Image recognition and early warning method for bedside pressure sore risk of ICU (Intensive Care Unit)

Abstract

The invention relates to an ICU bedside pressure sore risk image identification and early warning method. The method comprises the following steps: continuously acquiring images at a preset frame rate before and after the same skin area of a patient is pressed and decompressed to obtain a sequence; extracting color, brightness or texture features of the region from the sequence, comparing the color, brightness or texture features with baseline features before decompression, if a recovery threshold is met, judging that the region is reversible, otherwise, judging that the region is limited in recovery when continuous deviation exceeds a first time length; calculating regional displacement and deformation within a second duration under the condition that the limited state is recovered, and judging that the compression is hidden and continuous when the displacement is smaller than a threshold value; and comparing the characteristics of the region before and after body position adjustment or nursing intervention, and if the characteristics are not improved or deteriorated, outputting a pressure sore high-risk early warning. The method does not need an additional contact type sensor, is suitable for continuous and non-invasive monitoring beside an ICU bedside, can provide early warning in the early stage of formation of pressure sores and even before macroscopic injuries occur, and helps nurses to take targeted intervention measures in time, so that the occurrence rate of the pressure sores is reduced, and the nursing quality and the working efficiency of the ICU are improved.

Description

Technical Field

[0001] This invention relates to the field of medical image processing and intelligent medical detection technology, specifically to an image recognition and early warning method for the risk of pressure ulcers at the ICU bedside. Background Technology

[0002] Based on the technical solutions disclosed in Chinese patent document CN112509688A, "Automatic Analysis System, Method, Equipment, and Medium for Pressure Ulcer Images," it can be seen that current image recognition and early warning methods for pressure ulcer risk at the ICU bedside still have significant technical limitations and application drawbacks. These shortcomings are further amplified in real ICU continuous care scenarios, thus limiting their clinical value and promotion effectiveness. First, the core idea of ​​existing solutions is still based on the identification and analysis of images of existing pressure ulcers. The technical approach involves collecting single or a small number of pressure ulcer images, inputting the images into a deep learning model, and directly outputting diagnostic results such as pressure ulcer stage, location, or area. This mode is essentially a post-event identification and static judgment, rather than risk prediction and early warning for the pressure ulcer formation process. This means that the system can only function when there are relatively clear pathological changes on the skin surface or even when an identifiable pressure ulcer morphology has formed. For the large number of skin areas in the ICU that have not yet shown visible damage but are already in a high-risk pressure state, existing methods can hardly provide effective warnings. From a clinical nursing perspective, this easily leads to missing the optimal intervention window.

[0003] Secondly, these methods generally rely on single images or discrete photographs as the analysis object, lacking the ability to model the continuous changes in skin condition over time. Pressure ulcers are a dynamic process that gradually evolves from continuous pressure, impaired microcirculation, and tissue ischemia and hypoxia. Subtle changes in skin color, brightness, and texture often require comparison over time to demonstrate their medical significance. However, current technologies have not constructed a continuous image sequence analysis mechanism for the periods before and after decompression, and during the pressure application and recovery phases. This results in models that can only capture instantaneous features, making it difficult to distinguish between short-term physiological fluctuations and true pathological evolution, thus easily leading to misdiagnosis or missed diagnosis in ICU settings. Thirdly, the proposed approach heavily relies on the training effect of the deep learning model. Its performance is largely limited by the size, annotation quality, and distribution balance of the pressure ulcer dataset. In actual clinical environments, there are significant differences in body type, skin color, lighting conditions, and shooting angles among different hospitals and patients, creating a natural bottleneck in the model's generalization ability. If the training data does not adequately cover certain special body positions, early skin changes, or atypical pressure ulcer samples, the model may experience unstable recognition in real ICU bedside environments. The patent does not provide an adaptive mechanism for complex bedside environment changes and only updates the model by periodically importing datasets, which is insufficient to respond to real-time scene changes in a timely manner.

[0004] Furthermore, existing solutions primarily focus on outcome indicators such as pressure ulcer staging and area, failing to effectively correlate them with nursing behaviors and changes in body position. The solutions do not include the automatic identification of key events such as position adjustments and turning over, nor do they establish a mechanism for comparing and analyzing skin conditions before and after nursing interventions. Therefore, there is a significant disconnect between the system's output and actual nursing procedures. Nurses still need to rely on manual experience to determine when turning, decompression, or enhanced care is necessary. The system functions more as a post-event recording and diagnostic tool than a true bedside intelligent early warning system. Summary of the Invention

[0005] The purpose of this invention is to provide an image recognition and early warning method for the risk of pressure ulcers at the ICU bedside, thereby addressing some of the drawbacks and shortcomings pointed out in the background art.

[0006] The technical solution adopted by the present invention to solve the above-mentioned technical problems is as follows: an image recognition and early warning method for pressure ulcer risk at the ICU bedside, comprising: continuously acquiring images of the same skin area of ​​the patient before and after pressure and decompression at a preset frame rate to obtain a sequence; extracting the color, brightness or texture features of the area from the sequence and comparing them with the baseline features before decompression; if the recovery threshold is met, it is judged as reversible; otherwise, if the deviation continues to exceed a first duration, it is judged as limited recovery.

[0007] Under the condition of restricted recovery, the displacement and deformation of the area are calculated within the second time period. If the displacement is less than the threshold, it is judged as hidden continuous pressure. The characteristic changes of the area are compared before and after the body position adjustment or nursing intervention. If there is no improvement or the condition worsens, a high risk warning of pressure ulcers is output.

