Wound Management and Treatment Using Computer Vision and Machine Learning

Abstract

Certain aspects of the present disclosure provide techniques for wound management and treatment, including determining wound characteristics for a patient based on an image of the wound, including detecting characteristics based on analyzing the image using a first ML model. The techniques further include identifying patient medical data including characteristics related to a medical history for the patient, and predicting a first treatment plan for the patient based on providing the wound characteristics and the patient medical data to a second ML model. The first treatment plan is configured to be used to treat the wound for the patient.

Description

[Technical field]

[0001] (CROSS REFERENCE TO RELATED APPLICATIONS) This application claims priority to U.S. Patent Application No. 17 / 562,908, filed December 27, 2021, the contents of which are incorporated herein by reference in their entirety.

[0002] Introduction Aspects of the present disclosure relate to artificial intelligence and healthcare, and more specifically, to improved wound management and treatment using computer vision and machine learning (ML). [Background technology]

[0003] Managing and treating wounds in patients is a common healthcare goal. For example, a patient may incur a wound and seek treatment for the wound at a healthcare facility, or the patient may incur or aggravate the wound while entering a healthcare facility or receiving managed care at an outpatient facility. Managing and treating these wounds is difficult because different wounds may require different treatments and may take different amounts of time to heal, depending on the characteristics of the wound and the patient. Furthermore, generating a treatment plan for treatment is typically performed manually by a treatment provider. This requires an on-site inspection and assessment of the wound and manual creation or modification of a treatment plan to treat the wound. However, this approach is prone to inaccuracies, for example, due to the potential for human error, and is inefficient because it requires an on-site assessment and treatment plan generation by the treatment provider. Summary of the Invention [Means for solving the problem]

[0004] An embodiment provides a method. The method includes determining a plurality of characteristics of a wound for a patient based on an image of the wound, the plurality of characteristics including detecting a plurality of characteristics based on analyzing the image with a first machine learning (ML) model. The method further includes identifying patient medical data including a plurality of characteristics related to a medical history for the patient. The method further includes predicting a first treatment plan for the patient based on providing the plurality of characteristics of the wound and the patient medical data to a second ML model. The first treatment plan is configured to be used to treat the wound for the patient.

[0005] A further embodiment provides an apparatus including a memory and a hardware processor communicatively coupled to the memory, the hardware processor configured to perform operations. The operations include determining a plurality of characteristics of the wound for the patient based on the image of the wound, the plurality of characteristics including detecting the plurality of characteristics based on analyzing the image using the ML model. The operations further include identifying patient medical data including a plurality of characteristics related to a medical history for the patient. The operations further include predicting a first treatment plan for the patient based on providing the plurality of characteristics of the wound and the patient medical data to a second ML model. The first treatment plan is configured to be used to treat the wound for the patient.

[0006] A further embodiment provides a non-transitory computer readable medium including instructions that, when executed by a processor, cause the processor to perform operations. The operations include determining a plurality of characteristics of the wound for the patient based on an image of the wound, including detecting a plurality of characteristics based on analyzing the image using the ML model. The operations further include identifying patient medical data including a plurality of characteristics related to a medical history for the patient. The operations further include predicting a first treatment plan for the patient based on providing the plurality of characteristics of the wound and the patient medical data to a second ML model. The first treatment plan is configured to be used to treat the wound for the patient.

[0007] The following description and the annexed drawings set forth in detail certain illustrative features of the one or more embodiments. [Brief description of the drawings]

[0008] The accompanying drawings depict certain aspects of one or more embodiments and are not therefore to be considered as limiting the scope of the present disclosure.

[0009] [Figure 1] FIG. 1 depicts a computing environment for wound management and treatment using computer vision and ML, according to one embodiment.

[0010] [Diagram 2] FIG. 2 depicts a block diagram of a predictive controller for wound management and treatment using computer vision and ML, according to one embodiment.

[0011] [Diagram 3] FIG. 3 is a flow chart illustrating wound management and treatment using computer vision and ML, according to one embodiment.

[0012] [Figure 4] FIG. 4 illustrates detecting wound characteristics from captured images using computer vision, according to one embodiment.

[0013] [Diagram 5] FIG. 5 depicts an example of detecting wound characteristics from captured images using computer vision, according to one embodiment.

[0014] [Figure 6] FIG. 6 is a flowchart illustrating training a computer vision ML model for wound management and treatment, according to one embodiment.

[0015] [Figure 7] FIG. 7 depicts predicting a wound treatment plan using an ML model, according to one embodiment.

[0016] [Figure 8] FIG. 8 depicts wound characteristics for use in predicting a wound treatment plan using an ML model, according to one embodiment.

[0017] [Figure 9] FIG. 9 depicts patient characteristics for use in predicting a wound treatment plan using an ML model, according to one embodiment.

[0018] [Figure 10] FIG. 10 depicts a patient history for use in predicting a wound treatment plan using an ML model, according to one embodiment.

[0019] [Figure 11] FIG. 11 depicts historical wound treatment incident data for use in predicting a wound treatment plan using an ML model, according to one embodiment.

[0020] [Figure 12] FIG. 12 is a flowchart illustrating training an ML model for wound management and treatment using computer vision, according to one embodiment.

[0021] [Figure 13] FIG. 13 depicts using a wound treatment plan generated using an ML model, according to one embodiment.

[0022] [Figure 14] FIG. 14 depicts ongoing monitoring of a patient procedure for wound management and treatment using computer vision, according to one embodiment.

[0023] To facilitate understanding, the same reference numerals have been used, where possible, to designate like elements common to the figures. It is contemplated that elements and features of one embodiment may be beneficially incorporated in other embodiments without further recitation. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0024] Detailed Description Aspects of the present disclosure provide an apparatus, method, processing system, and non-transitory computer-readable medium for improved wound management and treatment using computer vision and ML. As discussed above, patient wounds are typically treated using a treatment plan that outlines various treatment tasks. In existing practices, the treatment plan is generally created manually by the treatment provider after inspecting the wound (e.g., in person). However, this is inefficient because it requires manual intervention, and it may not be effective because it is subject to human error and differences between treatment professionals. Alternatively, in existing practices, the treatment plan may be created using a predefined rubric or algorithm with predefined rules. This is also inefficient because it requires a large number of predefined rules and significant manual oversight, and is ineffective because using a specific rubric or algorithm is highly unlikely to be effective for all patients and all wounds. Thus, existing practices may generally lead to inconsistent and ineffective patient treatment outcomes.

[0025] In aspects described herein, a treatment plan for treating a patient wound can instead be automatically created using a trained ML model based on captured images of the wound or other captured sensor data. For example, a patient or a treatment provider can capture an image of the patient wound. Computer vision techniques (e.g., suitable ML models as further discussed below) can be used to analyze the image and detect various characteristics of the wound from the image. A suitable predictive ML model (e.g., a deep learning neural network (DNN)) can be trained to predict a treatment plan for the patient wound based on the detected wound characteristics and additional information about the patient. For example, the predictive ML model can use patient characteristics (e.g., demographic information, prescription drug information, and assessment information) and patient medical history (e.g., prior medical conditions and treatments for the patient) along with the detected wound characteristics to predict a treatment plan for the wound. The treatment plan can outline a set of treatment tasks to be followed when treating the wound. Beneficially, this can provide both technical advantages and advantages in treating the patient. For example, as discussed further below, this can provide technical advantages over conventional techniques in the healthcare field by reducing the computational burden and shifting it from prediction time (when computational resources may be in use and results are more likely to be time-dependent) to an earlier training phase (when computational resources can be scheduled and are more likely to be freely available). Additionally, and as discussed further below, this provides therapeutic benefits to patients by providing more accurate and consistent treatment plans, enabling preventative early treatment when high priority problems are identified, and enabling rapid adjustments to treatment plans based on real-time monitoring.

[0026] In an embodiment, a predictive ML model can be trained to predict treatment plans using data about historical wound treatment incidents. For example, the predictive ML model can receive data about prior patient wounds, including associated patient and wound characteristics, treatment plans used, facilities used, and treatment resolution. As described above, this data can be used to train the ML model to predict treatment plans for newly identified wounds based on wound (e.g., detected from images using computer vision techniques) and patient characteristics.

[0027] Further, the patient can be continuously monitored during treatment of the wound (e.g., automatically using suitable sensors or manually by the treatment provider), and the predictive ML model can update the predicted treatment plan based on the monitoring data. For example, additional images of the wound can be captured during treatment, computer vision techniques can be used to detect characteristics of the wound from the captured images as it is treated, and the predictive ML model can use the updated characteristics to predict a revised treatment plan for the wound. Further, the progress of the wound's treatment can be used to continuously train the predictive ML model and improve future predictions.

