Skin rejection prediction model

Abstract

The present disclosure describes techniques for generating a skin rejection prediction model. The skin rejection prediction model may predict the likelihood of skin rejection for a skin graft site. Tire skin rejection prediction model may be generated by using a plurality of feature selection models to select a model subset from training data. Candidate skin rejection prediction models may be trained using the model subset. The skin rejection prediction model may be selected from the candidate prediction models based on determining a performance metric associated with each candidate prediction using a test subset of tire training data.

Description

SKIN REJECTION PREDICTION MODELSTATEMENT REGARDING FEDERALLY SPONSORED RESEARCH OR DEVELOPMENT

[0001] This invention was made with government support under W81XWH-13-2-0053 awarded by the United States Army Medical Research and Development Command and under SUR-90-2747 and HU0001-17-2-0003 awarded by the Uniformed Services University of the Health Sciences. The government has certain rights in the invention.CROSS REFERENCE TO RELATED APPLICATIONS

[0002] This application claims the benefit of U.S. Provisional Patent Application 63 / 519,168, filed on August 11, 2023, which is hereby incorporated by reference in its entirety.FIELD OF THE DISCLOSURE

[0003] Described herein are methods, systems, and computational environments for predicting transplant rejection associated with skin-containing reconstructive transplantation using a skin rejection prediction model. Also described are methods, systems, and computational environments for generating the skin rejection prediction model.BACKGROUND

[0004] Skin-containing reconstructive transplantation such as Vascularized Composite Allotransplantation (VC A) has become a viable reconstructive alternative for devastating combat and civilian injuries where conventional treatment is not possible. However, routine clinical use of skin-containing reconstructive transplantation has been limited by the risk related to the use of long-term high-dose immunosuppression required to avoid transplant rejection. Therefore, a critical component in a successfill transplantation is careful maintenance of the immunosuppression to avoid overimmunosuppression or insufficient immunosuppression. Overimmunosuppression may result in chronic infections and accumulation of dangerous side effects that may be life threatening. Insufficient immunosuppression may lead to acute or chronic rejection episodes, loss of skin graft function and / or loss of the skin graft itself. Currently, monitoring immunosuppression depends on punch biopsies. However, punch biopsies can cause trauma and mechanical stress that may lead to an alloimmune response and even trigger rejection.SUMMARY OF THE DISCLOSURE

[0005] This Summary is provided to introduce a selection of concepts in a simplified form that arc further described below in the Detailed Description. This Summary' is not intended to identify all key features or essential features of the claimed subject matter, nor is it intended to be used alone as an aid in detennining the scope of the claimed subject matter.

[0006] Techniques for determining a prediction associated with skin rejection at a skin graft site of a subject and / or a treatment recommendation using a skin rejection prediction model and for generating the skin rejection prediction model are discussed herein.

[0007] In embodiments, there are provided a method of generating a skin rejection prediction model for a subject comprising: receiving first raw sensor data from a plurality of first imaging devices; receiving second raw sensor data from a plurality of second imaging devices; determining, from the first raw sensor data, first imaging data, the first imaging data comprising intensities associated with a plurality of color channels associated with the first imaging devices; determining, from the second raw sensor data, second imaging data, the second imaging data comprising intensities associated with the second imaging devices; generating a data structure storing training data, the training data comprising the first imaging data, the second imaging data, and clinical outcomes associated with the first imaging data and the second imaging data; executing a plurality of feature selection algorithms on the training data to generate a model subset, the model subset comprising a subset of the training data less than a total amount of the training data; training a plurality of candidate prediction models using the model subset; detennining a performance metric for each trained candidate prediction model of the plurality of trained candidate prediction models; and selecting, based on the performance metric, a candidate model from the plurality of trained candidate prediction models as the skin rejection prediction model, wherein the skin rejection prediction model is configured to determine a likelihood of skin rejection at a skin graft site of the subject.

[0008] In embodiments, there are provided a method of predicting skin rejection at a skin graft site of a subject comprising: receiving a first value of a first clinical parameter associated with the skin graft site and a second value of a second clinical parameter associated with the skin graft site; executing a skin rejection prediction model using the first value and the second value to generate a prediction associated with skin rejection at the skin graft site, wherein the skin rejection prediction model is generated by performing operations comprising: receiving first raw sensor data from a plurality of first imaging devices; receiving second raw sensor data from a plurality of second imaging devices; determining, from the first raw sensor data, first imaging data, the first imaging data comprising intensities associated with a plurality of color channels associated with the firstimaging devices; determining, from the second raw sensor data, second imaging data, the second imaging data comprising intensities associated with the second imaging devices; generating a data structure storing a training data, the training data comprising the first imaging data, the second imaging data, and clinical outcomes associated with the first imaging data and the second imaging data; executing a plurality of feature selection algorithms on tire training data to generate a model subset, the model subset comprising a subset of the training data less than a total amount of the training data; training a plurality of candidate prediction models using the model subset; determining a performance metric for each trained candidate prediction model of the plurality of trained candidate prediction models; and selecting, based on the performance metric, a trained candidate prediction model from the plurality of trained candidate prediction models as tire skin rejection prediction model; and outputting, by the skin rejection prediction model based on the first value and the second value, the prediction associated with skin rejection at the skin graft site.

[0009] In embodiments, there are provided a system for generating a skin rejection prediction model for a subject comprising: one or more processors; and one or more non -transitory computer- readable media storing computer-executable instructions that, when executed, cause the one or more processors to perform operations comprising: receiving first raw sensor data from a plurality of first imaging devices; receiving second raw sensor data from a plurality of second imaging devices; determining, from the first raw sensor data, first imaging data, the first imaging data comprising intensities associated with a plurality of color channels associated with the first imaging devices; determining, from the second raw sensor data, second imaging data, the second imaging data comprising intensities associated with the second imaging devices; generating a data structure storing training data, tire training data comprising the first imaging data, the second imaging data, and clinical outcomes associated with the first imaging data and the second imaging data; executing a plurality of feature selection algorithms on the training data to generate a model subset, the model subset comprising a subset of the training data less than a total amount of the training data; training a plurality of candidate prediction models using the model subset; determining a performance metric for each trained candidate prediction model of the plurality’ of trained candidate prediction models; and selecting, based on the perfonnance metric, a trained candidate prediction model from the plurality of trained candidate prediction models as the skin rejection prediction model, wherein the skin rejection prediction model is configured to determine a likelihood of skin rejection at a skin graft site of the subject.

[0010] In embodiments, there are provided a system for predicting skin rejection at a skin graft site of a subject comprising: one or more processors; an output component; and one or more computer-readable media storing computer-executable instructions that, when executed, cause theone or more processors to perform operations comprising: receiving a first value of a first clinical parameter associated with the skin graft site and a second value of a second clinical parameter associated with the skin graft site; executing a skin rejection prediction model using the first value and the second value to generate a prediction associated with skin rejection at the skin graft site, wherein the skin rejection prediction model is generated by performing operations comprising: receiving first raw sensor data from a plurality of first imaging devices; receiving second raw sensor data from a plurality of second imaging devices; determining, from the first raw sensor data, first imaging data, the first imaging data comprising intensities associated with a plurality of color channels associated with the first imaging device; determining, from the second raw sensor data, second imaging data, the second imaging data comprising intensities associated with the second imaging device; generating a data structure storing a training data, the training data comprising the first imaging data, the second imaging data, and clinical outcomes associated with the first imaging data and the second imaging data; executing a plurality of feature selection algorithms on the training data to generate a model subset, the model subset comprising a subset of the training data less than a total amount of the training data; training a plurality of candidate prediction models using the model subset; determining a perfonnance metric for each trained candidate prediction model of the plurality of trained candidate prediction models; and selecting, based on the perfonnance metric, a candidate model from the plurality of candidate models as the skin rejection prediction model; and outputting, by the skin rejection prediction model and the output component and based on the first value and the second value, the prediction associated with skin rejection at the skin graft site.

[0011] In embodiments, machine learning algorithms associated with candidate prediction models include random forest, extreme gradient boosting, support vector machine, naive Bayesian, neural network, and least absolute shrinkage and / or selection operator (e.g., Glmnet or LASSO).

[0012] In embodiments, the feature selection machine learning models include bootstrap resampling, random forest, and / or recursive feature elimination.BRIEF DESCRIPTION OF THE DRAWINGS

[0013] The detailed description is described with reference to the accompanying figures. In the figures, the left-most digit(s) of a reference number identifies the figure in which the reference number first appears. The use of the same reference numbers in different figures indicates similar or identical components or features. The figures are merely exemplary to illustrate certain features that may be used singularly or in combination with other features, and the present disclosure should not be limited to the embodiments shown.

[0014] FIG. 1 illustrates a block diagram of an example system for generating a skin rejection prediction model.

[0015] FIG. 2 illustrates a flow diagram of an example process of generating tire skin rejection prediction model.

[0016] FIG. 3 illustrates a flow diagram of an example process for performing the feature selection step while generating the skin rejection prediction model.

[0017] FIG. 4 illustrates a flow diagram of an example process of selecting the skin rejection prediction model from candidate models.

[0018] FIG. 5 illustrates a block diagram of an example system for using the skin rejection prediction model.

[0019] FIG. 6 illustrates a flow diagram of an example of using the skin rejection prediction model.DETAILED DESCRIPTION

[0020] The following detailed description is presented to enable any person skilled in the art to make and use the subject of the application. For purposes of explanation, specific nomenclature is set forth to provide a thorough understanding of the present disclosure. However, it will be apparent to one skilled in the art that these specific details are not required to practice the subject of the application. Descriptions of specific applications are provided only as representative examples. The present application is not intended to be limited to the embodiments shown, but is to be accorded the widest possible scope consistent with the principles and features disclosed herein.

[0021] The present disclosure provides methods for predicting whether a skin transplant may lead to rejection of the transplanted skin (also referred to as skin graft).