[0008] Furthermore, the preset frame rate is 1-10fps, and the duration of the continuous image acquisition covers at least the first preset acquisition period before the decompression event and the second preset acquisition period after the decompression event; the baseline features before decompression are obtained by taking the median or mean of the feature values ​​corresponding to the consecutive preset number of frames before the decompression event, and the baseline features are updated when it is determined that the skin area is in a reversible recovery state, and the baseline features are not updated when it is determined that the skin area is in a limited recovery state.

[0009] Furthermore, the recovery threshold includes the following relative deviation conditions: when the relative deviation between the post-decompression feature and the pre-decompression baseline feature does not exceed a preset proportional threshold, and the relative deviation does not increase within a preset time window, the recovery threshold is determined to be met.

[0010] Furthermore, for a continuous duration of not less than the first duration Within the observation interval, the feature deviations obtained frame by frame for the skin region are coupled with time accumulation and change suppression calculations to obtain the comprehensive skin recovery judgment value. The comprehensive judgment value of skin recovery mentioned above Obtained using the following formula:

[0011]

[0012] in, Indicates the start time of the first duration. This indicates the first duration; This represents the deviation of the skin region features at time t from the baseline features before decompression, and is a non-negative value; This represents a preset deviation threshold, used to normalize the deviation of the feature. This represents the deviation reinforcement coefficient, used to enhance the impact of continuous deviation on the determination quantity; This represents the change suppression coefficient, used to suppress instantaneous fluctuations in characteristic deviations; This indicates the rate of change of the deviation of the feature over time;

[0013] During the first duration Within, the comprehensive judgment value of skin recovery Under the condition of meeting the preset frame rate ratio, the value is greater than or equal to the recovery limitation judgment threshold. When this occurs, the skin area is determined to be in a state of limited recovery;

[0014] During the first duration The skin recovery comprehensive judgment value corresponding to the number of consecutive preset frames detected within the internal system. Less than or equal to the reversible recovery determination threshold When the skin area is in a reversible recovery state, the determination of the limited recovery state is terminated.

[0015] Furthermore, the displacement within the second time period is obtained by the change in the regional position of the skin region between adjacent frames, and the deformation is obtained by the rate of change of the area or the rate of change of the aspect ratio of the skin region contour.

[0016] Furthermore, the timing of the position adjustment or nursing intervention is determined by the spatial positional change conditions of the skin region in the bedside image sequence. The change conditions include the displacement of the center point of the skin region exceeding a preset displacement change threshold or the change in the orientation angle of the skin region exceeding a preset angle threshold.

[0017] Furthermore, the decompression event is triggered by the spatial position change of the skin region in the bedside image sequence. When the displacement of the center point of the skin region exceeds a preset displacement threshold or the change in the visible area of ​​the skin region exceeds a preset area threshold, the decompression event is determined to have occurred and the second preset acquisition period begins.

[0018] Furthermore, when the baseline feature before decompression is obtained using the median, it is used for bedside scenes with motion artifacts or local occlusion; when the baseline feature before decompression is obtained using the mean, it is used for bedside scenes with stable lighting and no occlusion, and the bedside scene is determined by whether the effective visible pixel ratio of the skin area is lower than a preset visibility threshold.

[0019] Furthermore, the preset frame percentage condition is as follows: among the total number of frames corresponding to the first duration, the percentage of frames that satisfy the condition that the feature deviation is greater than the preset deviation threshold is not less than the preset percentage threshold, and the frames that satisfy the condition cover at least two non-overlapping sub-time periods within the first duration.

[0020] Furthermore, when the skin region meets the occlusion condition within the first duration, the determination of the feature deviation is paused and the first duration is restarted after the occlusion is lifted; wherein, the occlusion condition includes the percentage of effective visible pixels in the skin region being lower than a preset visibility threshold.

[0021] The beneficial effects of this invention are as follows: By continuously acquiring images of the same skin area of ​​a patient at the ICU bedside and combining this with a temporal comparison of baseline features before and after decompression, the invention achieves dynamic identification of skin recovery capacity. It can not only distinguish between skin in a reversible recovery state and a limited recovery state, but also further identify hidden, persistent pressure conditions that are difficult for caregivers to detect visually. By introducing persistent feature deviation determination, frame rate constraints, and joint analysis of displacement and deformation, it effectively avoids misjudgments caused by instantaneous fluctuations, noise, or short-term posture changes, improving the stability and reliability of pressure ulcer risk identification and making risk assessment more consistent with the actual physiological evolution process.

[0022] Furthermore, this invention uses body positioning or nursing intervention as a key reference event, comparing changes in skin characteristics before and after the intervention. It only outputs a high-risk warning when the skin condition does not improve or shows a worsening trend, thus organically combining image recognition results with actual nursing actions and significantly reducing the probability of ineffective or excessive alarms. This method requires no additional contact sensors, is suitable for continuous, non-invasive bedside monitoring in the ICU, and can provide early warnings in the early stages of pressure ulcer formation or even before visible damage appears. This helps nursing staff take timely and targeted intervention measures, thereby reducing the incidence of pressure ulcers and improving the quality and efficiency of ICU nursing care. Attached Figure Description

[0023] Figure 1 This is a flowchart of the ICU bedside pressure ulcer risk identification and early warning logic of the present invention.

[0024] Figure 2 This is a diagram showing the relationship between bedside image acquisition and skin recovery assessment functions in the ICU according to the present invention.

[0025] Figure 3 This is a flowchart of the decompression event determination and skin feature monitoring process of the present invention.

[0026] Figure 4 This is a schematic diagram of the changes in skin brightness characteristics and baseline feature updates of the sacrum and coccyx before and after decompression in Embodiment 1 of the present invention.

[0027] Figure 5 This is a schematic diagram illustrating the calculation of the cumulative change of the comprehensive skin recovery judgment value over time in Embodiment 1 of the present invention.