[0028] Thus, aspects described herein provide significant advantages over conventional approaches for generating treatment plans. For example, using a trained ML model to automatically predict a treatment plan for treating a patient wound based on captured images of the wound or other captured sensor data provides an accurate treatment plan while minimizing the required computational resources for prediction and shifting the computational burden from prediction time (e.g., when a near real-time response may be required) to earlier training time (e.g., when resources can be easily dedicated to training). In an embodiment, generating a treatment plan using a specific rubric or algorithm with predefined rules can be computationally expensive because a large number of rules are required and parsing and following the rules is computationally expensive. Furthermore, this computationally expensive analysis is performed at the time the treatment plan is generated when a rapid response is likely to be required (e.g., so that the patient can be treated quickly).

[0029] Using a trained ML model to automatically predict a treatment plan for treating a patient wound, in contrast, is significantly less computationally expensive at the time the treatment plan is generated. For example, a predictive ML model can be trained in advance during a training phase when a rapid response is not required and computational resources are readily available. The trained ML model can then be used to quickly and computationally relatively inexpensively predict a treatment plan for the patient. This provides a significant technical advantage over conventional techniques by shifting the computational burden from prediction times, when a rapid response is required and computational resources can be engaged in other tasks, to planned training times, when a rapid response is not required and computational resources are available.

[0030] As another example, using a trained ML model to automatically predict a treatment plan for treating a patient wound based on captured images of the wound or other captured sensor data provides a more accurate and well-defined prediction. In an embodiment, the treatment plan for the wound may be created manually by the treatment provider. However, this leaves a risk of human error and allows for significant differences between human practitioners, which may result in a lack of certainty in the accuracy of the treatment plan. Predicting the treatment plan using a trained ML model can both reduce the risk of human error and provide more certainty in the level of accuracy of the treatment plan. Furthermore, the predicted treatment plan itself can be reviewed and refined by the treatment provider. This provides a starting point for the treatment provider with a more certain level of accuracy, reducing the burden on the treatment provider to generate the treatment plan themselves.

[0031] Exemplary Computing Environment FIG. 1 depicts a computing environment 100 for wound management and treatment using computer vision and ML, according to one embodiment. In an embodiment, a captured wound image 102 is provided to a detection layer 110. For example, a patient may have a wound (e.g., a pressure sore, a suture wound, an abrasion, an injury, or any other visible wound) that is detectable using an image capture device. A patient, a healthcare professional, a caregiver, or any other person may capture an image of the wound using an image capture device (e.g., a digital camera). For example, a patient or healthcare professional may use a camera integrated into a smartphone or tablet computer to capture the wound image 102 and may provide the image to the detection layer 110 using a suitable secure application. This is merely one example, and any suitable image capture device may be used by any suitable person or entity to capture the wound image 102. For example, an automated sensor may be used to automatically trigger image capture of the wound image 102 (e.g., during a medical exam). Additionally, the image capture device may operate outside the visible spectrum (eg, an infrared sensor, an x-ray sensor, or any other suitable sensor).

[0032] In an embodiment, the captured wound image 102 is provided to the detection layer 110 using a suitable communication network. For example, the wound image 102 can be captured using a camera in a computing device (e.g., a smartphone or tablet computer camera) and transferred to the detection layer using the computing device. The computing device can use any suitable communication network, including the Internet, a wide area network, a local area network, or a cellular network, and can use any suitable wired or wireless communication technique (e.g., WiFi or cellular communication). This is merely one example, and the wound image 102 can be captured by a camera and provided to the computing device using any suitable technique (e.g., using a storage medium or through wired or wireless transmission from the camera to the computing device).

[0033] The detection layer 110 includes a wound detection service 112, which includes a wound detection ML model 114. In an embodiment, the wound detection service 112 facilitates the transformation of incoming patient data (e.g., wound images 102). For example, as discussed below with respect to FIG. 2, the wound detection service 112 may be a computer software service implemented in a suitable controller (e.g., predictive controller 200 illustrated in FIG. 2) or combination of controllers. In an embodiment, the detection layer 110 and the wound detection service 112 can be implemented using any suitable combination of physical computing systems, cloud computing nodes and storage locations, or any other suitable implementation. For example, the detection layer 110 can be implemented using a server or a cluster of servers. As another example, the detection layer 110 can be implemented using a combination of computing nodes and storage locations in a suitable cloud environment. For example, one or more of the components of the detection layer 110 can be implemented using a public cloud, a private cloud, a hybrid cloud, or any other suitable implementation.

[0034] As one example, the wound detection service 112 may facilitate computer vision analysis of the wound image 102. In this example, the wound detection ML model 114 may be a suitable computer vision ML model (e.g., a DNN, a support vector machine (SVM), or any other suitable ML model). In an embodiment, the wound detection ML model 114 may receive the wound image 102 and be trained to recognize or detect various characteristics of the wound depicted in the image. These may include external characteristics (e.g., size and color), internal characteristics (e.g., size, color, and depth), location, and any other suitable characteristics. This is discussed further below with respect to Figures 4-5 and 8.

[0035] In an embodiment, the wound image 102 is simply one example of patient data that may be analyzed using the detection layer 110 (e.g., using the wound detection service 112 and the wound detection ML model 114). For example, captured sensor data 104 may also be provided to the detection layer 110. In an embodiment, the captured sensor data 104 includes data captured by sensors used during the treatment or rehabilitation of a patient (e.g., captured during treatment of a wound). For example, the captured sensor data 104 may include data from negative pressure wound therapy devices, oxygen and intubation devices, monitored pressure and drainage devices, or any other suitable devices.

[0036] In an embodiment, the wound detection service 112 can further facilitate analysis of the captured sensor data 104. For example, the wound detection service 112 can use a wound detection ML model 114 to detect and identify characteristics of a patient's wound based on the captured sensor data. In an embodiment, the wound detection ML model 114 can be any suitable ML model (e.g., DNN, decision tree, random forest, support vector machine, and other ML model types) that is trained to detect and identify characteristics of a patient's wound.

[0037] Further, in an embodiment, the wound detection ML model 114 can include multiple ML models trained to detect wound characteristics from different data. For example, one ML model can be trained to use computer vision techniques to identify wound characteristics from wound images 102, another ML model can be trained to detect wound characteristics based on sensor data from a wound therapy device, and another ML model can be trained to detect wound characteristics based on sensor data from a monitored pressure device. In some aspects, these different models may be ensembled to produce a prediction. This is merely an example, and the wound detection ML model can instead be trained to detect and identify characteristics of a patient's wound using data from multiple sources together (e.g., wound images 102 and captured sensor data 104).

[0038] In an embodiment, the detection layer 110 provides wound detection data to the prediction layer 120. For example, the wound detection service 112 can use a wound detection ML model 114 to detect characteristics of a patient wound using the wound images 102, the captured sensor data 104, or both. The detection layer 110 can provide these wound characteristics to the prediction layer 120.

[0039] The prediction layer 120 includes a wound prediction service 122 and a wound predictive ML model 124. In one embodiment, the wound prediction service 122 facilitates predicting treatment and rehabilitation information for a patient wound. For example, the wound prediction service 122 can use the wound predictive ML model 124 to determine a treatment prediction 150 (e.g., a wound treatment plan) and predict any other suitable treatment and rehabilitation information for a patient wound. This is discussed further below with respect to FIG. 7.

[0040] As discussed below with respect to FIG. 2, the wound prediction service 122 may be a computer software service implemented in a suitable controller (e.g., the predictive controller 200 illustrated in FIG. 2) or combination of controllers. In an embodiment, the prediction layer 120 and the wound prediction service 122 may be implemented using any suitable combination of physical computing systems, cloud computing nodes and storage locations, or any other suitable implementation. For example, the prediction layer 120 may be implemented using a server or a cluster of servers. As another example, the prediction layer 120 may be implemented using a combination of computing nodes and storage locations in a suitable cloud environment. For example, one or more of the components of the prediction layer 120 may be implemented using a public cloud, a private cloud, a hybrid cloud, or any other suitable implementation.

[0041] As discussed above, the prediction layer 120 uses the detected characteristics of the patient wound (e.g., output from the detection layer 110) to predict treatment and rehabilitation information for the patient wound. However, in some embodiments, the wound characteristics detected by the detection layer 110 are not sufficient to enable the prediction layer 120 to accurately predict treatment and rehabilitation information for the patient wound. For example, simply identifying the characteristics of the wound may not be sufficient to identify a suitable treatment plan for the patient, and may not be sufficient to identify a suitable treatment facility for the patient.