[0022] Technical and scientific terms used herein have the meanings commonly understood by one of ordinary skill in the art to which the present disclosure pertains, unless otherwise defined.

[0023] As used herein, the singular forms “a,” “an,"’ and “the'’ designate both the singular and the plural, unless expressly stated to designate the singular only.

[0024] As used herein, the term “skin graft” refers to skin tissue(s) that are surgically transplanted to treat wound(s), bum(s), area(s) of extensive skin loss due to infection, and surge ry or surgeries that require skin grafts for healing to occur (e g., after removing skin cancer tissues). The skin graft may be skin tissue from another site on the patient themselves (autologous / autograft), skin tissue from a donor that is of the same species as the patient, but not genetically identical(allogeneic / allograft), skin tissue from a donor that is genetically identical with the patient (isogeneic / isograft or syngraft), or skin tissue from a donor of a different species (xenogenic / xenograft or heterograft).

[0025] As used herein, the term “skin graft site" refers to a location on a subject or an individual's body where the skin graft is installed onto.

[0026] As used herein, the term “overimmunosuppression” refers to when a concentration of one or more of the immunosuppressants are determined to exceed a first concentration threshold.

[0027] As used herein, the term “insufficient immunosuppression” refers to when a concentration of the one or more of the immunosuppressants are determined to be below a second concentration threshold, the second concentration threshold being a different concentration threshold below the first concentration threshold.

[0028] As used herein, the terms “marker” and “biomarkers” are used interchangeably to refer to a measurable substance from a biological sample. For example, these can comprise one or more protein data markers, one or more nucleic acid data markers, one or more metabolite data markers, or a combination thereof.

[0029] As used herein, the term “pro-inflammatory biomarker” (also referred to as inflammatory biomarker”) refers to a biomarker whose presence is indicative of inflammation.

[0030] As used herein, tire term “anti-inflammatory biomarker” refers to a biomarker whose presence is indicative of lessening or lack of inflammation.

[0031] Examples of pro-inflammatory biomarkers may include, but are not limited to, granulocyte-monocyte colony stimulating factor (GMCSF), interleukin- 1 alpha (IL-1A), interleukin-1 beta (IL-1B), interleukin-1 receptor antagonist (IL-IRA), interleukin-2 (IL-2), interleukin-4 (IL-4), interleukin-6 (IL-6), interleukin- 10 (IL- 10), interleukin- 18 (IL- 18), and tumor necrosis factor alpha (TNFA). Hie enumerated pro-inflammatory biomarkers are cytokines.

[0032] Examples of anti-inflammatory biomarkers may include, but are not limited to, interleukin-8 (IL-8), interleukin- 12 (IL-12), and interferon gamma (IFNy). The enumerated antiinflammatory biomarkers are also cytokines.

[0033] As used herein, the term “3 charge-coupled device” (also referred to as 3CCD) refers to an imaging device or system that includes separate image sensors for each color channel. For example, a first image sensor may be configured to capture an intensity of red light, a second image sensor may be configured to capture an intensity of green light, and a third image sensor may be configured to capture an intensity of blue light.

[0034] As used herein, the term “near infrared imaging” (also referred to as near IR imaging) refers to using an imaging device or system to capture image(s) between a wavelength range of 780 nm to 2500 mu.

[0035] As used herein, the term “immunosuppressant” (also referred to as immunosuppressive) refers to a drug that reduces, inhibits, or prevents activities of a body’s immune system.

[0036] Examples of immunosuppressants may include, but are not limited to, prednisone, cyclosporine, tacrolimus, sirolimus, everolimus, and my cophenolate mofetil.

[0037] As used herein, the term “training data” refers to a data set used for generating the skin rejection prediction model and / or verifying an accuracy of the generated skin rejection prediction model.

[0038] As used herein, the term “training subset” refers to a first portion of the training data that is less than a whole of the training data, w here the training subset is used to generate the skin rejection prediction model.

[0039] As used herein, the term test subset refers to a second portion of the training data that is not used as the training subset, where the test subset is used for verifying the generated skin rejection prediction model

[0040] As used herein, the term “skin rejection” refers to inflammation and / or necrosis of the skin graft.

[0041] FIG. 1 illustrates a block diagram of an example system 100 for generating a skin rejection prediction model. The example system 100 includes body fluid collection device(s) 102, an immunosuppressant detection device 104, first computing device(s) 106, first camera(s) 122, second computing device(s) 124, and second camera(s) 132. The first computing device(s) 106 includes an input component 108, first processor(s) 110, first memory 112, a training data component 114. a clinical outcome component 116, a prediction model generating component 118, and an output component 120. The second computing device(s) 124 includes second processor(s) 126, second memory' 128, and a prediction model component 130. In embodiments, a skin rejection prediction model may be generated at the first computing dcvicc(s) 106 and transferred to the second computing device(s) 124. Alternatively, the skin rejection prediction model may be generated at the second computing device 124.

[0042] The body fluid collection device(s) 102 may be one or more devices that are positioned at or proximate to a plurality of subjects to collect body fluid samples from the subjects, the subjects being different than an individual or a subject that tire generated skin rejection prediction model is used for. In embodiments, the body fluid samples may include blood or other types of discharge.In embodiments, each body fluid collection device of the device(s) 102 may include a needle and a fluid storage container physically coupled to the needle.

[0043] In embodiments, the body fluid samples collected by the body fluid collection device(s) 102 may be transferred to an immunosuppressant detection device 104. The immunosuppressant detection device 104 may detect, identify, and quantify one or more immunosuppressant(s) from the body fluid. The immunosuppressant(s) may be detected, identified, and quantified (in other words, determined) by performing an immunoassay(s) and / or performing mass spectrometry (e.g., liquid-chromatography mass spectrometry) on the collected body fluid sample.

[0044] In embodiments, the detecting, identifying, and quantifying the immunosuppressant(s) may include detecting a presence of one or more immunosuppressants from the body fluid sample, identifying the one or more immunosuppressants, and / or determining a concentration of each detected and identified immunosuppressant.

[0045] In embodiments, as each immunosuppressant is detected, identified, and quantified by the immunosuppressant detection device 104, immunosuppressant data associated with the detected, identified, and quantified immunosuppressant (e.g., immunosuppressant name and concentration) may be transmitted or transferred to the first computing device(s) 106 and stored at the training data component 114. In embodiments, the immunosuppressant detection device may include a communication component that can transmit the immunosuppressant parameter(s) to the first computing device(s) 106 using a wired connection or a wireless connection. The first computing device(s) 106 may receive the immunosuppressant data using its communication component. Alternatively, the immunosuppressant data may be transmitted or transferred to the first computing device(s) 106 after all immunosuppressant parameter(s) are determined. In embodiments, the immunosuppressant data received by the first computing device(s) 106 from the immunosuppressant detection device 104 may be stored in tire training data component 114 as a feature to use to generate the skin rejection prediction model. In embodiments, the first computing device(s) may further include a list of immunosuppressants that are relevant to skin transplant such that the presence and / or absence of one or more immunosuppressants from the list may be included as a feature for the training data.

[0046] Alternatively, the immunosuppressant data may be determined by the processor(s) 110.

[0047] In embodiments, the first computing device(s) 106 is where the skin transplant prediction module is being generated. As illustrated, the first computing device(s) 106 is configured to generate the skin rejection prediction model, and transmits or transfers the generated skin rejection prediction model to the second computing device(s) 124 where the skin rejectionprediction model is used to generate a prediction metric associated with skin rejection. Alternatively, the first computing device(s) 106 may generate and utilize the skin rejection prediction model. Tire skin rejection prediction model may be generated by a single first computing device 106 or multiple first computing devices 106 (e.g., by a server cluster).

[0048] In addition to the immunosuppressant data, the first computing device(s) 106 may receive additional data that may be stored at the training data component 114 of the memory 112. In embodiments, the first computing device(s) 106 may receive first imaging data from the first camera(s) 122 and second imaging data from the second camera(s) 132.

[0049] In embodiments, each of the first camera(s) 122 may be a 3 CCD imaging device configured to emit a red light, a green light, and a blue light towards the skin graft site of a subject of the plurality of subjects and measure intensities of red, a blue, and a green lights scattered by the skin graft and captured by the image sensors of the first camera(s) 122. The intensities of each of the red, green, and blue lights captured by the first camera(s) 122 may be indicative of oxygenation at the skin graft site and the presence of pro-inflammatory biomarkers or anti-inflammatory biomarkers. In embodiments, the first camera(s) 122 may detennine as the first imaging data, features that include a total intensity based on all of the color channels (also referred to as a total 3CCD). an intensity of the red channel (also referred to as R2), an intensity of the blue channel (also referred to as B2), an intensity of the green channel (also referred to as G2), a difference between tire intensity of the red channel and the blue channel (also referred to as R2-B2), and a difference between an absorption between tire red channel and the blue channel (also referred to as R-B). First imaging data that are indicative of higher or more pro-inflammatory biomarker(s) than anti-inflammatory biomarker(s) or elevated pro-inflammatory biomarker(s) may be indicative of skin rejection.

[0050] Alternatively, the first camera(s) 122 may transmit or transfer raw color sensor data captured by the image sensors of the first camera(s) 122 to the computing device(s) 106. The processor(s) 110 may determine all features of the first image data from the raw color sensor data.

[0051] In embodiments, each of the second camera(s) 132 may be a near IR imaging device. The second camera may be configured to measure blood perfusion at the skin graft site. In embodiments, the second camera(s) 132 may determine as the second imaging data, features that include a mean IR intensity, a standard deviation of the IR intensity, a center IR intensity, a maximum IR intensity, and a minimum IR intensity. In embodiments, lower measured blood perfusion based on tire second imaging data may be indicative of skin rejection.

[0052] Alternatively, the second camera(s) 132 may transmit or transfer raw IR sensor data captured by the image sensors of the second camera(s) 132 to the computing device(s) 106. The proccssor(s) 110 may determine all features of the second image data from the raw IR sensor data.