[0028] Figure 6 This is a schematic diagram illustrating the changes in skin area displacement and the triggering of nursing intervention under concealed continuous pressure in Embodiment 1 of the present invention.

[0029] Figure 7 This is a schematic diagram of the decompression event triggering conditions and baseline feature selection in Embodiment 2 of the present invention.

[0030] Figure 8 This is a schematic diagram of the percentage of frames with skin features deviating from the frame and the determination of the occlusion area in Embodiment 2 of the present invention. Detailed Implementation

[0031] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0032] Combined with appendix Figure 1 This invention discloses an image recognition and early warning method for pressure ulcer risk at the ICU bedside. An image acquisition device is installed at a fixed location next to the ICU bed to continuously acquire images of the same skin area of ​​the patient under pressure and before and after decompression events at a preset frame rate, thereby obtaining an image sequence containing continuous changes in the skin area. The preset frame rate is set according to the monitoring needs of the ICU bedside, ensuring that the acquired image sequence fully reflects the changes in skin condition during pressure and decompression. During image acquisition, the acquisition area is locked to ensure that the image sequence corresponds to the same skin area, avoiding area shifts caused by changes in body position or camera shake.

[0033] After obtaining the image sequence, color features, brightness features, or texture features are extracted from the skin region in each frame of the image. These features are used to characterize changes in the skin's appearance. Before the decompression event occurs, baseline features before decompression are calculated from images corresponding to a preset number of consecutive frames, serving as reference features for the skin region under pressure. Subsequently, the skin features after the decompression event are compared frame by frame with the baseline features before decompression. When the comparison result meets a preset recovery threshold condition, the skin region is determined to be in a reversible recovery state, indicating that the skin can recover to its stable state before decompression.

[0034] When skin features after a decompression event fail to meet the recovery threshold condition, and the deviation of these features from the pre-decompression baseline persists for at least a first duration, the skin region is determined to be in a state of limited recovery. By introducing a duration condition, short-term fluctuations in skin features are suppressed, avoiding misjudgments caused by transient noise or occasional changes. This makes the assessment of skin recovery capacity more stable and reliable, providing a credible basis for subsequent identification of occult and persistent pressure and for early warning of pressure ulcer risks.

[0035] Once the skin region is determined to be in a state of limited recovery, the system continuously performs image analysis on the skin region for a subsequent second time period to further identify whether there is any hidden continuous pressure. During this second time period, based on continuously acquired bedside images, the system calculates the positional changes of the skin region between adjacent image frames to obtain regional displacement information of the skin region. Simultaneously, it analyzes the changes in the outer contour of the skin region to obtain deformation information reflecting changes in the skin surface morphology. The displacement characterizes the overall movement of the skin region on the bed surface or supporting structure, while the deformation characterizes the local deformation state of the skin region under continuous pressure conditions.

[0036] Within the second time period, when the detected regional displacement of the skin area is consistently less than a preset displacement threshold, and the deformation of the skin area is stable or limited, the skin area is determined to be in a state of hidden continuous pressure. By simultaneously considering both displacement and deformation, the system can effectively distinguish between apparent changes caused by short-term postural adjustments and skin condition fixation caused by genuine continuous pressure, thereby improving the accuracy of hidden continuous pressure identification. After determining the existence of a hidden continuous pressure state, the system further compares and analyzes the image features of the skin area before and after the postural adjustment or nursing intervention. When the color brightness or texture features of the skin area do not show an improvement trend after the nursing intervention, or show a worsening change compared to before the intervention, the system outputs corresponding high-risk pressure ulcer warning information to prompt nursing staff to take timely and targeted pressure relief or protective measures to reduce the risk of further pressure ulcer development.

[0037] Combined with appendix Figure 2Bedside image acquisition employs a preset frame rate for continuous acquisition, ranging from one to ten frames per second. This ensures that skin condition changes are adequately captured while minimizing unnecessary data redundancy and computational burden. The overall duration of continuous image acquisition covers at least a first preset acquisition period before the decompression event and a second preset acquisition period after the event. The first preset acquisition period characterizes the skin's stability under pressure, while the second acquisition period reflects the skin's recovery process after decompression. By maintaining continuous acquisition before and after the decompression event, the skin condition changes are ensured to have a complete temporal context, thus providing a reliable data foundation for subsequent judgment.

[0038] Before the decompression event occurs, color, brightness, or texture features of the skin region are extracted from images corresponding to a preset number of consecutive frames within a first preset acquisition period. The extracted feature values ​​are then statistically processed to obtain baseline features before decompression. The statistical processing is performed using either the median or the mean, depending on the quality of the bedside images. The median is used to suppress outliers when there is local occlusion or motion disturbance, while the mean is used when the image quality is stable to reflect the overall feature level. In subsequent analysis, if the skin region is determined to be in a reversible recovery state, it indicates that the skin can recover to a stable state after decompression. In this case, the baseline features are updated to reflect the latest health benchmark. If the skin region is determined to be in a limited recovery state, it indicates that the skin's recovery ability has decreased. In this case, the baseline features are not updated to avoid abnormal states being incorrectly introduced into the benchmark reference.

[0039] When determining whether skin meets the conditions for reversible recovery, a recovery threshold is introduced as a criterion. This recovery threshold includes a relative deviation condition: the skin features extracted after decompression are compared with the baseline features before decompression, and the relative deviation between the two is calculated. When this relative deviation does not exceed a preset proportional threshold, it indicates that the skin features have not undergone significant shift. Simultaneously, the trend of the relative deviation is analyzed within a preset time window. When the relative deviation remains stable or gradually decreases over time without increasing, it is determined to exhibit non-increasing change. When both the relative deviation amplitude condition and the trend condition are met, the skin region is determined to meet the recovery threshold, thus identifying the skin as being in a reversible recovery state.