[0042] In an embodiment, the predictive layer 120 can further receive and use patient medical data 130 and historical wound treatment data 140. For example, the patient medical data 130 can include patient characteristics 132 and patient medical history 134. In an embodiment, the patient characteristics 132 can include patient demographics (e.g., age, height, weight), patient prescriptions (e.g., a prescription drug list for the patient), patient assessment data (e.g., admission assessment data, discharge assessment data, activities of daily living (ADL) assessment data), or any other suitable patient characteristics. This is discussed further below with respect to FIG. 9. In an embodiment, the patient medical history 134 can include medical condition data (e.g., diagnosis, onset, treatment, and resolution) for any prior medical condition. This is discussed further below with respect to FIG. 10.

[0043] In an embodiment, the historical wound care data 140 may include data about inpatient outcomes 142 and outpatient outcomes 144 for various patients and various wounds. For example, the historical wound care data 140 may include wound characteristics (e.g., external characteristics, internal characteristics, and location) for the wound, patient characteristics (e.g., demographics, prescribed medications, assessments, and medical history) for the patient with the wound, treatment plan history (e.g., treatments used) for the wound, facility characteristics (e.g., type of facility, staffing at the facility, and available resources at the facility) for the treatment of the wound, resolution data (e.g., time and resources used in the treatment, and outcomes of the treatment), and any other suitable historical wound care data. In an embodiment, the patient medical data 130 provides data about a particular patient with a wound, while the historical wound care data 140 provides data about historical treatments and resolutions for various wounds and patients. Additionally, in an embodiment, the historical wound care data 140 has been stripped of any personally identifiable patient information.

[0044] In an embodiment, the patient medical data 130 and the historical wound treatment data 140 are provided to the predictive layer 120 using a suitable communication network. For example, the patient medical data 130 and the historical wound treatment data 140 can be stored in one or more suitable electronic databases (e.g., a relational database, a graph database, or any other suitable database) or other electronic repository (e.g., a cloud storage location, an on-premise network storage location, or any other suitable electronic repository). The patient medical data 130 and the historical wound treatment data 140 can be provided to the predictive layer 120 from the respective electronic repositories using any suitable communication network, including the Internet, a wide area network, a local area network, or a cellular network, and can use any suitable wired or wireless communication technique (e.g., WiFi or cellular communication).

[0045] As discussed above, in an embodiment, the wound prediction service 122 uses a wound prediction ML model 124 to predict treatment and rehabilitation information for a patient wound. For example, the wound prediction ML model 124 may be a suitable supervised ML model (e.g., DNN) that is trained to generate treatment predictions 150 for a patient wound from a combination of wound characteristics (e.g., output from the detection layer 110), patient medical data 130, and historical wound treatment data 140 for the particular wound in question. This is discussed further below with respect to FIG. 3. For example, the wound prediction ML model 124 may be selected based on an initial analysis of the input data (e.g., wound characteristics, patient medical data 130, and historical wound treatment data 140). In an embodiment, a basic technique may be selected first (e.g., logistic regression), the data may be converted to a numerical format, and based on the initial analysis, a data transformation and ML technique may be selected. This is merely an example, and any suitable supervised or unsupervised technique may be used.

[0046] For example, the wound predictive ML model can predict a treatment plan for the wound, including recommended treatments and prescribed medications. This is one example of a treatment prediction 150. In an embodiment, the treatment plan (or any other suitable treatment prediction 150) can be provided to a treatment facility 160. In an embodiment, the treatment facility 160 can be any suitable inpatient or outpatient treatment facility. Additionally, in an embodiment, the treatment plan can be provided directly to the patient or the patient's medical treatment provider. This is further discussed below with respect to FIG. 13. In an embodiment, the treatment prediction 150 is provided to any or all of the treatment facility, the patient, and the treatment provider using a suitable communication network. For example, the treatment prediction 150 can be provided from the prediction layer 120 to a destination (e.g., a treatment facility, a patient, or a treatment provider) using any suitable communication network, including the Internet, a wide area network, a local area network, or a cellular network, and can use any suitable wired or wireless communication technique (e.g., WiFi or cellular communication).

[0047] In an embodiment, the treatment prediction 150 is used to treat a patient. For example, the treatment prediction 150 can be a wound treatment plan provided to a treatment facility 160. A treatment provider at the treatment facility 160 or the patient themselves can use the wound treatment plan to treat the wound (e.g., using the identified treatments and prescribed medications). In an embodiment, the treatment of the wound can be monitored and ongoing patient monitoring data 170 can be collected. For example, repeated images of the wound can be captured, other sensor data can be provided, the treatment provider can provide assessment data, or any other suitable data can be collected. Further, in an embodiment, the captured data can be maintained in a suitable repository (e.g., an electronic database) and used for training (e.g., training the wound prediction ML model 124). This data and all training data can be stripped of any personally identifying patient information.

[0048] In an embodiment, this ongoing patient monitoring data 170 can be provided to the detection layer 110, the prediction layer 120, or both and used to refine the treatment prediction 150. For example, captured images or other captured sensor data can be provided to the detection layer 110 and analyzed in the same manner as the wound images 102 and captured sensor data 104 (e.g., to identify ongoing wound characteristics as the wound is treated). As another example, updated patient medical data can be provided to the prediction layer 120 and analyzed in the same manner as the patient medical data 130.

[0049] Further, in an embodiment, the ongoing patient monitoring data 170 can be used to continuously train the wound prediction ML model 124. For example, the wound prediction ML model 124 can determine from the ongoing patient monitoring data 170 (e.g., from detected wound characteristics of additional captured images of the wound as it is being treated) whether the wound is progressing in treatment and how quickly it is progressing. As an example, the wound color, shape, size, condition (e.g., oozing or drying), or depth may change during treatment to indicate progression in healing. The wound prediction service 122 can use the prior predicted treatment plan, and the outcome of the treatment as indicated by the ongoing patient monitoring data as additional training data, to further train the wound prediction ML model 124 and predict a treatment plan that provides a successful treatment to the patient.

[0050] 2 depicts a block diagram of a predictive controller 200 for wound management and treatment using computer vision and ML, according to one embodiment. The controller 200 includes a processor 202, a memory 210, and a network component 220. The memory 210 may take the form of any non-transitory computer-readable medium. The processor 202 generally reads and executes programming instructions stored in the memory 210. The processor 202 represents a single central processing unit (CPU), multiple CPUs, a single CPU with multiple processing cores, a graphics processing unit (GPU) with multiple execution paths, and the like.

[0051] Network component 220 includes components necessary for controller 200 to interface with a suitable communications network (e.g., a communications network that interconnects various components of computing environment 100 illustrated in FIG. 1 or that interconnects computing environment 100 with other computing systems). For example, network component 220 can include wired, WiFi, or cellular network interface components and associated software. Although memory 210 is illustrated as a single entity, memory 210 may include one or more memory devices having blocks of memory associated with physical addresses, such as random access memory (RAM), read-only memory (ROM), flash memory, or other types of volatile and / or non-volatile memory.

[0052] The memory 210 generally includes program code for implementing various functions associated with the use of the predictive controller 200. The program code is generally described as various functional "applications" or "modules" within the memory 210, although alternative implementations may have different functions and / or combinations of functions. Within the memory 210, the wound detection service 112 uses the wound detection ML model 114 to facilitate detecting wound characteristics from captured sensor data (e.g., captured images and other captured sensor data). This is discussed further below with respect to Figures 4-6. The wound prediction service 122 uses the wound prediction ML model 124 to facilitate predicting treatment and rehabilitation information related to wounds. This is discussed further below with respect to Figures 3 and 7.

[0053] Although controller 200 is illustrated as a single entity, in an embodiment, the various components may be implemented using any suitable combination of physical computing systems, cloud computing nodes and storage locations, or any other suitable implementation. For example, controller 200 may be implemented using a server or a cluster of servers. As another example, controller 200 may be implemented using a combination of computing nodes and storage locations in a suitable cloud environment. For example, one or more of the components of controller 200 may be implemented using a public cloud, a private cloud, a hybrid cloud, or any other suitable implementation.

[0054] 2 depicts the wound detection service 112, the wound prediction service 122, the wound detection ML model 114, and the wound prediction ML model 124 as being co-located with one another in the memory 210, but that representation is also provided merely as an illustration for clarity. More generally, the controller 200 may include one or more computing platforms, such as, for example, computer servers, that may be co-located or may form a bi-directionally linked but distributed system, such as, for example, a cloud-based system. As a result, the processor 202 and the memory 210 may correspond to distributed processor and memory resources within the computing environment 100. Thus, it should be understood that any or all of the wound detection service 112, the wound prediction service 122, the wound detection ML model 114, and the wound prediction ML model 124 may be stored remotely from one another in the distributed memory resources of the computing environment 100.

[0055] 3 is a flowchart 300 illustrating wound management and treatment using computer vision and ML, according to one embodiment. At block 302, a wound detection service (e.g., wound detection service 112 illustrated in FIGS. 1-2) receives captured sensor data associated with a patient wound. For example, as discussed above in connection with FIG. 1, in an embodiment, the wound detection service may receive a captured wound image (e.g., wound image 102 illustrated in FIG. 1), captured sensor data (e.g., captured sensor data 104 illustrated in FIG. 1), or both.