[0053] In embodiments, the training data component may further store as training data a first principal component, a second principal component, and a third principal component. Each principal component is the result of a performing principal component analysis (e.g., a dimension reduction method). The first principal component may be a linear combination of the features from the first imaging data, the second imaging data, and the immunosuppressant data that are associated with a first weight. The second principal component may be a linear combination of the features from the first imaging data, tire second imaging data, and the immunosuppressant data that are associated with a second weight that is different from the first weight. The third principal component may be a linear combination of the features from the first imaging data, the second imaging data, and the immunosuppressant data that are associated with a third weight that is different from the first weight and the second weight.

[0054] In embodiments, the clinical outcome component is configured to store clinical outcome(s) associated with the training data. In embodiments, the stored clinical outcome(s) may indicate whether there is skin rejection. In embodiments, the clinical outcome(s) may further or alternatively include a score associated with skin rejection. In embodiments, the score may be between 0 and 4. A score of 0 may indicate no or rare occurrence of inflammation at the skin graft site, and therefore may be indicative of no skin rejection (also referred to as non-rejection). A score of 1 may indicate mild occurrence of inflammation at the skin graft site, and therefore may also be indicative of no skin rejection. A score of 2 may be indicative of severe inflammation at the skin graft site, and therefore may be indicative of skin rejection. A score of 3 may be indicative of necrosis at the skin graft site, and therefore may also be indicative of skin rejection. In embodiments, the scores may correspond with a grade from the Banff 2007 Working Classification.

[0055] In embodiments, the input component 508 may include a mouse, a keyboard, a touch screen, and / or the like. In embodiments, a user may utilize the input component to input clinical outcome (s).

[0056] In embodiments, training data may be associated with the clinical outcome(s). In embodiments, the association may include clustering the features from the training data (e.g., the mean near IR intensity, the standard deviation of the near IR intensity, the center near IR intensity, the maximum near IR intensity, the minimum near IR intensity, the R-B, the total 3 CCD, the R2, the G2, the B2, the R2-B2, the first principal component, tire second principal component, the third principal component, and the presence or absence of one or more immunosuppressants) with theclinical outcome(s). For example, each feature or a combination of the features may be associated with one of the four scores. Based on the score, each feature or the combination of the features may be clustered as having a likelihood of skin rejection or having a likelihood of skin non-rejection.

[0057] In embodiments, tire training data may be stored at the training data component 114 in a first data structure (e.g., a database) and the clinical outcome(s) and the clustering may be stored at the clinical outcome component 116 in a second data structure. Alternatively, the training data and the clustering may be stored in the memory 112 in a single data structure. In embodiments, the processor(s) 110 may be used to generate the skin rejection prediction model at the prediction model generating component using the training data and the clustering. Generating the skin rejection prediction model is described in additional detail at FIG. 2-4, as well as throughout this disclosure. In embodiments, the generated skin rejection prediction model may be stored at the prediction model generating component 118 or at another location within the memory 112. The skin rejection prediction model being stored within the memory 112 may be transferred or transmitted to the prediction model component 130 within the memory 128 of the second computing device(s) 124. The processor(s) 126 may use the skin rejection prediction model to determine a skin rejection prediction metric at a skin graft site of the subject. Using the prediction model to determine the skin rejection prediction metric is discussed in additional detail in FIGS. 5 and 6, as well as throughout this disclosure.

[0058] FIG. 2 illustrates a flow diagram of an example process 200 of generating a skin rejection prediction model. In embodiments, some or all of process 200 can be performed by one or more components described in association with FIG. 1, as described herein. Additionally, some portions of process 200 can be omitted, replaced, and / or reordered. For example, the example process 200 may be performed by the processor(s) 110 of the first computing device(s) 106, and the generated skin rejection prediction model may correspond to the generated wound prediction model detailed in association with FIG. 1.

[0059] At operation 202, the process can include receiving training data. In embodiments, tire training data may correspond to data stored within the training data component 114 and data stored within the clinical outcome component 116 described in association with FIG. 1, as well as throughout this disclosure. In embodiments, the training data may include the first imaging data, the second imaging data, the immunosuppressant data, and the clustering of the first imaging data, the second imaging data, and the immunosuppressant data with the clinical outcomes described in association with FIG. 1, as well as throughout this disclosure. Examples of the features in the training data may include, but are not limited to. the mean near IR intensity, the standard deviation of the near IR intensity, the center near IR intensity, the maximum near IR intensity, the minimumnear IR intensity, the R-B, the total 3CCD, the R2, the G2, the B2, the R2-B2, the first principal component, the second principal component, the third principal component, and the presence or absence of one or more immunosuppressants.

[0060] At operation 204, the process can include determining a training subset from the training data. In embodiments, the training subset may be used for generating the skin rejection prediction model. In embodiments, the operation 204 may include selecting 80% of the training data as the training subset. Selecting the training data can include selecting features from the training data that are clustered with skin rejection and features from the training data that are clustered with non-rcjcction.

[0061] At operation 206, the process can include detennining a test subset from the training data. In embodiments, the test subset may be used to verify the skin rejection prediction model. In embodiments, the test subset may be the training data that are not part of the training subset.

[0062] At operation 208, the process can include performing feature selection on the training subset to determine a model subset. Additional details with respect to the operation 210 are described in association with FIG. 3, as well as throughout this disclosure.

[0063] At operation 210, the process can include training candidate prediction models. In embodiments, the operation 212 may include inputting the model subset into each candidate prediction model to train each candidate prediction model. Additional details with respect to the operation 212 are described in association with FIG. 4, as well as throughout this disclosure.

[0064] At operation 212, the process can include performing subgroup analysis to determine whether each candidate prediction model performs well. The operation 212 can include determining, during the subgroup analysis, whether the result of the subgroup analysis for each candidate prediction model meets or exceeds a threshold analysis metric. In embodiments, only the candidate prediction modcl(s) that meets or exceeds the threshold analysis metric may be considered for selection as the skin rejection prediction model.

[0065] At operation 214. the process can include detennining first performance metric(s) for each candidate prediction model and determining, based on the first performance metric(s), whether any of the candidate prediction models requires calibration. In embodiments, the first performance metric(s) may be determined by inputting the test subset into the candidate prediction models and / or performing subset analysis of the candidate prediction models using the subgroups. In embodiments, the first performance metric(s) may include a first calibration metric that is indicative of whether any of the candidate prediction models require calibration. Additional details with respect to the operation 214 are described in association with FIG. 4, as well as throughout this disclosure.

[0066] At operation 216, the process can include determining the second performance metric(s) for each candidate prediction model and determining, based on the second performance mctric(s), whether any of the candidate prediction models requires calibration. In embodiments, the second performance metric(s) may be determined by inputting the external test data into the candidate prediction models. In embodiments, the second performance metric(s) may include a second calibration metric that is indicative of whether any of the candidate prediction models require calibration. Additional details with respect to tire operation 216 are described in association with FIG. 4, as well as throughout this disclosure.

[0067] At operation 218, the process can include performing tire calibration. In embodiments, the operation 218 may further include selecting, based on the first and / or the second perfonnance metric(s), the subgroup analysis, and / or after calibration, one of the candidate prediction models as the skin rejection prediction model. Additional details with respect to the operation 218 are described in association with FIG. 4, as well as throughout this disclosure.

[0068] FIG. 3 illustrates a flow diagram of an example process 300 of performing feature selection during the generation of the skin rejection prediction model. In embodiments, some or all of process 300 can be performed by one or more components described in association with FIG. 1, as described herein. Additionally, some portions of process 300 can be omitted, replaced, and / or reordered. For example, the example process 300 may be performed by the processor(s) 110 of the first computing device(s) 106, and the generated skin rejection prediction model may correspond to the generated wound prediction model detailed in association with FIG. 1. In embodiments, the example process 300 may correspond to the operation 210 of FIG. 2.

[0069] At operation 302, the process can include receiving training data. In embodiments, the training data may include some or all of the data stored in the training data component 114. The training data may include features such as the mean near IR intensity, the standard deviation of the near IR intensity, the center near IR intensity, the maximum near IR intensity, the minimum near IR intensity, tire R-B, the total 3 CCD, the R2, the G2, the B2, the R2-B2, the first principal component, the second principal component, the third principal component, and the presence or absence of one or more immunosuppressants.

[0070] At operation 304, the process can include performing a data quality check on the training data. In embodiments, the operation 304 may include determining a completeness of the training data. Determining the completeness may include determining whether the training data contains missing value(s) for one or more features, and if so, the operation 304 may include filling in the missing value(s) to ensure completeness of the training data. In embodiments, the total number of values or entries of the training data after the quality check may be at least 1000.

[0071] At operation 306, the process can include selecting a training subset. In embodiments, the operation 306 may include randomly selecting a portion of the training data as the training subset. In embodiments, the operation 306 may include selecting at an 8:2 ratio, the training subset and a test subset. The test subset may correspond to the test subset. In embodiments, the operation 304 may further or alternatively include selecting at least one value or entry associated with each feature of the training data. In embodiments, the operation 306 may include randomly selecting one or more values or entries from each feature as a part of the training subset until a threshold ratio (e.g., 8:2 ratio) has been met. In embodiments, the operation 306 may, additionally or alternatively, include ensuring values that arc associated with both skin rejection and skin non-rcjcction arc selected for the training subset.

[0072] At operation 308. the process can include generating a resampled training subset (also referred to as a resampled subset). In embodiments, the operation 308 may include randomly selecting features from the training subset with replacement, that is when a value or entry is selected from the training subset to include in the resampled subset, that value or entry available to be selected again for the resampled subset such that the resample subset may include one or more duplicate entries and therefore result in excluding one or more entries from the training subset. Tire operation 308 may further include randomly selecting values or entries from the training subset with replacement until the number of values or entries within the resampled subset equals the total number of values or entries in the training subset. Alternatively, the operation 308 may be performed until the number of values or entries in the resampled subset reaches a threshold number of entries, where the threshold number of entries is less than tire total number of values or entries in the training subset. For example, the operation 308 may be performed until the resampled subset reaches 70% of the total number of values or entries of the training subset.