[0040] Once the skin area enters the monitoring and analysis phase, it will be continuously monitored for no less than the first duration. Within the observation interval, the feature deviation values ​​obtained frame by frame for the skin region are comprehensively analyzed and processed. These feature deviation values ​​characterize the degree of change in the current features of the skin region relative to the baseline features before decompression. By accumulating these deviation values ​​over time and simultaneously introducing a mechanism to suppress the rate of change, a comprehensive skin recovery assessment metric reflecting the skin's recovery ability is obtained. This comprehensive judgment metric can simultaneously reflect both the persistence and stability of skin feature deviations, thus avoiding errors caused by judging solely based on instantaneous deviation values.

[0041] The comprehensive assessment of skin recovery Calculated using the following formula

[0042]

[0043] in, Indicates the start time of the first duration. The first duration is used to define the continuous observation interval of the skin recovery state; Indicates at time The deviation of the skin region features described below from the baseline features before decompression, wherein the deviation is a non-negative value, is used to reflect the magnitude of changes in skin condition; This represents a preset deviation threshold, used to normalize the deviation of the features, so that the deviations between different skin areas or different patients are comparable. This represents the deviation enhancement coefficient, used to enhance the contribution of feature deviation to the overall judgment value when it persists, thereby highlighting the impact of long-term deviation on the skin's recovery ability; This represents the change suppression coefficient, used to suppress the interference of rapid fluctuations in feature deviation over a short period of time on the judgment result; This indicates the rate of change of the deviation of the feature over time, used to characterize the smoothness of changes in skin condition.

[0044] During the first duration Internally, the comprehensive assessment of skin recovery is obtained. A status determination is performed. If, within the first time duration, the comprehensive determination value is greater than or equal to the recovery-limited determination threshold while satisfying the preset frame rate percentage condition,... When the skin region is determined to be in a state of restricted recovery, it indicates that the deviation in skin features persists in both time and magnitude, and the recovery ability is significantly reduced. When a comprehensive judgment value corresponding to a preset number of consecutive frames is detected within the first time period... Less than or equal to the reversible recovery determination threshold When the skin area is in a reversible recovery state, the determination of the limited recovery state is terminated, thereby avoiding the continued triggering of risk identification when the skin condition has obviously improved.

[0045] In the second analysis phase, based on continuously acquired bedside images, the spatial positional changes of the skin region between adjacent image frames are calculated to obtain displacement information. This displacement is obtained by comparing the changes in the center or overall position of the skin region in consecutive frames, reflecting the actual movement of the skin region on the bed surface or supporting structure. Simultaneously, the outer contour of the skin region is extracted, and its changes in consecutive frames are analyzed. Deformation information of the skin region is obtained by calculating the rate of change of the contour area or the rate of change of the contour aspect ratio. This deformation characterizes the degree of local deformation of the skin region under pressure, and is used in conjunction with the displacement information to distinguish between true continuous pressure and apparent changes caused by short-term postural adjustments.

[0046] The timing of postural adjustments or nursing interventions is determined through real-time analysis of bedside image sequences. A postural adjustment or nursing intervention is identified when a sudden change in the spatial position of the skin region is detected in consecutive image frames. The conditions for such a spatial change include a displacement of the skin region's center point exceeding a preset displacement change threshold within adjacent frames or time periods, or a significant change in the overall orientation of the skin region causing the orientation angle to exceed a preset angle threshold.

[0047] Combined with appendix Figure 3 By continuously analyzing bedside image sequences, the occurrence of a decompression event is determined using changes in the spatial location of the skin region as a trigger condition. When a significant change in the center point position of the skin region in consecutive image frames is detected, and this change exceeds a preset displacement threshold, it is determined that the pressure state experienced by the skin region has changed, thus identifying a decompression event. Furthermore, when the visible area of ​​the skin region in the bedside image changes significantly relative to the previous stable state, and this change exceeds a preset area threshold, a decompression event is also identified.

[0048] When determining the baseline features before the decompression event, different feature statistics methods are adaptively selected based on the quality of the bedside image and the visibility of the skin area. When motion artifacts caused by patient micro-movements or nursing operations exist in the bedside image, or when there is local occlusion in the skin area, the median method is preferentially used to statistically analyze the feature values ​​corresponding to a preset number of consecutive frames to reduce the impact of outliers on the baseline features. When the bedside image has stable lighting conditions and the skin area is unobstructed, the mean method is used to statistically analyze the feature values ​​corresponding to a preset number of consecutive frames to more accurately reflect the overall feature level of the skin area. The bedside scene is distinguished by whether the proportion of effectively visible pixels in the skin area is lower than a preset visibility threshold, thus ensuring that reliable baseline features before decompression can be obtained under different acquisition conditions, providing a stable reference for subsequent assessment of skin recovery status.

[0049] A preset frame percentage condition is introduced to determine the amount of feature deviation within the first time period. Specifically, in the total number of frames corresponding to the first time period, the number of image frames with skin region feature deviation greater than a preset deviation threshold is counted, and the percentage of such image frames in the total number of frames is calculated. When the percentage is not less than the preset percentage threshold, it indicates that the skin feature deviation is persistent over time. At the same time, to avoid misjudgment caused by deviations concentrated in a short period of time, it is further required that the image frames satisfying the percentage condition cover at least two non-overlapping sub-periods within the first time period in terms of time distribution, thereby ensuring that the feature deviation is not an instantaneous or sporadic change, but occurs repeatedly throughout the entire observation interval.