[0056] In block 304, the wound detection service uses the ML model to detect wound characteristics from the captured data. For example, the wound detection service may use the captured image, the sensor data, or both to detect external characteristics (e.g., size and color), internal characteristics (e.g., size, color, and depth), location, and any other suitable characteristics of the wound. As discussed above in connection with the wound detection ML model 114 illustrated in FIG. 1, the wound detection service may use any suitable ML model or combination of ML models to detect wound characteristics from the captured sensor data. This is discussed further below in connection with FIGS. 4-6.

[0057] At block 306, a prediction service (e.g., wound prediction service 122 illustrated in Figures 1-2) receives the patient medical data. For example, the prediction service may receive the patient medical data 130 illustrated in Figure 1. This may include patient characteristics (e.g., patient demographics, patient prescription medications, patient assessment data, or any other suitable patient characteristics) and patient medical history (e.g., medical condition data for any prior medical conditions). This is discussed further below with respect to Figures 9-10.

[0058] At block 308, the prediction service receives historical wound treatment data. For example, the prediction service may receive historical wound treatment data 140 illustrated in FIG. 1. This may include historical data on inpatient and outpatient outcomes for various patients and various wounds. This is discussed further below with respect to FIG. 11. In an embodiment, the prediction service uses the historical wound treatment data for ongoing training of the predictive ML model. Alternatively, the prediction service does not receive the historical wound treatment data. In this example, the historical wound treatment data is used to train the predictive ML model (e.g., as discussed below with respect to FIG. 12), but is not used for inference (e.g., for prediction).

[0059] In block 310, the predictive service uses the ML model to predict a treatment plan for the patient wound. For example, the predictive service may use the wound predictive ML model 124 illustrated in FIG. 1-2 to predict a treatment plan. The predictive ML model may be any suitable ML model trained to predict a treatment plan for the patient wound using wound characteristics (e.g., detected from the sensor data captured using the ML model in block 304), the patient medical data received in block 306, and the historical wound treatment data received in block 308. This may include predicting medical treatments, prescription medications, and any other suitable treatments for the patient wound. This is discussed further below with respect to FIG. 7.

[0060] As shown, the predictive ML model uses all of the wound characteristics, patient medical data, and historical wound treatment data to predict a treatment plan. However, this is merely an example. Alternatively, or in addition, the predictive ML model can use any subset of this data (e.g., if some of this data is not available for a given patient wound). For example, the predictive ML model can use wound characteristics and patient medical data without the historical wound treatment data, or wound characteristics and historical wound treatment data without the patient medical data. In an embodiment, this may result in a small loss of accuracy in predicting the treatment plan, but the predicted treatment plan will still be significantly improved over conventional techniques (e.g., manual creation of a treatment plan).

[0061] In an embodiment, the predictive service can further identify preventative treatment tasks for the wound (e.g., treatment tasks intended to quickly prevent further disease or problems associated with the wound). For example, the predictive service can use wound characteristics, patient medical data, including but not limited to specific health-related data associated with one or more patients, such as age, weight, medical conditions, demographics, or other such data, or both, to identify high priority treatment tasks (e.g., prescription medication, bandage, or another medical procedure) required for a wound (e.g., a pressure sore, a wound, an abrasion, a wound, or any other wound). As an example, a wound can be identified as requiring immediate medical treatment (e.g., a bandage, a surgical procedure, a specific prescription medication, or any other suitable treatment) to prevent further disease or problems associated with the wound. Thus, for example, a pressure sore, a wound, an abrasion, or a wound can be identified as requiring immediate medication, immediate bandage, or another immediate medical procedure. The predictive service can transmit an alert (e.g., an email, SMS message, phone call, or another form of electronic message) describing the treatment task to a treatment provider for the patient (e.g., via a treatment facility for the patient) or to the patient himself / herself. The treatment provider or patient can then use the treatment task to treat the wound. In an embodiment, the predictive service can identify this treatment task prior to completing a prediction of the treatment plan. For example, the predictive service can identify high priority treatment tasks while predicting the treatment plan and transmit an alert prior to completing a prediction of the treatment plan. In an embodiment, this allows for a quick alert regarding the treatment task without waiting for a complete prediction of the treatment plan.

[0062] In block 312, the predictive service receives ongoing data from the treatment monitoring. For example, the predictive service can receive additional sensor data (e.g., additional images) captured during treatment and rehabilitation of the patient's wound. This data can be captured at a treatment facility (e.g., an inpatient or outpatient facility) by a suitable medical professional or by the patient themselves. In an embodiment, the predictive service can use the ongoing data to further refine the wound treatment plan.

[0063] Example of Detecting Wound Characteristics from Captured Images Figure 4 illustrates detecting wound characteristics from captured images using computer vision, according to one embodiment. In an embodiment, Figure 4 provides an example of detecting wound characteristics from captured data using an ML model, discussed above in connection with block 304 illustrated in Figure 3. A wound image 102 (e.g., as discussed above in connection with Figure 1) is provided to a computer vision service 410 and a computer vision ML model 412. In an embodiment, the wound image 102 is an image of a patient wound captured using any suitable image capture device (e.g., a camera, a medical imaging device, or any other suitable image capture device).

[0064] In one embodiment, the computer vision service 410 is an example of the wound detection service 112, and the computer vision ML model 412 is an example of the wound detection ML model 114, both of which are illustrated in Figures 1-2. As discussed above, in one embodiment, the wound detection service 112 can use the wound detection ML model to detect wound characteristics from various captured sensor data, including captured images or captured sensor data from a treatment device. The computer vision service 410 uses the computer vision ML model 412 to detect wound characteristics 420 from the wound image 102.

[0065] In an embodiment, the computer vision ML model 412 can be any suitable ML model. For example, a non-neural network ML model can be used (e.g., SVM), which can use any suitable object detection, recognition, or identification technique. As another example, a neural network ML model can be used (e.g., CNN), which can use any suitable object detection, recognition, or identification technique.

[0066] As discussed above, wound characteristics 420 can include any suitable wound characteristics. These can include external characteristics (e.g., size and color), internal characteristics (e.g., size, color, and depth), location, and any other suitable characteristics. This is discussed further below with respect to FIG. 8.

[0067] FIG. 5 illustrates an example of detecting wound characteristics from a captured image using computer vision, according to one embodiment. In an embodiment, the captured image depicts a wound on a patient. As discussed above, a suitable wound detection service (e.g., computer vision service 410 illustrated in FIG. 4) detects wound characteristics from the image using a suitable wound detection ML model (e.g., computer vision ML model 412 illustrated in FIG. 4). For example, the wound detection service can detect exterior size 502 and exterior color 508. As another example, the wound detection service can detect interior size and color 506 and depth 504.

[0068] Example of training a computer vision ML model FIG. 6 is a flow chart 600 illustrating training a computer vision ML model for wound management and treatment, according to one embodiment. This is merely an example, and in an embodiment, a suitable unsupervised technique may be used (e.g., without requiring training). In block 602, a training service (e.g., a human administrator or a software or hardware service) collects historical wound image data. For example, a wound detection service (e.g., wound detection service 112 illustrated in FIGS. 1 and 2) may be configured to act as a training service and collect previously captured (e.g., collected over time) images of a patient wound. This is merely an example, and any suitable software or hardware service may be used (e.g., a wound detection training service).

[0069] At block 606, a training service (or other suitable service) preprocesses the collected historical wound image data. For example, the training service may create a feature vector reflecting values ​​of various features for each collected wound image. At block 608, the training service receives the feature vectors and uses them to train a trained computer vision ML model 412 (e.g., computer vision model 412 illustrated in FIG. 4).

[0070] In an embodiment, at block 604, the training service also collects additional wound data (e.g., data generated from an in-person assessment of the wound). At block 606, the training service may also pre-process this additional wound data. For example, feature vectors corresponding to the historical wound image data may be further annotated using the additional wound data. Alternatively, or in addition, additional feature vectors corresponding to the additional wound data may be created. At block 608, the training service uses the pre-processed additional wound data during training to generate a trained computer vision ML model 412.

[0071] In an embodiment, the pre-processing and training can be performed as batch training. In this embodiment, all data is pre-processed at once (e.g., all historical wound image data and additional wound data) and provided to the training service in block 608. Alternatively, the pre-processing and training can be performed in a streaming manner. In this embodiment, the data is streaming and is continuously pre-processed and provided to the training service. For example, it may be desirable to take a streaming approach for scalability. The training data set may be very large, and therefore it may be desirable to pre-process the data and provide it to the training service in a streaming manner (e.g., to avoid computation and storage limitations). Furthermore, in an embodiment, a federated learning approach may be used in which multiple healthcare entities contribute to training a shared model.

