, and then to P3; it uses 2x upsampling; and 1×1 convolution adjusts the channels. The top-up path propagates from P3 to P4, and then to P5; a 3×3 convolution with a stride of 2 is used. Lateral connection: feature maps are aligned and then added element by element; 3×3 convolution is used to fuse features.

6. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the program, it implements the method as described in any one of claims 1 to 4.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1 to 4.

Images