[0050] The visibility of the skin region is monitored to avoid interference from occlusion in the judgment results. When the skin region is detected to meet the occlusion condition within the first time period, the determination of feature deviation is paused. The occlusion condition is determined by analyzing the proportion of effectively visible pixels in the image of the skin region. When the proportion of effectively visible pixels is lower than a preset visibility threshold, the skin region is considered to be occluded. After the occlusion is removed, the first time period is restarted, and the determination of feature deviation is re-executed, thereby avoiding erroneous state judgments caused by local occlusion or short-term invisibility, and ensuring the continuity and accuracy of the skin recovery state analysis.

[0051] Example 1:

[0052] In this embodiment, a long-term bedridden patient in the ICU ward of a hospital was selected as the monitoring subject, with a focus on monitoring the skin area of ​​the sacrum and coccyx. A fixed image acquisition device was installed beside the bed, with the acquisition view covering the area of ​​the patient's sacrum and coccyx in contact with the bed surface. The system was set to a preset frame rate of 5fps, i.e., acquiring 5 frames per second. This frame rate balances the accuracy of capturing changes in skin appearance with the consumption of computational resources. The overall duration of continuous image acquisition covered a first preset acquisition period of 10 minutes before the decompression event and a second preset acquisition period of 20 minutes after the decompression event, thus ensuring complete image data support during both the pressure and decompression phases. Figure 4 As shown in the figure, the time-series changes in skin brightness characteristics before and after decompression, as well as the trend of baseline characteristics, are presented.

[0053] During the first preset acquisition period prior to the decompression event, the system acquired a total of 3000 images. Color and brightness features of the sacrococcygeal skin region were extracted from 300 consecutive frames, and the average brightness value was used as the feature quantity. Because the patient's position was generally stable during this period, but there were slight respiratory fluctuations, local brightness fluctuations appeared in some frames. Therefore, the system used the median method to calculate the pre-decompression baseline features. Statistically, the median of the skin region brightness features in the 300 images was 128, and this value was used as the pre-decompression baseline feature for subsequent comparative analysis. Figure 4 The dashed line represents the baseline brightness level before the decompression.

[0054] After the nursing staff turns the patient over, the system automatically recognizes the decompression event and enters the second preset data acquisition period. Images are continuously acquired and skin region features are extracted frame by frame within 20 minutes after decompression. Taking the first 5 minutes after decompression as an example, a total of 1500 frames are acquired, with the brightness features corresponding to the first 200 frames being 134, 132, 131, 130, and 129, respectively. The system calculates the relative deviation of the post-decompression features from the baseline feature 128. Taking the brightness value 134 as an example, its relative deviation is 6, accounting for approximately [percentage missing] of the baseline features. .

[0055] The system's preset ratio threshold is 10%, therefore the relative deviations corresponding to the above frames do not exceed the threshold. For example... Figure 4 As shown in the brightness curve after medium-pressure reduction, the overall brightness characteristics show a gradual downward trend.

[0056] Simultaneously, the system analyzes the trend of relative deviation within a preset time window. Using 60 consecutive seconds (300 frames) of images as a time window, the relative deviation gradually decreased from an initial 6 to 2 within this window, without showing an increasing trend, thus satisfying the non-increasing change condition. Based on this, the system determines that the skin region's characteristic changes after decompression meet the recovery threshold condition, classifying the skin as being in a reversible recovery state.

[0057] In the reversible recovery state, the system updates the baseline features before decompression. Specifically, 300 consecutive frames of images are selected from the stable phase after decompression, and the corresponding brightness feature values ​​are mainly concentrated in the range of 126 to 129. The system calculates the new baseline features using the mean method, and the updated baseline feature value is 127. Figure 4 The midpoint line indicates the updated baseline brightness level, which is used for subsequent monitoring to reflect the new stable state of the patient's skin after recovery.

[0058] In contrast, during another monitoring period, the system found that the skin area brightness characteristic remained around 140 for a long period after a certain decompression event, with a relative deviation of approximately .

[0059] Although the percentage did not exceed the threshold, it did not show a decreasing trend within 10 consecutive minutes, and even increased from 9.4% to 10.1% in some time windows. Based on this, the system determined that the relative deviation did not meet the non-incremental change condition, and the continuous deviation time exceeded the preset first duration of 15 minutes. Therefore, the system determined that the skin area was in a state of restricted recovery and maintained the original baseline characteristics without updating.

[0060] After completing the decompression event identification and entering the post-decompression monitoring phase, the system initiates the calculation process for the comprehensive skin recovery assessment. In this embodiment, the first duration... Set to 900 seconds, or 15 minutes, as the start time of the first duration. This is the moment immediately following the decompression event. The bedside image acquisition frame rate remains at 5fps, thus acquiring a total of 4500 valid images within the first duration. The system extracts the brightness features of the sacrococcygeal skin region frame by frame and calculates the feature deviation for each frame using the pre-decompression baseline feature value of 128 as a reference.

[0061] Within the first 15 minutes after decompression, the system statistically obtained the characteristic deviations at some representative moments as follows. At 60 seconds after decompression, the average skin brightness was 134, corresponding to the following deviation... The average brightness was 132 after 300 seconds of decompression, with a deviation of 4. The average brightness was 130 after 600 seconds of decompression, with a deviation of 2. The average brightness was 129 after 900 seconds of decompression, with a deviation of 1.

[0062] Preset deviation threshold The value is set to 10 to normalize the deviation; this is the deviation reinforcement coefficient. Set to 2 to enhance the impact of persistent deviation; change suppression coefficient. The value is set to 0.5 to suppress transient fluctuations. The system performs a time-cumulative and change-suppression coupled calculation on the deviation within the first time period according to the following formula to obtain the comprehensive skin recovery judgment value. :

[0063]

[0064] in, Indicates the start time of the first duration. Indicates the first duration. Indicates time The deviation of lower skin region features from baseline features before decompression. This indicates a preset deviation threshold. Indicates deviation from the strengthening coefficient. Indicates the coefficient of inhibition of change. This indicates the rate of change of the characteristic deviation over time.