[0072] Example of predicting wound treatment plan FIG. 7 depicts predicting a wound treatment plan using an ML model, according to one embodiment. In an embodiment, FIG. 7 corresponds to block 310 illustrated in FIG. 3 above. A wound prediction service 122, as discussed above in connection with FIGS. 1-2, is associated with a treatment plan predictive ML model 712. In an embodiment, the treatment plan predictive ML model 712 is an example of a wound predictive ML model (e.g., an example of the wound predictive ML model 124 illustrated in FIGS. 1-2). For example, as illustrated, the wound prediction service 122 uses the treatment plan predictive ML model 712 to predict a wound treatment plan 720.

[0073] In an embodiment, the wound prediction service 122 uses multiple types of data to predict the wound treatment plan 720 using the treatment plan predictive ML model 712. For example, the wound prediction service 122 can use the detected wound characteristics 702. In an embodiment, the detected wound characteristics 702 are generated by detecting the wound characteristics from the captured data (e.g., the wound image 102, the captured sensor data 104, or both) by a wound detection service (e.g., the wound detection service 112 illustrated in FIG. 1-2) using a wound detection ML model (e.g., the wound detection ML model 114 illustrated in FIG. 1-2). For example, as illustrated in FIG. 4, the computer vision service 410 can detect the wound characteristics 420 from the wound image 102 using the computer vision ML model 412. As discussed below in connection with FIG. 8, in an embodiment, the detected wound characteristics 702 can include external characteristics (e.g., size, color), internal characteristics (e.g., size, color, depth), location, and any other suitable characteristics.

[0074] Additionally, the wound prediction service 122 can use patient characteristics 132 (e.g., as discussed above in connection with FIG. 1 ) to predict a wound treatment plan 720 using the treatment plan predictive ML model 712. As discussed below in connection with FIG. 9 , the patient characteristics 132 can include patient demographics (e.g., age, height, weight), patient prescription medications (e.g., a prescription drug schedule for the patient), patient assessment data (e.g., admission assessment data, discharge assessment data, activities of daily living (ADL) assessment data), or any other suitable patient characteristics.

[0075] Additionally, the wound prediction service 122 can use the patient history 134 (e.g., as discussed above in connection with FIG. 1 ) to predict the wound treatment plan 720 using the treatment plan predictive ML model 712. As discussed below in connection with FIG. 10 , the patient history 134 can include medical condition data (e.g., diagnosis, onset, treatment, and resolution) for any prior medical conditions.

[0076] The wound prediction service 122 can further use historical wound treatment data 140 (e.g., as discussed above in connection with FIG. 1) to predict wound treatment plans 720 using the treatment plan predictive ML model 712. As discussed below in connection with FIG. 11, the historical wound treatment data 140 can include wound characteristics (e.g., external characteristics, internal characteristics, and location) for the wound, patient characteristics (e.g., demographics, prescribed medications, assessments, and medical history) for the patient with the wound, treatment plan history (e.g., treatments used), facility characteristics for the treatment of the wound (e.g., type of facility, staffing at the facility, and available resources at the facility), resolution data (e.g., time and resources used in the treatment, and outcomes of the treatment), and any other suitable historical wound treatment data. As discussed above in connection with FIG. 1, in an embodiment, the patient characteristics 132 and patient medical history 134 provide data about a particular patient with a wound, while the historical wound treatment data 140 provides data about historical treatments and resolutions for various wounds and patients.

[0077] In an embodiment, the wound prediction service 122 uses the historical wound treatment data 140 for ongoing training of the treatment predictive ML model 712. For example, because training the treatment predictive ML model 712 can be computationally expensive, the wound prediction service can train the treatment predictive ML model 712 at a suitable interval (e.g., hourly, daily, weekly) or based on a trigger event (e.g., after a threshold number of new observations are received, upon request from an administrator, or at any other suitable interval). Alternatively, the wound prediction service 122 does not receive the historical wound treatment data 140. In this example, the historical wound treatment data 140 is used to train a predictive ML model (e.g., as discussed below in connection with FIG. 12), but is not used for inference (e.g., for predicting the wound treatment plan 720).

[0078] In an embodiment, the wound treatment plan 720 provides a treatment plan for treating a patient wound. For example, the wound treatment plan 720 can include a set of tasks (e.g., medication tasks, treatment tasks, rehabilitation tasks, physical training tasks, or any other suitable tasks) to be performed by the patient, the patient's healthcare provider, the patient's caregiver, or other support personnel. The wound treatment plan 720 can be predicted by the treatment plan predictive ML model 712 such that adherence to the wound treatment plan 720 would provide the patient with an optimal or preferred treatment. As discussed above, treatment plans are typically generated manually (e.g., by a healthcare provider) or programmatically using specific rubrics or algorithms. This can be inefficient (e.g., because it requires manual intervention) and ineffective. In an embodiment, the wound treatment plan 720 generated using the treatment plan predictive ML model 712 provides a treatment that is both effective and efficient. Additionally, in an embodiment, the healthcare provider can review the generated wound treatment plan 720 and provide any suitable revisions. This can further greatly improve the efficiency and effectiveness of developing treatment plans by assisting health care providers.

[0079] In an embodiment, the wound treatment plan 720 may include treatment tasks related to actions to be taken by the patient. For example, the wound treatment plan 720 may include information related to preferred nutrition for the patient. In this example, the patient's compliance with the preferred nutrition may further be identified during treatment (e.g., using sensors available at the location where the patient is being treated). As another example, the wound treatment plan 720 may include information about preferred humidity levels for treating the patient's wound. Humidity levels in the patient's living facility may be monitored (e.g., using suitable sensors) and the patient may be prompted or assisted to maintain the preferred humidity levels for treatment of the wound.

[0080] As another example, the wound treatment plan 720 may include a sleep therapy task. For example, the wound treatment plan 720 may outline the amount of sleep and the sleep position. In this example, a patient with a wound (e.g., a bedsore) on a particular location on his / her body may be treated by describing a sleep position, duration, or both for the patient to help treat the wound (e.g., a position or duration that relieves pressure on the wound). The patient's sleep may be monitored and the patient may be assisted in complying with the sleep therapy task. For example, one or more sensors (including, for example, a smart wearable device, a smart sleep device, an image capture sensor, or any other suitable sensor) may monitor the patient while sleeping and identify when the patient is not in REM sleep. If the patient is sleeping in a position that is not recommended for treating the patient's wound, the patient may be awakened when not in REM sleep and prompted or assisted to move to a therapy position for further sleep.

[0081] Exemplary Wound and Patient Characteristics 8 depicts exemplary wound characteristics 800 for use in predicting a wound treatment plan using an ML model, according to one embodiment. In an embodiment, the wound characteristics 800 provide an example of detected wound characteristics 702 illustrated in FIG. 7 and generated using a wound detection ML model suitable for detecting characteristics from captured wound data (e.g., captured wound images). For example, the wound characteristics 800 may include one or more wounds 802.

[0082] In one embodiment, each wound 802 includes external characteristics 810. External characteristics 810 include size 812. For example, size 812 can describe the external size of wound 802 (e.g., the size of the area surrounding the open area of ​​the wound or the area surrounding the more severe portion of the wound). In one embodiment, size 812 can describe an area (e.g., mm 2), dimensions, perimeter circumference, or using any other suitable technique. For example, size 812 can be expressed as a function that describes the external size of the wound.

[0083] The external characteristics 810 may further include color 814. For example, the color 814 may describe the color of an external portion of the wound. The color 814 may be the average color across the external area, the most extreme color across the external area (e.g., the darkest color, the lightest color, the color containing the greatest percentage of a particular shade, etc.), or any other suitable color. Additionally, the color 814 may be represented using a numerical value, a tuple (e.g., red, green, blue (RGB) values), a textual indicator, or using any other suitable technique.

[0084] In an embodiment, the external characteristics can further include a regularity 816 (e.g., a regularity in the shape of the wound) and a condition 818 (e.g., a condition of the exterior of the wound). For example, the condition 818 can describe whether the wound is dry or oozing, whether it is sutured or stapled, or any other suitable condition. These are merely examples, and the external characteristics 810 can include any suitable characteristics.

[0085] In some embodiments, the wound 802 further includes interior characteristics 820. The interior characteristics 820 include a size 822. For example, the size 812 can describe the interior size of the wound (e.g., the open area of ​​the wound or the size of the more severe portions of the wound). In some embodiments, the size 822 can describe the area (e.g., mm 2 ), dimensions, perimeter circumference, or using any other suitable technique. For example, size 822 can be expressed as a function that describes the internal size of the wound.