[0073] At operation 310, the process can include performing a random forest algorithm on the resampled subset. In embodiments, performing the random forest algorithm includes generating a plurality of decision trees, each decision tree of the plurality of decision trees includes randomly generated nodes, where each decision tree may include a different number of randomly generated nodes. The number of decision trees may be based on the size of the resampled subset (e.g., the total number of values or entries of the resampled subset), such that the greater the size of the resampled subset, the higher the number of decision trees. The operation 310 may further include generating a minimum threshold number of decision trees. For example, for the operation 310, the number of decision trees to be generated may be 20% of the total number of values or entries of the resampled subset, and the minimum threshold number of decision trees may be 50 decisiontrees. Therefore, if 20% of the total number of values or entries is below the 50 decision trees minimum, the operation 310 would generate 50 decision trees.

[0074] At operation 312, the process can include ranking features based on the result of the random forest algorithm performed at the operation 310. In the operation 310. each tree has a vote to support the importance of at least one feature. A majority rule is applied to the vote. In embodiments, during the operation 310, the features in the resampled subset are evaluated for their contributions to an accuracy in discriminating the outcomes (e.g., in the accuracy in discriminating skin rejection or skin non -rejection). Evaluating a feature’s contribution to the accuracy in discriminating tire outcome includes, for each decision tree, determining tire decision tree’s vote to support the importance of the feature, the importance is determined based on majority rule, and determining a mean decrease accuracy (MDA) associated with the feature, where the MDA measures the decrease in accuracy when the feature is excluded. Each feature is then ranked based on the MDA. For example, a first feature that has a first MDA that indicates a higher decrease in accuracy if the first feature is excluded from the resampled subset would be ranked higher than a second MDA that indicates a lower decrease in accuracy than the first feature if the second feature is excluded from tire resampled subset..

[0075] At operation 314. the process can include eliminating the lowest ranked feature from the resampled subset. In embodiments, the operation 314 may include eliminating the feature with the lowest MDA determined in the operation 312. In embodiments, the operation 314 may be performed after each time the operation 312 is performed. In embodiments, the operation 314, may alternatively be perfonned after a threshold number or iteration of the operations 310 and 312 are perfonned. For example, the operation 314 may be performed after every 10 iterations of the operations 310 and 312 are performed. In such an example, an average ranking may be performed based on the 10 iterations, and the operation 314 may be performed to eliminate the lowest average ranked feature. In embodiments, when performing the operation 314, eliminating the feature from the resampled subset may include excluding all values or entries of the eliminated feature from future iterations of the process 300. In embodiments, the operations 310-316 may be performed as a part of a recursive feature elimination algorithm.

[0076] At operation 316, tire process can include determining whether the number of times that the operation 310 has been perfonned has reached a target number of iterations. If no, then the process may return to the operation 310 where tire operation 310 may be performed on a resampled subset that excludes the eliminated feature(s). If yes, then the process may proceed to operation 318 where a model subset is generated from a resampled subset that includes the remaining features.

[0077] Alternatively, the operation 310 can include determining whether a target number of features remains or whether a target number of eliminated features has been reached. Similarly, if the target number of features remaining or the target number of eliminated features has been reached, then the process may proceed to the operation 318. In embodiments, the features of the model subset may include total 3 CCD, G2. R-B, maximum near IR intensity, minimum near IR intensity, center near IR intensity, and the presence or absence of one or more immunosuppressant! s) .

[0078] FIG. 4 illustrates a flow diagram of an example process 400 of generating a skin rejection prediction model. In embodiments, some or all of process 400 can be performed by one or more components described in association with FIG. 1. as described herein. Additionally, some portions of process 400 can be omitted, replaced, and / or reordered. For example, the example process 400 may be performed by the processor(s) 110 of the first computing device(s) 106, and the generated skin rejection prediction model may correspond to the generated wound prediction model detailed in association with FIG. 1. In embodiments, the example process 400 may correspond to the operations 212-218 of FIG. 1.

[0079] At operation 402. the process can include a receiving model subset. In embodiments, the model subset may correspond to the model subset determined at the operation 318 of FIG. 3.

[0080] At operation 404, the process can include training first candidate prediction model using the model subset. In embodiments, the first candidate prediction model may be configured to predict a likelihood of skin rejection using a random forest algorithm. In embodiments, the operation 404 may include inputting the model subset into the first candidate prediction model to generate a first prediction, determining a first difference between the first prediction and an expected result, the expected result being based on one or more clinical outcomes, and adjusting, based on the first difference, one or more parameters of tire first candidate prediction model to generate updated first candidate prediction model such that subsequent updated first prediction(s) approaches the expected result. In embodiments, training the first candidate prediction model may include iteratively updating the first candidate model using the model subset until the updated first prediction approaches a threshold similarity with the expected result or until a threshold iteration has been reached.

[0081] At operation 406, the process can include a training second candidate prediction model using the model subset. In embodiments, the second candidate prediction model may be configured to predict a likelihood of skin rejection using an extreme gradient boosting algorithm. In embodiments, the operation 406 may include inputting the model subset into the second candidate prediction model to generate a second prediction, determining a second difference between thesecond prediction and the expected result, and adjusting, based on the second difference, one or more parameters of the second candidate prediction model to generate updated second candidate prediction model such that subsequent updated second prediction(s) approaches the expected result. In embodiments, training the second candidate prediction model may include iteratively updating the second candidate model using the model subset until the updated second prediction approaches a threshold similarity with the expected result or until a threshold iteration has been reached.

[0082] At operation 408, the process can include training a third candidate prediction model using the model subset. In embodiments, the third candidate prediction model may be configured to predict a likelihood of skin rejection using a support vector machine algorithm. In embodiments, the operation 404 may include inputting the model subset into the third candidate prediction model to generate a third prediction, determining a third difference between the third prediction and the expected result, and adjusting, based on the third difference, one or more parameters of the third candidate prediction model to generate updated third candidate prediction model such that subsequent updated third prediction(s) approaches the expected result. In embodiments, training the third candidate prediction model may include iteratively updating the third candidate model using the model subset until the updated third prediction approaches a threshold similarity with the expected result or until a threshold iteration has been reached.

[0083] At operation 410, the process can include training n-th candidate prediction model using the model subset. In embodiments, training the n-th candidate prediction model may include training a fourth candidate prediction model, a fifth candidate prediction model, a sixth candidate prediction model, up to an n-th candidate prediction model is trained. For example, the n-th candidate prediction model may be the sixth candidate prediction model, and the operation 410 may include training the fourth candidate prediction model, the fifth candidate prediction model, and the sixth candidate prediction model. In embodiments, the fourth candidate prediction model may be configured to predict a likelihood of skin rejection using a naive Bayesian algorithm, the fifth candidate prediction model may be configured to predict a likelihood of skin rejection using neural networks, and the sixth candidate prediction model may be configured to predict a likelihood of skin rejection using a generalized linear modeling based on the least absolute shrinkage and selection operator algorithm.

[0084] In embodiments, the operation 410 may include inputting the model subset into the third candidate prediction model and up to the n-th candidate prediction model to generate a third prediction and up to an n-th prediction, determining a difference between each of the prediction(s) up to n-th prediction and the expected result, and adjusting one or more parameters of each of the prediction model(s) up to the n-th candidate prediction model to generate updated candidateprediction model(s) such that subsequent updated prediction(s) approaches the expected result. In embodiments, training each of the prediction model(s) up to n-th candidate prediction model may include iteratively updating the candidate model(s) using the model subset until the updated prediction(s) approaches a threshold similarity with the expected result or until a threshold iteration has been reached.

[0085] In the example where the n-th candidate prediction model is the sixth candidate prediction model, the operation 410 can include determining a fourth prediction, a fifth prediction, and a sixth prediction, determining a fourth difference between the fourth prediction and the expected result, detennining a fifth difference between the fifth prediction and the expected result, determining a six difference between tire sixth prediction and the expected result, adjusting, based on the fourth difference, one or more parameters of the fourth candidate prediction model to generate updated fourth candidate prediction model, such that subsequent updated fourth updated prediction(s) approaches the expected result, adjusting, based on the fifth difference, one or more parameters of the fifth candidate prediction model to generate updated fifth candidate prediction model, such that subsequent updated fifth updated prediction(s) approaches the expected result, and adjusting, based on the sixth difference, one or more parameters of the sixth candidate prediction model to generate updated sixth candidate prediction model, such that subsequent updated sixth updated prediction(s) approaches the expected result.

[0086] At operation 412, the process can include determining first performance metric(s) associated with the first candidate prediction model. In embodiments, the operation 412 may include perfonning internal cross-validation of the first candidate prediction model with the test subset and external cross-validation of the first candidate prediction model with an independent dataset unrelated to the training data.

[0087] In embodiments, performing the internal cross-validation of the first candidate prediction model may include determining, using the test subset and / or using a subgroup analysis of the subgroups, a first area under the receiver operating characteristic curve (AUROC) associated with the first candidate prediction model and a first calibration metric associated with first candidate prediction model as first performance metrics. In embodiments, performing the external cross-validation of the first candidate prediction model may include detennining, using the independent dataset, a second AUROC associated with the first candidate prediction model and a second calibration metric associated with the first candidate prediction model to include as also as the first perfonnance metrics. In embodiments, the calibration metrics may be Fl scores.

[0088] At operation 414. the process can include determining second performance metric(s) associated with the second candidate prediction model. In embodiments, the operation 414 mayinclude performing internal cross-validation of the second candidate prediction model with the test subset and external cross-validation of the second candidate prediction model with the independent dataset.