[0065] In the discrete implementation, the system uses the difference in feature deviation between adjacent frames and the inter-frame time interval to approximate the rate of change. Taking an inter-frame time interval of 0.2 seconds as an example, when the deviation in adjacent frames gradually decreases from 6, the rate of change per unit time is a slightly negative value, indicating that the deviation is showing a slow decreasing trend.

[0066] The corresponding change suppression term is:

[0067] For the time period with a deviation of 6, the normalized deviation term is:

[0068]

[0069] The overall contribution per frame is approximately:

[0070]

[0071] The system calculates the overall skin recovery assessment value for the first time period by multiplying the contribution value of each frame within the first time period by the inter-frame time interval of 0.2 seconds and then summing and approximating the result. .like Figure 5 As shown, this value is significantly lower than the reversible recovery threshold.

[0072] Recovery Limitation Threshold Set to 300, the reversible recovery determination threshold. The value was set to 220. The system further checked the preset frame rate ratio condition. Within the first time period, there were 4500 frames, of which 0 frames had a deviation greater than the preset deviation threshold of 10, resulting in a ratio of 0, which did not meet the frame rate ratio condition for limited recovery. Simultaneously, the system repeatedly calculated the comprehensive judgment value within the observation window corresponding to 600 consecutive frames, and all results were obtained... The result is less than 220. Based on the above results, the system determines that the skin area is in a reversible recovery state and terminates further determination of the limited recovery state.

[0073] In contrast, during another monitoring period for the same patient, post-decompression skin brightness remained around 145, corresponding to a deviation of approximately 17, with a rate of change close to 0 within 15 minutes. Using the same parameters, the system calculated a single-frame normalized deviation of approximately:

[0074]

[0075] The change suppression term is close to 1, and the final cumulative result is... Meanwhile, 3600 out of 4500 frames had deviations exceeding the preset deviation threshold, accounting for 80%, and these deviations covered multiple non-overlapping time periods within the first duration. The system made this determination based on this. Greater than And if the frame rate ratio condition is met, it can be determined that the skin area is in a state of limited recovery. For example... Figure 5 As shown, in this case, the comprehensive judgment value is significantly higher than the recovery limitation threshold.

[0076] Once the system determines that the patient's sacrococcygeal skin region is in a state of limited recovery, it automatically enters the second-duration concealed continuous pressure analysis phase. The second duration is set to 600 seconds (10 minutes), with the bedside image acquisition frame rate remaining at 5fps, thus acquiring a total of 3000 consecutive images within the second duration. The system locates the sacrococcygeal skin region in each frame and calculates the positional change of this region between adjacent frames to obtain regional displacement data. Specifically, using the coordinates of the skin region's center point in the first frame as a reference, the system calculates the displacement distance of the center point in subsequent frames relative to this reference position.

[0077] During the actual monitoring process in the second time period, the system statistically determined that the maximum displacement of the skin region's center point in 3000 frames was 1.6 mm, and the average displacement was 0.9 mm. In this embodiment, the preset displacement threshold was set to 3 mm. Clearly, the above displacement data were all less than this threshold, indicating that the skin region did not experience significant overall movement during this time period. Simultaneously, the system continuously extracted the outer contour of the skin region and calculated the contour area change rate and aspect ratio change rate. Statistical results showed that the contour area change rate of the skin region in the second time period was mainly distributed between 1.2% and 1.8%, and the aspect ratio change rate was mainly distributed between 0.02 and 0.04, both within a stable range without significant fluctuations. Combining the results of displacement and deformation, the system determined that the skin region was under hidden continuous pressure during the second time period. Figure 6 As shown in the shaded area corresponding to the second duration, the displacement curve during this stage is always below the displacement threshold.

[0078] After determining the state of concealed and persistent pressure, the system continues to analyze the bedside image sequence to pinpoint the time of any repositioning or nursing intervention. Approximately 18 minutes after the start of monitoring, the nursing staff turned the patient over. Analysis of the image sequence revealed a sudden increase in the displacement of the skin region's center point to 12 mm in adjacent frames before and after this time point, significantly exceeding the preset displacement abrupt change threshold of 8 mm. Simultaneously, the overall orientation of the skin region changed significantly, with its orientation angle varying by approximately 25 degrees between adjacent frames, exceeding the preset angle threshold of 15 degrees. Based on this, the system automatically determined that a repositioning or nursing intervention had occurred and recorded this moment as the time of the intervention. Figure 6 The mutation location is shown in the figure.

[0079] After determining the timing of the nursing intervention, the system extracted and compared skin feature data from 5 minutes before and 5 minutes after the intervention. The results showed that before the intervention, the skin area brightness characteristic remained around 145, a deviation of approximately 17 from the pre-decompression baseline of 128. However, within 5 minutes after the intervention, the brightness characteristic only decreased to 143, with a corresponding deviation of 15, and no further downward trend was observed in the subsequent 300 frames; some frames even saw an increase to 146. Based on this, the system determined that the skin characteristics did not improve after the intervention and showed a slight deterioration trend. Based on the displacement and deformation analysis results and the feature comparison before and after the intervention, the system ultimately confirmed a persistent and hidden pressure ulcer risk in the sacrococcygeal skin area and output a high-risk pressure ulcer warning.