[0086] The interior characteristics 820 may further include color 824. For example, the color 824 may describe the color of an interior portion of a wound. The color 824 may be the average color over the interior area, the most extreme color over the interior area (e.g., the darkest color, the lightest color, the color containing the greatest percentage of a particular hue, etc.), or any other suitable color. Additionally, the color 824 may be represented using a numerical value, a tuple (e.g., red, green, blue (RGB) values), a textual indicator, or using any other suitable technique.

[0087] The interior characteristics 820 can further include depth 826. For example, the depth 826 can describe the depth of a wound. This can include tissue depth for an open or closed wound and can be expressed using measurements (e.g., mm), relative to the surface area of ​​the skin, using markings, or using any other suitable technique. These are merely examples and the interior characteristics 820 can include any suitable characteristics.

[0088] In an embodiment, the interior characteristics can further include regularity 828 (e.g., regularity of the shape of the wound) and condition 830 (e.g., condition of the interior of the wound). For example, condition 830 can describe whether the wound is dry or oozing, whether it is sutured or stapled, or any other suitable condition.

[0089] In an embodiment, wound 802 further includes location 840. For example, location 840 can describe the location of the wound on the patient's body. In an embodiment, location 840 can be described relative to a portion of the patient's body, using a measurement system, or using any other suitable technique. External characteristics 810, internal characteristics 820, and location 840 are merely examples, and wound 802 can include any suitable characteristics organized in any suitable manner.

[0090] 9 depicts patient characteristics 900 for use in predicting a wound treatment plan using an ML model, according to one embodiment. In an embodiment, the wound characteristics 900 provide examples for the patient characteristics 132 described above in connection with FIG. 1. The patient 902 includes patient demographics 910. For example, the patient demographics 910 may include age 912, height 914, and weight 916. These are merely examples, and the patient demographics 910 may include any suitable characteristics.

[0091] Patient 902 may further include patient prescriptions 920. In one embodiment, patient prescriptions 920 includes one or more prescription drugs 922A-N. These are merely examples, and patient prescriptions 920 may include any suitable data.

[0092] Additionally, patient 902 may include one or more patient assessments 930 (e.g., a patient assessment 930 corresponding to each healthcare facility at which the patient is admitted). In an embodiment, patient assessment 930 includes an admission assessment 932. For example, an admission assessment may be performed for the patient upon admission to the healthcare facility (e.g., performed by a suitable healthcare professional, using a suitable automated assessment system, or both). The admission assessment may be recorded as an admission assessment 932.

[0093] In an embodiment, patient assessment 930 further includes a discharge assessment 934. For example, a discharge assessment can be performed for a patient upon discharge from a healthcare facility (e.g., performed by a suitable healthcare professional, using a suitable automated assessment system, or both). The discharge assessment can be recorded as discharge assessment 934.

[0094] Patient assessment 930 may further include activities of daily living (ADL) assessment 936. For example, the ADL assessment may record the patient's ability to dress, eat, walk, toilet, and perform their own hygiene. The ADL assessment may be recorded as ADL assessment 936. These are merely examples, and patient assessment 930 may include any suitable data. Additionally, patient demographics 910, patient prescriptions 920, and patient assessment 930 are merely examples. Patient 902 may include any suitable patient data organized in any suitable manner.

[0095] 10 depicts a patient history 1000 for use in predicting a wound treatment plan using an ML model, according to one embodiment. In an embodiment, the patient history 1000 provides an example for the patient history 134, described above in connection with FIG.

[0096] The patient 1002 includes one or more medical conditions 1010A-N. Each medical condition includes a respective diagnosis 1012A-N, a respective onset description 1014A-N (e.g., a date or a text description), a respective treatment 1016A-N (e.g., a treatment history for the medical condition), and a respective resolution 1018A-N (e.g., a date of resolution or an indication that the medical condition is ongoing). These are merely examples, and each medical condition 1010A-N may include any suitable data. Additionally, the medical conditions 1010A-N are merely examples, and the patient 1002 may include any suitable medical history data.

[0097] 11 depicts historical wound treatment incident data 1100 for use in predicting a wound treatment plan using an ML model, according to one embodiment. In an embodiment, the historical wound treatment incident data 1100 provides an example for the historical wound treatment data 140, described above in connection with FIG. 1. Additionally, in an embodiment, the historical wound treatment incident data 1100 corresponds to any suitable patient (e.g., in addition to the patient whose wound is being treated). For example, the historical wound treatment incident data 1100 can be maintained (e.g., in a suitable anonymized or private format) by a health care provider.

[0098] The historical wound treatment incidents 1102 include patient characteristics 1110. In an embodiment, the patient characteristics 1110 correspond to the patient characteristics 900 illustrated in Figure 9 (e.g., for patients with historical wounds). The patient characteristics 1110 include demographics 1112 (e.g., age, height, weight) and medical history 1114. These are merely examples and the patient characteristics 1110 may include any suitable data.

[0099] The historical wound treatment incidents 1102 further include wound characteristics 1120. In an embodiment, the wound characteristics 1120 correspond to the patient characteristics 800 illustrated in Figure 8 (e.g., for the associated historical wound). The wound characteristics 1120 include external characteristics 1122 (e.g., size, color), internal characteristics 1124 (e.g., size, color, depth), and location 1126. These are merely examples and the wound characteristics 1120 may include any suitable data.

[0100] The historical wound treatment incident 1102 further includes a treatment plan history 1130. For example, the treatment plan history 1130 may describe one or more treatments 1132A-N that were used to treat the associated wound. These are merely examples, and the treatment plan history 1130 may include any suitable data.

[0101] The historical wound treatment incidents 1102 further include one or more facility characteristics 1140 (e.g., describing any facility used to treat the wound, including outpatient and inpatient facilities). The facility characteristics 1140 include type 1142 (e.g., inpatient, outpatient, or any other suitable type), staffing data 1144 (e.g., describing the number and type of staffing at the facility), and resource data 1146 (e.g., describing available resources, including equipment, staffing, prescription medications, and any other suitable resources). These are merely examples, and the facility characteristics 1140 may include any suitable data.

[0102] The historical wound care incident 1102 further includes a resolution 1150. For example, the resolution 1150 may include a time 1152 (e.g., time of resolution), resources 1154 (e.g., equipment, staffing, and other resources used in the resolution), and an outcome 1156 (e.g., end result of treatment). These are merely examples, and the resolution 1150 may include any suitable data. Additionally, the patient characteristics 1110, wound characteristics 1120, treatment plan history 1130, facility characteristics 1140, and the resolution 1150 are merely examples. The historical wound care incident 1102 may include any suitable data.

[0103] Example of training an ML model to predict wound treatment regimens FIG. 12 is a flowchart 1200 illustrating training an ML model for wound management and treatment using computer vision, according to one embodiment.

[0104] In block 1202, a training service (e.g., a human administrator or a software or hardware service) collects historical wound treatment data. For example, a wound prediction service (e.g., wound prediction service 122 illustrated in FIGS. 1 and 2) can be configured to act as the training service and collect the historical wound treatment data. This is merely an example and any suitable software or hardware service can be used (e.g., a wound prediction training service).

[0105] At block 1204, a training service (or other suitable service) pre-processes the collected historical wound treatment data. For example, the training service may create, for each historical wound, a feature vector reflecting values ​​of various features.

[0106] In block 1206, the training service receives the feature vectors and uses them to train a trained treatment plan predictive ML model 712 (e.g., as discussed above in connection with FIG. 7).

[0107] In an embodiment, the pre-processing and training can be performed as batch training. In this embodiment, all data is pre-processed at once (e.g., all historical wound image data and additional wound data) and provided to the training service at 1206. Alternatively, the pre-processing and training can be performed in a streaming manner. In this embodiment, the data is streaming and continuously pre-processed and provided to the training service. For example, it may be desirable to take a streaming approach for scalability. The training data set may be very large, and therefore it may be desirable to pre-process the data and provide it to the training service in a streaming manner (e.g., to avoid computation and storage limitations).

[0108] Example of using predicted wound treatment plan 13 depicts using a wound treatment plan generated using an ML model, according to one embodiment. In an embodiment, a predictive controller 1310 (e.g., the predictive controller 200 illustrated in FIG. 2) generates a predicted treatment plan 1320. For example, as discussed above in connection with block 310 of FIGS. 3 and 7, a wound prediction service (e.g., the wound prediction service 122 illustrated in FIGS. 1-2) can use a wound prediction ML model (e.g., the wound prediction ML model 124 illustrated in FIGS. 1-2) to predict a treatment plan.