[0089] In embodiments, performing the internal cross-validation of the second candidate prediction model may include detennining, using the test subset and / or using a subgroup analysis of the subgroups, a first AUROC associated with the second candidate prediction model and a first calibration metric associated with second candidate prediction model as second performance metrics. In embodiments, performing the external cross-validation of the second candidate prediction model may include detennining. using the independent dataset, second AUROC associated with the second candidate prediction model and second calibration metric associated with the second candidate prediction model to include also as the second performance metrics. In embodiments, the calibration metrics may be Fl scores.

[0090] At operation 416, the process can include determining a third performance metric(s) associated with the third candidate prediction model. In embodiments, tire operation 414 may include perfonning internal cross-validation of the third candidate prediction model with the test subset and external cross-validation of the third candidate prediction model with the independent dataset.

[0091] In embodiments, performing the internal cross-validation of the third candidate prediction model may include determining, using the test subset and / or using a subgroup analysis of the subgroups, a first AUROC associated with the third candidate prediction model and a first calibration metric associated with third candidate prediction model as third performance metrics. In embodiments, performing the external cross-validation of the third candidate prediction model may include determining, using the independent dataset, a second AUROC associated with the third candidate prediction model and a third calibration metric associated with the second candidate prediction model to include also as the third performance metrics. In embodiments, the calibration metrics may be Fl scores.

[0092] At operation 418, the process can include detennining n-th second perfonnance metric(s) associated with the n-th candidate prediction model. In embodiments, the operation 414 may include performing internal cross-validation of the n-th candidate prediction model with the test subset and external cross-validation of the n-th candidate prediction model with the independent dataset.

[0093] In embodiments, performing the internal cross-validation of the n-th candidate prediction model may include detennining, using the test subset and / or using a subgroup analysis of the subgroups, a first AUROC associated with the n-th candidate prediction model and a firstcalibration metric associated with n-th candidate prediction model as n-th performance metrics. In embodiments, performing the external cross-validation of the n-th candidate prediction model may include determining, using the independent dataset, a second AUROC associated with the n-th candidate prediction model and a second calibration metric associated with the n-th candidate prediction model to include also as the n-th performance metrics. In embodiments, the calibration metrics may be Fl scores.

[0094] In the example where the n-th candidate prediction model is the sixth candidate prediction model, the operation 418 can include performing internal cross-validation and external cross-validation of the fourth candidate prediction model to determine fourth performance metrics, perfonning internal and external cross-validation of the fifth candidate prediction model to determine fifth performance metrics, and performing internal and external cross-validation of the sixth candidate prediction model to determine sixth performance metrics. The fourth, fifth, and sixth performance metrics may also include AUROCs and calibration metrics associated with the fourth candidate prediction model, AUROCs and calibration metrics associated with the fifth candidate prediction model, and AUROCs and calibration metrics associated with the sixth candidate prediction model.

[0095] At operation 420. the process can include calibrating the candidate prediction models. In embodiments, the operation 420 may include determining whether one or more of the calibration metrics associated with each of the candidate prediction models are below a calibration threshold. For example, tire calibration threshold may be an Fl score of 0.5, and if the first candidate prediction model includes one or more calibration metrics below an Fl score of 0.5, then the operation 420 may include calibrating the first candidate prediction model. However, if the first candidate prediction model's calibration metrics are at or above the Fl score of 0.5, then the operation 420 may include not calibrating the first candidate prediction model. In the example, where the first candidate prediction model has an Fl score below 0.5, the operation 420 can further include determining update AUROC(s) using tire test subset and / or the independent dataset after calibration. In embodiments, determining updated AUROC(s) may be performed for all calibrated candidate prediction model(s).

[0096] Alternatively, the operation 420 may include determining whether each of the candidate prediction models includes one or more AUROCs that meet or exceed a threshold AUROC value (e.g., meeting or exceeding a threshold of 0.8). In such an example, the operation 420 may include excluding the candidate model(s) that have AUROCs below the threshold AUROC from calibration and selection as the skin rejection prediction model.

[0097] At operation 422, the process can include selecting one of the candidate prediction model(s) as the skin rejection prediction model. In embodiments, the operation 422 may include selecting the candidate prediction model of the candidate prediction models with the highest AUROC value. In embodiments, the operations 422 may include determining the highest AUROC value after calibration.

[0098] FIG. 5 illustrates a block diagram of an example system 500 for using a skin rejection prediction model to determine the likelihood of skin rejection in a subject. In embodiments, the skin rejection prediction model may correspond to tire skin rejection prediction model generated as described in association with FIGS. 1-4. The example system 500 includes a first camera 502, a second camera 520, a computing device 506, a body fluid collection device 522, and an immunosuppressant detection device 524. In embodiments, first camera 502 may correspond with the first camera(s) 122, the second camera 520 may correspond with the second camera(s) 132, the computing device 506 may correspond with the computing device 124, the body fluid collection device 522 may correspond with the body fluid collection device(s) 102, and the immunosuppressant detection device 524 may correspond with the immunosuppressant detection device 104 of FIG. 1. Tire computing device 506 includes an input component 508, processor(s), memory 512. which further includes a clinical parameter component 514. a prediction model component 516, and an output component 518.

[0099] In embodiments, the body fluid collection device 522 may collect body fluid sample(s), such as blood, from a skin transplant subject. The body fluid collection device 522 may be positioned at or proximate to the subject to collect the body fluid sample(s) from the subject. In embodiments, the body fluid collection device 522 may include a needle and a fluid storage container physically coupled to the needle.

[0100] In embodiments, the body fluid collected by the body fluid collection device 522 may be transferred to the immunosuppressant detection device 524. Tire immunosuppressant detection device 524 may detect, identify, and quantify one or more immunosuppressant(s) from the body fluid. Hie immunosuppressant(s) may be detected, identified, and quantified (in other words, determined) by performing an immunoassay(s) and / or performing mass spectrometry (e.g., liquidchromatography mass spectrometry) on the collected body fluid sample.

[0101] In embodiments, the detecting, identifying, and quantify ing the immunosupprcssant(s) may include detecting a presence of one or more immunosuppressants from the body fluid, identifying the one or more immunosuppressants, and / or determining a concentration of each detected and identified immunosuppressant.

[0102] In embodiments, as each immunosuppressant is detected, identified, and quantified by the immunosuppressant detection device 524, immunosuppressant data associated with the detected, identified, and quantified immunosuppressant (e g., immunosuppressant name and concentration) may be transmitted or transferred to the computing device 506 and stored at the clinical parameter component 514. In embodiments, the immunosuppressant detection device 524 may include a communication component that can transmit the immunosuppressant data to the computing device 506 using a wired connection or a wireless connection. The computing device 506 may receive the immunosuppressant data using its communication component. Alternatively, the immunosuppressant data may be transmitted or transferred to the computing device 506 after all immunosuppressant parameter(s) are determined.

[0103] In embodiments, the first camera(s) 122 may be a 3CCD imaging device configured to emit a red light, a green light, and a blue light towards a skin graft site 504 of the subject and receive, by image sensors of the first camera(s) 122, signals associated with red, blue, and green lights scattered by tire skin graft site 504 in response to the emitted lights. Raw sensor data from the image sensors may be transmitted to the computing device 506, where the raw sensor data may be used by the processor(s) 510 to determine, as the first imaging data, the intensity of the red light scattered by the skin graft site 504, an intensity of the blue light scattered by the skin graft site 504, an intensity of the green light scattered by the skin graft site 504. The intensities of each of the red, green, and blue lights captured by the first camera(s) 122 may be indicative of oxygenation at the skin graft site 504 and the presence of pro-inflammatory biomarkers or anti-inflammatory biomarkers. From the intensity of the red light, the intensity of the blue light, and the intensity of the green light, the processor(s) 510 may further determine, as the first imaging data, a total 3CCD associated with the skin graft site 504.

[0104] Tire processor(s) 510 may further determine, from the raw sensor data and as first imaging data, an absorption of the red light by the skin graft site 504 and an absorption of the blue light by the skin graft site 504. Each absorption may be determined based on determining a difference between the intensity of the scattered light. From the absorption of the red light and the absorption of the blue light, the processor(s) 510 may further determine, as the first imaging data, a difference between the absorption of the red light and the absorption of the green light (R-B at the skin graft site 504). Tire processor(s) 510 may store within the clinical parameter component, the total 3 CCD, tire intensity of the green light at the skin graft site, and the R-B at the skin graft site. Alternatively, the intensities of the colors, the total 3CCD, the absorptions, the R-B, and others may be determined by the first camera 502 and transmitted to the clinical parameter component 514.

[0105] In embodiments, the second camera 520 may be a near IR imaging device. The second camera may be configured to emit near IR light toward the skin graft site 504 and capture, with its image sensor, near IR light scatter by the skin graft site 504 in response to the emitted near IR light. Raw sensor data from the second camera 520 may be transmitted to the computing device 506. Hie processor(s) 510 may determine and store within tire clinical parameter component a maximum intensity of the scattered near IR light associated with the skin graft site 504, a center intensity of the scattered near IR light associated with the skin graft site 504, a minimum intensity of the scattered near IR light associated with the skin graft site 504. The intensity of the scattered near IR light may correspond to blood perfusion at the skin graft site 504. Alternatively, the second camera 520 may determine the maximum intensity, the center intensity and / or the minimum intensity and transmit the intensities to the clinical parameter component 514.

[0106] In embodiments, the input component 508 may include a mouse, a keyboard, a touch screen, and / or the like . In embodiments, a user may utilize the input component to supplement data within the clinical parameter component or input information about the subject such as identifying information about the subject (e.g., name, age, gender, etc.).

[0107] In embodiments, the processor(s) 510 may execute the skin rejection prediction model stored at the prediction model component 516 based on the data from the clinical parameter component 514 to determine the likelihood of skin rejection at the skin graft site 504. Determining the likelihood of skin rejection is described in association with FIG. 6, as well as throughout this disclosure.