[0080] Example 2:

[0081] In this embodiment, a long-term bedridden patient in a hospital's ICU ward is still used as the monitoring subject, but the monitoring focus is expanded from the sacrum and coccyx to the patient's left greater trochanter region. The bedside image acquisition device remains in a fixed installation position, continuously acquiring image data of the patient's left greater trochanter region at a frame rate of 5fps. During continuous monitoring, the system automatically identifies the time of decompression events by performing real-time analysis of the bedside image sequence, using this time as the time boundary for subsequent analysis. Figure 7 As shown in the figure, the displacement of the center point of the skin region and the visible area change over time are illustrated.

[0082] The system triggers decompression events by analyzing the spatial position changes of the skin region across consecutive image frames. In each frame, the system calculates the coordinates of the skin region's center point and compares them to the coordinates of the center point in the previous stable state. A decompression event is determined to have occurred when the displacement of the skin region's center point exceeds a preset displacement threshold. In this embodiment, the preset displacement threshold is set to 6 mm. During monitoring, the patient adjusted to a side-lying position with the assistance of a caregiver. In adjacent frames, the system detected a sudden increase in the displacement of the skin region's center point from approximately 1.2 mm to approximately 8.5 mm, significantly exceeding the preset displacement threshold. Figure 7 As can be seen from the displacement curve, the displacement change has obvious temporal concentration. Based on this, the system determines that a decompression event has occurred and marks this moment as the starting reference moment of the second preset acquisition period.

[0083] In addition to center point displacement, the system also simultaneously monitors changes in the visible area of ​​the skin region in bedside images. When the patient's position changes, the skin area that was originally obscured by the bed surface or supporting structure is gradually exposed. The system detects that the visible area of ​​the skin region increases from approximately 3200 pixels to approximately 4100 pixels, corresponding to an area change ratio of approximately:

[0084]

[0085] This change exceeds a preset area threshold of 20%. For example... Figure 7 As shown in the visible area curve, the area change and the displacement abrupt change are highly consistent in time, and the system also determines that a decompression event has occurred based on this. By simultaneously introducing two triggering conditions, namely the displacement of the center point and the change in visible area, this embodiment can stably identify decompression events under different body position adjustment methods, avoiding missed or false judgments caused by relying on only a single feature.

[0086] Before a decompression event occurs, the system needs to determine the baseline characteristics prior to decompression. Due to slight spontaneous movement by the patient during monitoring, and occasional passing of caregivers by the bedside, localized occlusion and motion artifacts appear in the images. The system determines the bedside scene by analyzing the percentage of effectively visible pixels in the skin area. In 300 consecutive frames before the decompression event, the system calculated that the average percentage of effectively visible pixels in the skin area was approximately 68%, lower than the preset visibility threshold of 75%. Therefore, the current bedside scene was determined to have localized occlusion or motion artifacts.

[0087] Under the aforementioned scenario determination, the system uses the median method to statistically analyze the skin feature values ​​corresponding to 300 consecutive frames before the decompression event, obtaining the baseline features before decompression. In this embodiment, the brightness feature of the left greater trochanter region is distributed between 120 and 134 within 300 frames, with a median of 127. The system uses this value as the baseline feature before decompression for subsequent skin condition comparison and analysis. Using the median method can effectively suppress the influence of outliers caused by short-term occlusion or localized high-brightness reflection, making the baseline features more stable and reliable.

[0088] During another monitoring period, the patient's position remained stable, the bedside lighting conditions were good, and there were no significant obstructions. Before the decompression event, the system calculated that the percentage of effectively visible pixels in the skin area remained stable at around 92%, higher than the preset visibility threshold. Based on this, the system determined that the current bedside scene was a scene with stable lighting and no obstructions, and used the mean method to statistically analyze the skin feature values ​​of 300 consecutive frames. In this scene, the skin area brightness features were mainly concentrated in the range of 124 to 129, and the calculated mean value was 126.5. This value was used as the baseline feature before decompression for subsequent analysis.

[0089] After identifying the decompression event and determining the baseline features before decompression, the system enters the stability determination stage of the skin recovery state to distinguish between short-term fluctuations and true, persistent abnormalities. The first duration is set to 900 seconds, and the bedside image acquisition frame rate is maintained at 5fps, thus theoretically allowing for 4500 consecutive images within the first duration. Within this time frame, the system calculates the deviation of skin region features from the baseline features before decompression frame by frame, and introduces a preset frame rate ratio condition to constrain the temporal distribution characteristics of feature deviation. Figure 8 As shown in the figure, the temporal distribution of feature deviation frames and the changes in the proportion of visible pixels are presented.

[0090] The system counts the number of image frames with feature deviations exceeding a preset deviation threshold out of the 4500 frames corresponding to the first time period. In this embodiment, the preset deviation threshold is set to 10, corresponding to a significant deviation of the brightness feature from the baseline feature. Statistics show that within the 15-minute monitoring interval after decompression, approximately 3150 images have feature deviations exceeding this threshold, representing approximately the following percentage of the total frames:

[0091]

[0092] This ratio is significantly higher than the preset percentage threshold of 60%, thus meeting the frame rate percentage requirement.

[0093] Furthermore, the system analyzes the distribution of image frames that meet the deviation conditions along the time axis. Figure 8 It can be seen that the aforementioned feature deviation frames are mainly distributed in the two time intervals of 2 to 6 minutes and 9 to 14 minutes after decompression, with obvious intervals in between. Since the above two time intervals do not overlap in time and are both within the first duration, the system determines that the condition of covering at least two disjoint sub-time periods within the first duration is met, thereby further confirming that the feature deviation is persistent and repetitive, rather than caused by instantaneous anomalies.