[0109] For example, the wound prediction service can use detected wound characteristics generated from captured sensor data (e.g., captured images of the wound) using a wound detection service (e.g., wound detection service 112 illustrated in FIG. 1) and a wound detection ML model (e.g., wound detection ML model 114 illustrated in FIGS. 1-2). As discussed above, FIG. 8 provides an example of wound characteristics. The wound prediction service can further use any or all of patient characteristics (e.g., as illustrated in FIG. 9), patient medical history (e.g., as illustrated in FIG. 10), and historical wound treatment incidents (e.g., as illustrated in FIG. 11). In an embodiment, the wound prediction service uses historical wound treatment data for ongoing training of the wound detection ML model. Alternatively, the wound prediction service does not receive historical wound treatment data.

[0110] In an embodiment, the predictive controller 1310 transmits the predicted treatment plan 1320 to any or all of the patient 1340, the treatment provider 1350, and the healthcare facility 1360 via a communications network 1330. The communications network 1330 may be any suitable communications network, including the Internet, a wide area network, a local area network, or a cellular network, and may use any suitable wired or wireless communications technique (e.g., WiFi or cellular communications).

[0111] In an embodiment, any or all of the patient 1340, treatment provider 1350, and healthcare facility 1360 receive the predicted treatment plan. The predicted treatment plan 1320 can then be used to treat the patient wound. For example, the patient 1340 can receive the predicted treatment plan 1320 on a suitable electronic device (e.g., a smart phone, tablet, laptop computer, desktop computer, or any other suitable device) and use it for treatment (e.g., using a mobile or local application running on the patient device or accessing the predicted treatment plan 1320 via the communication network 1330).

[0112] Similarly, the treatment provider 1350 or the healthcare facility 1360 (e.g., a health care professional at the healthcare facility 1360) can receive the predicted treatment plan 1320. In an embodiment, any or all of the patient 1340, the treatment provider 1350, and the healthcare facility 1360 store the predicted treatment plan 1320. For example, this can allow the recipient to access the predicted treatment plan 1320 without requiring a continuous network connection.

[0113] 14 depicts ongoing monitoring of a patient treatment for wound management and treatment using computer vision, according to one embodiment. As discussed above in connection with the ongoing patient monitoring data 170 illustrated in FIG. 1, in an embodiment, the patient treatment plan can be revised based on ongoing monitoring of the treatment progress of the patient's wound. In an embodiment, the patient is treated at an outpatient facility 1430. The outpatient facility 1430 continues to monitor the treatment of the wound.

[0114] For example, the patient or treatment provider can continue to capture electronic images of the wound as it is treated or capture electronic sensor data during treatment. The patient or treatment provider can transmit this outpatient monitoring data 1432 (e.g., captured images or other sensor data) to the predictive controller 1410 (e.g., the predictive controller 200 illustrated in FIG. 2) using a communications network 1420. The communications network 1420 can be any suitable communications network, including the Internet, a wide area network, a local area network, or a cellular network, and can use any suitable wired or wireless communications technique (e.g., WiFi or cellular communications).

[0115] In an embodiment, the predictive controller 1410 can use the outpatient monitoring data 1432 to revise the predicted treatment plan. For example, as discussed above in connection with FIG. 4, a computer vision service can use a computer vision ML model to identify wound characteristics from captured wound images. These wound characteristics can then be used to predict a wound treatment plan using a treatment plan predictive ML model (e.g., as discussed above in connection with FIG. 7). The outpatient monitoring data 1432 can include one or more additional captured images of the wound, and a suitable computer vision ML model can be used to detect wound characteristics from these images. The predictive controller 1410 can then use the updated wound characteristics to predict an updated wound treatment plan.

[0116] Alternatively or additionally, the patient is treated at a healthcare facility 1440. As in the outpatient facility 1430, the patient's wound can be continuously monitored (e.g., by the treatment provider or by the patient) at the healthcare facility 1440. The treatment provider or patient can transmit facility monitoring data 1442 (e.g., updated captured sensor data regarding the wound) to the predictive controller 1410 using the communication network 1420. The predictive controller 1410 can use the facility monitoring data 1442 to revise the predicted treatment plan. For example, as discussed above in connection with FIG. 4, a computer vision service can use a computer vision ML model to identify wound characteristics from captured wound images. These wound characteristics can then be used to predict the wound treatment plan using a treatment plan predictive ML model (e.g., as discussed above in connection with FIG. 7). The facility monitoring data 1442 can include one or more additional captured images of the wound, and a suitable computer vision ML model can be used to detect wound characteristics from these images. The predictive controller 1410 can then use the updated wound characteristics to predict an updated wound treatment plan.

[0117] Further, in an embodiment, the outpatient monitoring data 1432 and the facility monitoring data 1442 can be used to continuously train the treatment plan predictive ML model. For example, the outpatient monitoring data 1432 and the facility monitoring data 1442 can include additional captured images of the wound during treatment. Computer vision services can be used to identify characteristics of these wounds, and from these characteristics, the predictive ML model can identify the extent to which treatment is progressing for the patient. This indication of progress, along with previously predicted treatment plans, can be used as training data to further refine the treatment plan predictive ML model.

[0118] Example notes Implementation examples are described in the numbered appendices below.

[0119] Appendix 1: A method comprising: determining a plurality of characteristics of a wound for a patient based on an image of the wound, the plurality of characteristics including detecting the plurality of characteristics based on analyzing the image using a first machine learning (ML) model trained to detect wound characteristics from a captured image; and predicting a first treatment plan for the patient based on identifying patient medical data comprising a plurality of characteristics related to a medical history for the patient and providing the plurality of characteristics of the wound and the patient medical data to a second ML model, the second ML model trained to predict the first treatment plan using prior wound treatment outcome data comprising a plurality of prior wound treatment outcomes associated with a plurality of prior patients, the first treatment plan being configured to be used to treat the wound for the patient.

[0120] Appendix 2: The method of any of Appendixes 1 or 3-10, further comprising determining a second plurality of wound characteristics for the patient based on analyzing a second image of the wound captured during treatment of the wound associated with the predicted first treatment plan using the first ML model, and predicting a second treatment plan for the patient based on providing the second plurality of wound characteristics to a second ML model.

[0121] Appendix 3: The method of any of appendices 1-2 or 4-10, wherein the second plurality of characteristics of the wound is further used to modify a second ML model through further training based on the second plurality of characteristics and the first treatment plan.

[0122] Addendum 4: The method of any of Addendums 1-3 or 5-10, further comprising identifying a preventive treatment task for the wound based on at least one of a plurality of characteristics of the wound or a treatment plan, and transmitting an electronic alert related to the treatment task.

[0123] Appendix 5: The method of any of appendices 1-4 or 6-10, wherein identifying the preventative treatment task is identified using a second ML model based on at least one of a plurality of characteristics of the wound, and further comprising electronically transmitting an alert to a treatment provider for the patient using a communications network prior to completing predicting the treatment plan for the patient.

[0124] Addendum 6: The method of any of Addendums 1-5 or 7-10, wherein detecting a plurality of characteristics of the wound further includes determining at least one of wound depth, color, or size based on an image of the wound.

[0125] Appendix 7: The method of any of appendices 1-6 or 8-10, wherein the prior wound treatment outcome data comprises data reflecting wound characteristics, treatments, and resolution for each of a plurality of prior wounds associated with a plurality of prior patients.

[0126] Appendix 8: The method of any of appendices 1-7 or 9-10, wherein the treatment plan comprises one or more recommended treatment tasks for the wound, comprising at least one of a prescription medication, a patient action, or a treatment provider action for treating the wound.

[0127] Addendum 9: The method of any of Addendums 1-8 or 10, further comprising modifying treatment of the wound based on one or more recommended treatment tasks.

[0128] Appendix 10: The method of any of Appendixes 1-9, further comprising identifying additional data captured by the sensor during wound treatment or assessment, and predicting the first treatment plan is further based on providing the identified additional data to a second ML model.

[0129] Addendum 11: A processing system comprising a memory having computer-executable instructions and one or more processors configured to execute the computer-executable instructions and cause the processing system to perform a method according to any one of Addendums 1-10.

[0130] Addendum 12: A processing system, comprising means for carrying out the method according to any one of Addendums 1-10.

[0131] Addendum 13: A non-transitory computer-readable medium comprising computer-executable instructions that, when executed by one or more processors of a processing system, cause the processing system to perform a method as described in any one of Addendums 1-10.

[0132] Addendum 14: A computer program product, embodied on a computer-readable storage medium, comprising code for performing the method according to any one of Addendums 1-10.