[0108] In embodiments, the output component 518 may be a display used to output a report associated with the likelihood of skin rejection. The report may include, as the likelihood of skin rejection, a probability of skin rejection. The report may also include a reason associated with the likelihood of skin rejection (e.g., overimmunosuppression or insufficient immunosuppression), and a recommendation to lower the likelihood of skin rejection. The recommendation may include increasing the dose of one or more immunosuppressants during insufficient immunosuppression or lowering the dose of one or more immunosuppressants during overimmunosuppression. The amount of dose to increase or decrease may be determined by the skin rejection prediction model based on the dose within the immunosuppressant data. The report may further be transmitted to a user device of the clinician and / or the surgeon (e.g., to a mobile phone, a tablet, a personal computer, and / or the like) using the communication component of the computing device 506.

[0109] FIG. 6 illustrates a flow diagram of an example process 600 of using a skin rejection prediction model to predict the likelihood of skin rejection at a skin graft site such as the skin graft site 504. In embodiments, the skin rejection prediction model may be the skin rejection predictionmodel described in association with FIGS. 1-5. In addition to what is discussed below in association with FIG. 6, using the skin rejection prediction model is further discussed in association with the Examples section, as well as throughout this disclosure.

[0110] At operation 602, the process can include determining a first camera data. In embodiments, the first camera data may include data from a 3CCD imaging device. Additional details with respect to the operation 602 are described in association with FIG. 5, as well as throughout this disclosure.

[0111] At operation 604, tire process can include determining second camera data. In embodiments, the first camera data may include data from a near IR imaging device. Additional details with respect to the operation 604 are described in association with FIG. 5, as well as throughout this disclosure.

[0112] At operation 606, the process can include determining immunosuppressant data. Additional details with respect to the operation 606 are described in association with FIG. 5, as well as throughout this disclosure.

[0113] At operation 608, the process can include determining the likelihood of skin rejection using the skin rejection prediction model based on the first camera data, the second camera data, and the immunosuppressant data. Additional details with respect to the operation 608 are described in association with FIG. 5, as well as throughout this disclosure.

[0114] At operation 610, the process can include determining whether the likelihood is greater than or equal to a threshold likelihood. If yes, then at the operation 612, the process can include outputting a report indicating skin rejection. The report can indicate a probability associated with the skin rejection, a reason for the likelihood of skin rejection, and a treatment recommendation. Additional details with respect to the operation 610 are described in association with FIG. 5, as well as throughout this disclosure. If no, then the process can return to the operation 602, where the process continues to gather imaging data and immunosuppressant data from the skin graft site and predict the likelihood of skin rejection at the skin graft site. In embodiments, the skin rejection prediction model may accurately predict the likelihood of skin rejection through 5 days, and therefore, for a skin graft site that is approaching the 5 day limit, a clinician may use the skin rejection prediction model again to predict an updated likelihood of skin rejection for the skin graft site.

[0115] While one or more examples of the techniques described herein have been described, various alterations, additions, permutations, and equivalents thereof are included within the scope of the techniques described herein.

[0116] In the description of examples, reference is made to the accompanying drawings that form a part hereof, which show by way of illustration specific examples of the claimed subject matter. It is to be understood that other examples may be used and that changes or alterations, such as structural changes, may be made. Such examples, changes or alterations are not necessarily departures from the scope with respect to the intended claimed subject matter. While the steps herein may be presented in a certain order, in some cases the ordering may be changed so that certain inputs are provided at different times or in a different order without changing the function of the systems and methods described. The disclosed procedures could also be executed in different orders. Additionally, various computations that arc herein need not be performed in the order disclosed, and other examples using alternative orderings of the computations could be readily implemented. In addition to being reordered, the computations could also be decomposed into subcomputations with the same results.

[0117] Although the subject matter has been described in language specific to structural features and / or methodological acts, it is to be understood that the subject matter defined in the appended claims is not necessarily limited to the specific features or acts described. Rather, the specific features and acts are disclosed as example fonns of implementing the claims.

[0118] The components described herein represent instructions that may be stored in any type of computer-readable medium (also referred to as a computer-readable storage medium or computer-readable storage medium) and may be implemented in software and / or hardware. All of the methods and processes described above may be embodied in, and fully automated via, software code modules and / or computer-executable instructions executed by one or more computers or processors, hardware, or some combination thereof. Some or all of the methods may alternatively be embodied in specialized computer hardware. A computer-readable medium may be, for example, but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of the computer-readable medium would include the following: an electrical connection having one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing. In the context of this document, a computer-readable storage medium may be any tangible medium that can contain or store a program for use by or in connection with an instruction execution system, apparatus, or device. In embodiments, the memories 112, 128. and 512 of FIGS. 1 and 5 respectively may be computer-readable mediums.

[0119] Program code embodied on a computer-readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing. Computer program code for carrying out operations for aspects of the present disclosure may be written in any combination of one or more programming languages, including an object-oriented programming language such as Java, Smalltalk, C++, or the like, and conventional procedural programming languages, such as the “C” programming language or similar programming languages. The program code may execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer, and partly on a remote computer or entirely on the remote computer or server. In the latter scenario, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or the connection may be made to an external computer (for example, through the Internet using an Internet Service Provider).

[0120] Conditional language such as, among others, “may,” “could,” “may” or “might,” unless specifically stated otherwise, are understood within the context to present that certain examples include, while other examples do not include, certain features, elements and / or steps. Thus, such conditional language is not generally intended to imply that certain features, elements, and / or steps are in any way required for one or more examples or that one or more examples necessarily include logic for deciding, with or without user input or prompting, whether certain features, elements and / or steps arc included or arc to be performed in any particular example.

[0121] Conjunctive language such as the phrase “at least one ofX, Y or Z,” unless specifically stated otherwise, is to be understood to present that an item. term. etc. may be either X, Y. or Z. or any combination thereof, including multiples of each element. Unless explicitly described as singular, “a” means singular and plural.

[0122] Any routine descriptions, elements, or blocks in the flow diagrams described herein and / or depicted in the attached figures should be understood as potentially representing modules, segments, or portions of code that include one or more computer-executable instructions for implementing specific logical functions or elements in the routine. Alternate implementations are included within the scope of the examples described herein in which elements or functions may be deleted, or executed out of order from that shown or discussed, including substantially synchronously, in reverse order, with additional operations, or omitting operations, depending on the functionality involved as would be understood by those skilled in the art.

[0123] Many variations and modifications may be made to the above-described examples, the elements of which are to be understood as being among other acceptable examples. All suchmodifications and variations are intended to be included herein within the scope of this disclosure and protected by the following claims.

[0124] Aspects of the present disclosure arc described herein with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the disclosure. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general-purpose computer, special purpose computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions / acts specified in the flowchart and / or block diagram block or blocks.

[0125] These computer program instructions may also be stored in a computer-readable medium that can direct a computer, other programmable data processing apparatus, or other devices to function in a particular manner, such that the instructions stored in tire computer-readable medium produce an article of manufacture including instructions which implement the function / act specified in the flowchart and / or block diagram block or blocks. The computer program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other devices to cause a series of operational steps to be performed on the computer, other programmable apparatus, or other devices to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide processes for implementing the functions / acts specified in the flowchart and / or block diagram block or blocks.

[0126] The flowchart and block diagrams in the Figures illustrate the architecture, functionality', and operation of possible implementations of systems, methods, and computer program products according to embodiments of the present disclosure. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing the specified logical function(s). It should also be noted that, in some alternative implementations, the functions noted in the blocks may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and / or flowchart illustration, and combinations of blocks in tire block diagrams and / or flowchart illustration, can be implemented by special purpose hardware-based systems thatperform the specified functions or acts, or combinations of special purpose hardware and computer instructions.

[0127] Although the figures show a specific order of method steps, the order of the steps may differ from what is depicted. Also, two or more steps may be performed concurrently or with partial concurrence. Such variation will depend on the software and hardware systems chosen and on the designer's choice. All such variations are within the scope of the disclosure. Likewise, software implementations could be accomplished with standard programming techniques with rule-based logic and other logic to accomplish the various connection steps, processing steps, comparison steps, and decision steps.

[0128] As will be understood by one of ordinary skill in the art. each embodiment disclosed herein can comprise, consist essentially of. or consist of its particular stated element, step, ingredient, or component. Thus, the terms “include” or “including” should be interpreted to recite: “comprise, consist of, or consist essentially of.” Tire transition term “comprise” or “comprises” means includes, but is not limited to, and allows for the inclusion of unspecified elements, steps, ingredients, or components, even in major amounts. Tire transitional phrase “consisting of’ excludes any element, step, ingredient, or component not specified. The transition phrase “consisting essentially of’ limits the scope of the embodiment to the specified elements, steps, ingredients, or components and to those that do not materially affect the embodiment.EXEMPLARY EMBODIMENTS

[0129] 1 : A method of generating a skin rejection prediction model for a subject comprising: receiving first raw sensor data from a plurality of first imaging devices: receiving second raw sensor data from a plurality of second imaging devices: determining, from the first raw sensor data, first imaging data, the first imaging data comprising intensities associated with a plurality of color channels associated with the first imaging devices; determining, from the second raw sensor data, second imaging data, the second imaging data comprising intensities associated with the second imaging devices; generating a data structure storing training data, the training data comprising the first imaging data, the second imaging data, and clinical outcomes associated with the first imaging data and the second imaging data; executing a plurality of feature selection algorithms on the training data to generate a model subset, the model subset comprising a subset of the training data less than a total amount of the training data; training a plurality of candidate prediction models using the model subset; determining a performance metric for each trained candidate prediction model of the plurality of trained candidate prediction models; and selecting, based on the performance metric, a candidate model from the plurality of trained candidate prediction models asthe skin rejection prediction model, wherein the skin rejection prediction model determines a likelihood of skin rejection at a skin graft site of the subject.

[0130] 2: Tire method of embodiment 1, further comprising: Determining whether one or more values associated with the training data is missing; and filling in, based on determining one or more missing values associated with the training data, the one or more missing values.