[0094] During the initial monitoring period, the system also simultaneously detected the visibility of the skin area to avoid interference from occlusion in the judgment results. Due to a caregiver briefly approaching the bedside to adjust the infusion device, the skin area was partially obscured approximately 7 to 8 minutes after decompression. The system detected that the percentage of effectively visible pixels in the skin area decreased to approximately 58% over approximately 120 consecutive frames, falling below the preset visibility threshold of 75%. Figure 8 As shown in the occlusion interval, the system determines that the skin area meets the occlusion condition within that time period.

[0095] When the system detects that the occlusion condition is met, it immediately pauses the process of determining the feature deviation and freezes the timing state of the current first duration. After the occlusion is removed, the system recalculates the percentage of effectively visible pixels in the skin region. When this percentage recovers to at least 80% and remains so for more than 50 frames, the occlusion is determined to be removed. Subsequently, the system uses the consecutive effective image frames after the occlusion is removed as a new starting point to recalculate the effective observation interval corresponding to the first duration and continues to perform the frame percentage and temporal distribution analysis of the feature deviation.

[0096] Based on the combined judgment results of the frame rate ratio and time distribution conditions, the system finally confirmed that the skin region had significant feature deviations that occurred continuously and in multiple time periods during the first time period, providing a reliable basis for subsequent determination of the limited recovery state and analysis of hidden continuous pressure.

[0097] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. An image recognition and early warning method for bedside pressure ulcer risk in the ICU, characterized in that, include: Images of the same skin area of ​​the patient were continuously acquired at a preset frame rate before and after pressure and decompression to obtain a sequence. The color, brightness or texture features of the region are extracted from the sequence and compared with the baseline features before decompression. If the recovery threshold is met, it is judged as reversible; otherwise, if the deviation continues to exceed the first duration, it is judged as limited recovery. Under the restricted recovery state, the regional displacement and deformation are calculated within the second time period. If the displacement is less than the threshold, it is judged as hidden continuous pressure. Compare the changes in characteristics of the area before and after body positioning or nursing intervention. If there is no improvement or the condition worsens, a high risk warning for pressure ulcers will be issued.

2. The image recognition and early warning method for bedside pressure ulcer risk in the ICU according to claim 1, characterized in that, The preset frame rate is 1-10fps, and the duration of the continuous image acquisition covers at least the first preset acquisition period before the decompression event and the second preset acquisition period after the decompression event. The baseline features before decompression are obtained by taking the median or mean of the feature values ​​corresponding to a preset number of consecutive frames before the decompression event occurs. The baseline features are updated when the skin area is determined to be in a reversible recovery state, and the baseline features are not updated when the skin area is determined to be in a limited recovery state.

3. The image recognition and early warning method for bedside pressure ulcer risk in the ICU according to claim 1, characterized in that, The recovery threshold includes the following relative deviation conditions: when the relative deviation between the post-decompression feature and the pre-decompression baseline feature does not exceed a preset proportional threshold, and the relative deviation does not increase within a preset time window, the recovery threshold is determined to be met.

4. The image recognition and early warning method for bedside pressure ulcer risk in the ICU according to claim 1, characterized in that, The determination of limited recovery for continuous deviation exceeding the first duration includes: when the feature deviation of the skin region is greater than the preset deviation threshold under the condition of a preset frame rate for a continuous period of not less than the first duration, the skin region is determined to be in a limited recovery state; when the feature deviation corresponding to a consecutive preset number of frames is detected to meet the recovery threshold within the first duration, the skin region is determined to be in a reversible recovery state, and the determination of the limited recovery state is terminated.

5. The image recognition and early warning method for bedside pressure ulcer risk in the ICU according to claim 1, characterized in that, The displacement within the second time period is obtained by the change in the regional position of the skin region between adjacent frames, and the deformation is obtained by the rate of change of the area or the rate of change of the aspect ratio of the skin region contour.

6. The image recognition and early warning method for bedside pressure ulcer risk in the ICU according to claim 1, characterized in that, The timing of the position adjustment or nursing intervention is determined by the spatial positional change conditions of the skin region in the bedside image sequence. The change conditions include the displacement of the center point of the skin region exceeding a preset displacement change threshold or the change in the orientation angle of the skin region exceeding a preset angle threshold.

7. The image recognition and early warning method for bedside pressure ulcer risk in the ICU according to claim 2, characterized in that, The decompression event is triggered by the spatial position change of the skin region in the bedside image sequence. When the displacement of the center point of the skin region exceeds a preset displacement threshold or the change in the visible area of ​​the skin region exceeds a preset area threshold, the decompression event is determined to have occurred and the second preset acquisition period begins.

8. The image recognition and early warning method for bedside pressure ulcer risk in the ICU according to claim 2, characterized in that, When the baseline feature before decompression is obtained using the median, it is used for bedside scenes with motion artifacts or local occlusion; when the baseline feature before decompression is obtained using the mean, it is used for bedside scenes with stable lighting and no occlusion, and the bedside scene is determined by whether the effective visible pixel ratio of the skin area is lower than a preset visibility threshold.

9. The image recognition and early warning method for bedside pressure ulcer risk in the ICU according to claim 4, characterized in that, The preset frame percentage condition is as follows: among the total number of frames corresponding to the first duration, the percentage of frames that satisfy the condition that the feature deviation is greater than the preset deviation threshold is not less than the preset percentage threshold, and the frames that satisfy the condition cover at least two non-overlapping sub-time periods within the first duration.

10. The image recognition and early warning method for bedside pressure ulcer risk in the ICU according to claim 4, characterized in that, When the skin region meets the occlusion condition within the first duration, the determination of the feature deviation is paused and the first duration is restarted after the occlusion is lifted; wherein, the occlusion condition includes the percentage of effective visible pixels in the skin region being lower than a preset visibility threshold.

Images