[0133] Additional Considerations The foregoing description is provided to enable any person skilled in the art to practice the various embodiments described herein. The examples discussed herein are not limitations of the scope, applicability, or embodiments described in the claims. Various modifications of these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be applied to other embodiments. For example, changes may be made in the function and arrangement of elements discussed without departing from the scope of the disclosure. Various examples may omit, substitute, or add various procedures or components, as appropriate. For example, the methods described may be performed in different orders than those described, and various steps may be added, omitted, or combined. Also, features described with respect to some examples may be combined in some other examples. For example, an apparatus may be implemented, or a method may be practiced, using any number of aspects described herein. In addition, the scope of the disclosure is intended to cover such apparatus or methods practiced using other structure, functionality, or structure and functionality in addition to or other than the various aspects of the disclosure described herein. It should be understood that any aspect of the disclosure disclosed herein may be embodied by one or more elements of a claim.

[0134] As used herein, the word "exemplary" means "serving as an example, instance, or illustration." Any aspect described herein as "exemplary" is not necessarily to be construed as preferred or advantageous over other aspects.

[0135] As used herein, phrases referring to "at least one of" a list of items refer to any combination of those items, including single members. By way of example, "at least one of a, b, or c" is intended to cover a, b, c, ab, ac, bc, and abc, as well as any combination with multiples of the same element (e.g., aa, aaa, aab, aac, abb, acc, bb, bbb, bbc, cc, and ccc, or any other permutation of a, b, and c).

[0136] As used herein, the term "determining" encompasses a wide variety of actions. For example, "determining" may include calculating, computing, processing, deriving, investigating, looking up (e.g., looking up in a table, database, or another data structure), ascertaining, and the like. "Determining" may also include receiving (e.g., receiving information), accessing (e.g., accessing data in a memory), and the like. "Determining" may also include resolving, selecting, choosing, establishing, and the like.

[0137] The methods disclosed herein include one or more steps or actions for achieving the method. Method steps and / or actions may be interchanged with one another without departing from the scope of the claims. In other words, unless a specific order of steps or actions is specified, the order and / or use of specific steps and / or actions may be modified without departing from the scope of the claims. Furthermore, various operations of the methods described above may be performed by any suitable means capable of performing the corresponding functions. The means may include various hardware and / or software components and / or modules, including, but not limited to, circuits, application specific integrated circuits (ASICs), or processors. Generally, where operations illustrated in a figure exist, those operations may have corresponding counterparts of means+function components with similar numbering.

[0138] The following claims are not intended to be limited to the embodiments set forth herein, but are to be accorded the full scope consistent with the claim language. Within the claims, reference to an element in the singular is not intended to mean "one and only one" unless specifically so recited, but rather, "one or more." Unless specifically recited otherwise, the term "some" refers to one or more. No claim element is to be construed under the provisions of 35 USC §112(f) unless the element is expressly recited using the phrase "means for," or, in the case of a method claim, the element is recited using the phrase "step for." All structural and functional equivalents to the elements of the various aspects described throughout this disclosure that are known or later become known to those of skill in the art are expressly incorporated herein by reference and are intended to be encompassed by the claims. Additionally, nothing disclosed herein is intended to be made available to the public, regardless of whether such disclosure is expressly recited in a claim.

Claims

1. A method, the method comprising: determining a plurality of characteristics of the wound related to the patient based on an image of the wound, wherein determining includes detecting the plurality of characteristics based on analyzing the image using a first machine learning (ML) model trained to detect wound characteristics from captured images; identifying patient medical data comprising a plurality of characteristics relating to a medical history of the patient; predicting a first treatment plan for the patient based on providing the plurality of characteristics of the wound and the patient medical data to a second ML model, the second ML model being trained to predict the first treatment plan using prior wound treatment outcome data comprising a plurality of prior wound treatment outcomes associated with a plurality of prior patients, the first treatment plan being configured to be used to treat the wound for the patient; A method comprising:

2. Using the first ML model, determining a second plurality of characteristics of the wound for the patient based on analyzing second images of the wound captured during treatment of the wound associated with the predicted first treatment plan; and predicting a second treatment plan for the patient based on providing the second plurality of characteristics of the wound to the second ML model; and The method of claim 1 further comprising:

3. The method described in claim 2, wherein the second plurality of characteristics of the wound are further used to modify the second ML model through further training based on the second plurality of characteristics and the first treatment plan.

4. The method of claim 1, further comprising identifying a preventative treatment task for the wound based on at least one of the plurality of characteristics of the wound or the treatment plan.

5. The method of claim 1, wherein detecting the multiple characteristics of the wound further includes determining at least one of the depth, color, or size of the wound based on the image of the wound.

6. The method of claim 1, wherein the prior wound treatment outcome data comprises data reflecting wound characteristics, treatment, and resolution for each of a plurality of past wounds associated with the plurality of prior patients.

7. The method of claim 6, further comprising identifying additional data captured by a sensor during treatment or assessment of the wound; The method of claim 1 , wherein predicting the first treatment plan is further based on providing the identified additional data to the second ML model.

8. The method of claim 1, wherein the treatment plan comprises one or more recommended treatment tasks for the wound, the one or more recommended treatment tasks comprising at least one of prescription medications, patient actions, or treatment provider actions for treating the wound.

9. The method of claim 8, further comprising modifying treatment of the wound based on the one or more recommended treatment tasks.

10. The method of claim 1, further comprising identifying a preventative treatment task for the wound based on at least one of the plurality of characteristics of the wound or the treatment plan.

11. An apparatus, comprising: Memory and a hardware processor communicatively coupled to the memory, the hardware processor comprising: determining a plurality of characteristics of the wound related to the patient based on an image of the wound, wherein determining includes detecting the plurality of characteristics based on analyzing the image using a first machine learning (ML) model trained to detect wound characteristics from captured images; identifying patient medical data comprising a plurality of characteristics relating to a medical history of the patient; predicting a first treatment plan for the patient based on providing the plurality of characteristics of the wound and the patient medical data to a second ML model, the second ML model being trained to predict the first treatment plan using prior wound treatment outcome data comprising a plurality of prior wound treatment outcomes associated with a plurality of prior patients, the first treatment plan being configured to be used to treat the wound for the patient; a hardware processor configured to perform operations including: An apparatus comprising:

12. The operation is determining a second plurality of characteristics of the wound for the patient based on analyzing second images of the wound captured during treatment of the wound associated with the predicted first treatment plan using the first ML model; predicting a second treatment plan for the patient based on providing the second plurality of characteristics of the wound to the second ML model; and The apparatus of claim 11 further comprising:

13. The apparatus of claim 12, wherein the second plurality of characteristics of the wound are further used to modify the second ML model through further training based on the second plurality of characteristics and the first treatment plan.

14. The operation is identifying a preventative treatment task for the wound based on at least one of the plurality of characteristics of the wound or the treatment plan; transmitting an electronic alert related to said treatment task; The apparatus of claim 11 further comprising:

15. The method of claim 14, wherein identifying the preventive treatment task is identified using the second ML model based on the at least one of the plurality of characteristics of the wound; The operation is electronically transmitting the alert to a treatment provider for the patient using a communications network prior to completing the predicting the treatment plan for the patient. The apparatus of claim 14 further comprising:

16. The device of claim 11, wherein the prior wound treatment outcome data comprises data reflecting wound characteristics, treatment, and resolution for each of a plurality of past wounds associated with the plurality of prior patients.

17. A non-transitory computer-readable medium comprising instructions that, when executed by a processor, determining a plurality of characteristics of the wound related to the patient based on an image of the wound, wherein determining includes detecting the plurality of characteristics based on analyzing the image using a first machine learning (ML) model trained to detect wound characteristics from captured images; identifying patient medical data comprising a plurality of characteristics relating to a medical history of the patient; predicting a first treatment plan for the patient based on providing the plurality of characteristics of the wound and the patient medical data to a second ML model, the second ML model being trained to predict the first treatment plan using prior wound treatment outcome data comprising a plurality of prior wound treatment outcomes associated with a plurality of prior patients, the first treatment plan being configured to be used to treat the wound for the patient; 10. A non-transitory computer-readable medium that causes the processor to perform operations including:

18. The operation is determining a second plurality of characteristics of the wound for the patient based on analyzing second images of the wound captured during treatment of the wound associated with the predicted first treatment plan using the first ML model; predicting a second treatment plan for the patient based on providing the second plurality of characteristics of the wound to the second ML model; and further comprising 18. The non-transitory computer-readable medium of claim 17, wherein the second plurality of characteristics of the wound is further used to modify the second ML model through further training based on the second plurality of characteristics and the first treatment plan.

19. The operation is using the second ML model to identify preventative treatment tasks for the wound based on the plurality of characteristics of the wound; electronically transmitting, using a communications network, an electronic alert to a treatment provider for the patient prior to completing the predicting of the treatment plan for the patient; 20. The non-transitory computer-readable medium of claim 17, further comprising:

20. The non-transitory computer-readable medium of claim 17, wherein the prior wound treatment outcome data comprises data reflecting wound characteristics, treatment, and resolution for each of a plurality of past wounds associated with the plurality of prior patients.

Images