[0131] 3: The method of embodiments 1 or 2, wherein the first imaging data further comprises: total intensities associated with the first raw sensor data; intensities associated with green lights scattered by skin graft sites of a plurality of individuals different from the subject; intensities associated with red lights scattered by the skin graft sites; intensities associated with blue lights scattered by the skin graft sites; differences between absorptions between tire red lights and the blue lights; and / or differences between the intensities of the red lights and the blue lights.

[0132] 4: The method of any one of embodiments 1-3, wherein the second imaging data further comprises: mean intensities associated with infrared lights scattered by skin graft sites of a plurality of individuals different from the subject; standard deviations of the intensities associated with the infrared lights; center intensities associated with the infrared lights; maximum intensities associated with the infrared lights; and / or minimum intensities associated with the infrared lights.

[0133] 5: The method of any one of embodiments 1-4, further comprising: determining a training subset from the training data, the training subset being less than the training data; and determining a test subset from tire training data, the test subset comprising a portion of the training data excluded from the training subset.

[0134] 6: The method of embodiment 5, wherein generating the model subset comprises: generating a resampled subset from the training subset, wherein each entry of the resampled subset is selected from the training subset with replacement; executing a random forest algorithm on the resampled subset; ranking, based on a result of the random forest algorithm, features associated with the resampled subset; eliminating a lowest ranking feature of the features; and generating the model subset, the model subset comprising remaining features of the resampled subset.

[0135] 7: The method of embodiments 5 or 6, wherein determining the performance metric comprises determining, by inputting the test subset into tire trained candidate prediction model, a magnitude of an area under a receiver operator curve (AUROC) associated with the trained candidate prediction model and a calibration metric associated with the candidate model.

[0136] 8: The method of embodiment 7, wherein selecting the skin rejection prediction model comprises selecting a candidate prediction model of the plurality of candidate prediction models with a highest magnitude of AUROC.

[0137] 9: The method of any one of embodiments 1-8, wherein training the plurality of candidate prediction models comprises: determining, for each candidate prediction model of the plurality of candidate prediction models and based on inputting the model subset into the candidate model, a prediction associated with the candidate prediction model; determining a difference between the prediction and the clinical outcomes; and adjusting, based on the difference, a parameter of the candidate prediction model.

[0138] 10: The method of any one of embodiments 1-9, wherein the training data further comprises immunosuppressant data, the immunosuppressant data being associated with body fluid samples from a plurality of individuals different from the subject.

[0139] 11: The method of embodiment 10, wherein the immunosuppressant data comprises a presence or absence of one or more immunosuppressants.

[0140] 12: A method of predicting skin rejection at a skin graft site of a subject comprising: receiving a first value of a first clinical parameter associated with the skin graft site and a second value of a second clinical parameter associated with the skin graft site; executing a skin rejection prediction model using the first value and the second value to generate a prediction associated with skin rejection at the skin graft site, wherein the skin rejection prediction model is generated by performing operations comprising: receiving first raw sensor data from a plurality of first imaging devices; receiving second raw sensor data from a plurality of second imaging devices; determining, from the first raw sensor data, first imaging data, the first imaging data comprising intensities associated with a plurality of color channels associated with the first imaging devices; determining, from the second raw sensor data, second imaging data, the second imaging data comprising intensities associated with the second imaging devices; generating a data structure storing a training data, the training data comprising the first imaging data, the second imaging data, and clinical outcomes associated with the first imaging data and the second imaging data; executing a plurality of feature selection algorithms on the training data to generate a model subset, the model subset comprising a subset of the training data less than a total amount of the training data; training a plurality of candidate prediction models using the model subset; determining a performance metric for each trained candidate prediction model of the plurality of trained candidate prediction models; and selecting, based on tire performance metric, a trained candidate prediction model from the plurality of trained candidate prediction models as the skin rejection prediction model; and outputting, by the skin rejection prediction model based on the first value and the second value, the prediction associated with skin rejection at the skin graft site.

[0141] 13: The method of embodiment 12, wherein the prediction associated with skin rejection comprises a likelihood of skin rejection at the skin graft site.

[0142] 14: The method of embodiments 12 or 13, wherein the first clinical parameter comprises: total intensity associated with light scattered by the skin graft site; an intensity associated with green light scattered by the skin graft site; a difference of an absorption of blue light by the skin graft site and a red light by the skin graft site; or a difference of an intensity of blue light scattered by the skin graft site and an intensity of red light scattered by the skin graft site.

[0143] 15: The method of any one of embodiments 12-14, wherein the second clinical parameter comprises: center intensity associated with infrared light scattered by the skin graft site; minimum intensity associated with infrared light scattered by the skin graft site; and / or maximum intensity associated with infrared light scattered by the skin graft site.

[0144] 16: The method of any one of embodiments 12-15, further comprising: receiving third value associated with immunosuppressant data, the immunosuppressant data being determined from a body fluid sample associated with the subject; and wherein the prediction associated with the skin rejection is further determined based on the third value.

[0145] 17: Tire method of embodiment 16, wherein outputting the prediction associated with skin rejection comprises: determining, based on the immunosuppressant data, overimmunosuppression or insufficient immunosuppression associated with the subject.

[0146] 18: The method of embodiment 17, wherein outputting the prediction associated with skin rejection further comprises: determining, based on determining overimmunosuppression and the immunosuppressant data, a first amount to decrease a dose of an immunosuppressant; determining, based on determining insufficient immunosuppression and the immunosuppressant data, a second amount to increase the dose of the immunosuppressant; and outputting a recommendation associated with increasing a first amount of the dose or decreasing a second amount of the dose.

[0147] 19: A system for generating a skin rejection prediction model for a subject comprising: one or more processors; and one or more non-transitory computer-readable media storing computer-executable instructions that, when executed, cause the one or more processors to perform operations comprising: receiving first raw sensor data from a plurality of first imaging devices; receiving second raw sensor data from a plurality of second imaging devices; determining, from the first raw sensor data, first imaging data, the first imaging data comprising intensities associated with a plurality of color channels associated with the first imaging devices; determining, from the second raw sensor data, second imaging data, the second imaging data comprising intensities associated with the second imaging devices; generating a data structure storing training data, the training data comprising the first imaging data, the second imaging data, and clinical outcomes associated with the first imaging data and the second imaging data; executing a pluralityof feature selection algorithms on the training data to generate a model subset, the model subset comprising a subset of the training data less than a total amount of the training data; training a plurality of candidate prediction models using the model subset; determining a performance metric for each trained candidate prediction model of the plurality of trained candidate prediction models; and selecting, based on the performance metric, a trained candidate prediction model from the plurality of trained candidate prediction models as the skin rejection prediction model, wherein the skin rejection prediction model is configured to determine a likelihood of skin rejection at a skin graft site of the subject.

[0148] 20: A system for predicting skin rejection at a skin graft site of a subject comprising: one or more processors; an output component; and one or more computer-readable media storing computer-executable instructions that, when executed, cause the one or more processors to perform operations comprising: receiving a first value of a first clinical parameter associated with the skin graft site and a second value of a second clinical parameter associated with the skin graft site: executing a skin rejection prediction model using the first value and the second value to generate a prediction associated with skin rejection at the skin graft site, wherein tire skin rejection prediction model is generated by performing operations comprising: receiving first raw sensor data from a plurality of first imaging devices; receiving second raw sensor data from a plurality of second imaging devices; determining, from the first raw sensor data, first imaging data, the first imaging data comprising intensities associated with a plurality of color channels associated with the first imaging device; determining, from the second raw sensor data, second imaging data, the second imaging data comprising intensities associated with the second imaging device; generating a data structure storing a training data, the training data comprising the first imaging data, tire second imaging data, and clinical outcomes associated with the first imaging data and the second imaging data; executing a plurality of feature selection algorithms on the training data to generate a model subset, the model subset comprising a subset of the training data less than a total amount of the training data; training a plurality of candidate prediction models using the model subset; determining a performance metric for each trained candidate prediction model of the plurality of trained candidate prediction models; and selecting, based on the perfonnance metric, a candidate model from the plurality of candidate models as the skin rejection prediction model; and outputting, by the skin rejection prediction model and the output component and based on the first value and the second value, the prediction associated with skin rejection at the skin graft site.

[0149] While the example clauses described above are described with respect to one particular implementation, it should be understood that, in the context of this document, the content of the example clauses can also be implemented via a method, device, system, computer-readable medium, and / or another implementation. Additionally, any of examples 1-20 may be implemented alone or in combination with any other one or more of the examples 1-20.EXAMPLES

[0150] Example 1: Materials and Methods.

[0151] To create the training data, a total of 24 miniature swine were purchased from the Massachusetts General Hospital. Twenty-two swine received an SLA -mismatched hindlimb allograft, and two additional swine received three elevated latissimus flaps (two in one animal) to serve as controls. Allograft animals underwent no more than three cycles of immunosuppression / no immunosuppression to simulate graft rejection, which ran up to 196 post-operative days (POD). Sw ine skin is comparable to human skin in clinical and histopathological settings. Assessment of tissue perfusion was conducted by using infrared (IR) images to capture local temperature gradients related to blood flow' across the skin on the allograft and its surrounding area. Three charge coupled device (3CCD) imaging measures oxygenation by deconstructing a color image into red, green, and blue channels to create a contrast enhanced image that is sensitive to the absorption spectrum of oxygenated hemoglobin. Both IR and 3CCD data were collected from the first POD until the end of the third rejection cycle. Blood samples were collected weekly and for cause, and a Luminex assay of 13 pro- or anti-inflammatory biomarkers was done. To determine skin rejection status, visual evaluation of the graph skin was conducted by a transplant surgeon using grades from 0 (no difference between graft skin and native skin) to 4 (full graft epidermolysis and necrosis) based on the Banff grading criteria established for human VCAs. The swine data sets were used for exploratory data analysis and the development of machine learning models.

[0152] To create the independent dataset for the external cross-validation, human patient data from upper extremity transplant recipients was used for translational validation of tire models developed above. The data was obtained from four human patients, three wdth bilateral hand allotransplants and one patient with a unilateral hand allotransplant. Data collection for IR and 3CCD measures and rejection grading was conducted as previously described for swine but all measures were taken on both sides of each hand. Samples were collected between 97 and 3155 PODs for a total of 74 samples.

[0153] Example 2: Data Quality Check.

[0154] Considering that the spectroscopic data (including the IR and 3CCD measures) may have both true signal and background noises, the following steps were taken for data processing: 1) Filtering background noise and false signals; 2) Locally smoothing of the filtered signals; and 3) Data normalization to the mean of 0 for each measurement using the inverse normal transformation (or rankZ). Time-series modeling was then used to identify signal differences between the transplanted animals and the controls, and to examine predictability of the signals. Time series modeling is a regression technique that forecasts future values of a fluctuated curve based on its known history. A model that includes both auto-regression and moving average, known as ARMA was used for prediction. In the ARMA model, the impact of previous lags along with the residuals is considered for forecasting the future values of the time series. Missing values were estimated by using the random forest algorithm.

[0155] Example 3: Principal Components Analysis.

[0156] In addition to the spectroscopic data, a total of 13 pro- or anti-inflammatory cytokines were assayed weekly, which generated 103 measurement samples. The heatmap of the cytokine data showed that the cytokines can be divided into two clusters corresponding to the pro- or antiinflammatory ones, respectively. The proinflammatory biomarkers include 10 cytokines, such as GMCSF, IL-1A, IL-1B, IL-IRA, IL-2, IL-4, IL-6, IL-10, IL-18, and TNFA, while the antiinflammatory biomarkers include only 3 cytokines: IL-8, IL- 12, and IFN-y. Similar to the spectroscopic data, the samples can be generally divided into two big clusters (red or green color enriched (e.g., high red color intensity or high green color intensity)). Interestingly, the red cluster, which has high values of proinflammatory biomarkers (such as IL-1A, IL-2, and IL- 10), is enriched with samples (or POD-recipient combinations) in the high probability of rejection state, and vice versa for the blue cluster.

[0157] Principal components analysis on the combined IR and 3CCD data showed that the top three principal components (PCs) can explain 85% of the variation in tire spectroscopic data. The first PC loaded heavily towards red channel 3CCD measures, while the second PC loaded mainly to IR measures. The third PC negatively loaded to total 3 CCD and red minus blue light absorbance indicating low oxygenation, while positively to the IR standard deviation measure indicating fluctuating blood perfusion.

[0158] Tire hcatmap of correlations demonstrated that each of the PCs was associated with specific pro- and / or anti-inflammatory cytokines. Hie first PC was positively associated with IL- 12 and negatively with GMCSF, while the second PC was positively correlated with IFN-y and negatively with IL-2 and TNFA. The third PC was specifically associated with IFN-y, displaying a similar time-course pattern to the anti -infl a atory cytokine known to play a role in woundhealing and non-rejection of allograft. These clearly demonstrate that the spectroscopic data can capture the immune responses of swine after the allograft.

[0159] Example 4: skin rejection prediction model development.

[0160] Since each recipient animal has a different or individual immune response towards the donor, thus the spectroscopic measurements obtained from the recipients over the post-operational course, provide a large heterogeneous sample size (N=l,124) for machine learning model development and testing. A standardized pipeline for supervised machine learning was employed to accurately classify or predict the samples as “rejection” and “non-rejection” based on the Banff grading.

[0161] First, POD-matched spectroscopic and clinical data were used first for machine learning. Seven features (F7) were identified to approach the maximal classification accuracy, including three 3CCD measures of oxygenation (i.e. total 3CCD, relative enrichment (e.g., intensity) of the green color channel, and the difference between the red and blue channel), three measures of blood perfusion (i.e., maximum, minimum, and center near IR intensities), and the presence or absence of immunosuppressive drugs. Multiple models were then trained using six different algorithms (see, e.g.. the candidate prediction models described in association with FIG. 4). All of the six candidate models showed Fl scores > 0.5, indicating that there is no significant class imbalance problem. Of the six algorithms, the random forest model (F7-RF) demonstrated the best discrimination with an area under tire receiver operating characteristic curve (AUROC) of 0.91, followed by the F7-XGB model with an AUROC of 0.89. The six models were evaluated by individual swine recipients (or subgroup analysis) and similar results were obtained. However, the top two models displayed S-shaped curves in the calibration plots, which indicated over- and underestimation in the low and high ends, respectively. In addition, calibration statistics also showed that the two models needed a post-training step for calibration. Sigmoid correction (or Platt scaling) is the appropriate method for S-shaped curves. After the correction, the F7-RF model displayed excellent calibration, as was the F7-XGB model.

[0162] Translational validation was conducted with data from four human patients, three with bilateral hand allotransplants and one patient with a unilateral hand allotransplant. Samples were collected between 97 and 3155 PODs for a total of 74 samples. In these samples, the F7-RF and F7-XGB models displayed superior discrimination with AUROCs of 0.82 and 0.87, respectively, and adequate calibration after the sigmoid correction.

[0163] Second. POD-mismatched data was used to establish the seven features’ predictability for clinical outcomes (“rejection” and “non-rejection”). The Banff grading data was deliberately delayed from the F7 predictors data by a series of time differences ranging from 1 to 7 days.Compared to the POD-matched modeling results, minor degradation was observed in the POD- mismatched models in terms of the magnitude of correlation (in absolute value) between total 3CCD and tire Banff grading, and of the F7-RF model’s and F7-XGB model’s predictive accuracies as well (Table 1). However, up to the time difference of five days, both the F7-RF and F7-XGB models still maintained good discrimination (with AUROC of 0.824 and 0.810, respectively) and calibration in the swine training test data. Furthermore, evidence from translational validation with the human patient data indicated that the two models had AUROC of 0.744 and 0.776, respectively, up to a time difference of five days. These clearly indicate that the seven features have reliable predictability for clinical outcomes in a timeframe of up to five days before the onset of skin rejection.Table 1* The clinical data was deliberately delayed by a series of days from 1 to 7, compared to the spectroscopic data, in order to test the seven features’ predictability. 0 represents the post- operational day (POD)-matched data as a control. When the area under the curve (AUROC, but referred to as AUC in Table 1) in the testing data is 0.70 or below, we stopped further testing.

Claims

CLAIMSWHAT IS CLAIMED IS:

1. A method of generating a skin rejection prediction model for a subject comprising: receiving first raw sensor data from a plurality of first imaging devices; receiving second raw sensor data from a plurality of second imaging devices; determining, from the first raw sensor data, first imaging data, the first imaging data comprising intensities associated with a plurality of color channels associated with the first imaging devices; determining, from the second raw sensor data, second imaging data, the second imaging data comprising intensities associated with the second imaging devices; generating a data structure storing training data, the training data comprising the first imaging data, the second imaging data, and clinical outcomes associated with the first imaging data and the second imaging data; executing a plurality of feature selection algorithms on the training data to generate a model subset, the model subset comprising a subset of the training data less than a total amount of the training data; training a plurality of candidate prediction models using the model subset; determining a performance metric for each trained candidate prediction model of the plurality of trained candidate prediction models; and selecting, based on the performance metric, a candidate model from the plurality of trained candidate prediction models as the skin rejection prediction model, wherein the skin rejection prediction model determines a likelihood of skin rejection at a skin graft site of the subject.

2. Tire method of claim 1, further comprising:Determining whether one or more values associated with the training data is missing; and filling in, based on determining one or more missing values associated with the training data, the one or more missing values.

3. The method of claim 1, wherein the first imaging data further comprises: total intensities associated with the first raw sensor data; intensities associated with green lights scattered by skin graft sites of a plurality of individuals different from the subject; intensities associated with red lights scattered by the skin graft sites;intensities associated with blue lights scattered by the skin graft sites; differences between absorptions between the red lights and the blue lights; and / or differences between the intensities of the red lights and the blue lights.

4. The method of claim 1, wherein the second imaging data further comprises: mean intensities associated with infrared lights scattered by skin graft sites of a plurality of individuals different from the subject; standard deviations of the intensities associated with the infrared lights; center intensities associated with the infrared lights; maximum intensities associated with the infrared lights; and / or minimum intensities associated with the infrared lights.

5. The method of claim 1, further comprising: determining a training subset from the training data, the training subset being less than the training data; and determining a test subset from the training data, the test subset comprising a portion of the training data excluded from the training subset.

6. The method of claim 5, wherein generating the model subset comprises: generating a resampled subset from the training subset, wherein each entry of the resampled subset is selected from the training subset with replacement; executing a random forest algorithm on the resampled subset; ranking, based on a result of tire random forest algorithm, features associated with tire resampled subset; eliminating a lowest ranking feature of the features; and generating the model subset, the model subset comprising remaining features of the resampled subset.

7. Tire method of claim 5, wherein determining the performance metric comprises determining, by inputting the test subset into the trained candidate prediction model, a magnitude of an area under a receiver operator curve (AUROC) associated with the trained candidate prediction model and a calibration metric associated with the candidate model.

8. The method of claim 7, wherein selecting the skin rejection prediction model comprises selecting a candidate prediction model of the plurality of candidate prediction models with a highest magnitude of AUROC.

9. Tire method of claim 1, wherein training the plurality of candidate prediction models comprises: determining, for each candidate prediction model of the plurality of candidate prediction models and based on inputting the model subset into the candidate model, a prediction associated with the candidate prediction model; determining a difference between the prediction and the clinical outcomes; and adjusting, based on the difference, a parameter of the candidate prediction model.

10. The method of claim 1, wherein the training data further comprises immunosuppressant data, the immunosuppressant data being associated with body fluid samples from a plurality of individuals different from the subject.

11. Tire method of claim 10, wherein the immunosuppressant data comprises a presence or absence of one or more immunosuppressants.

12. A method of predicting skin rejection at a skin graft site of a subject comprising: receiving a first value of a first clinical parameter associated with the skin graft site and a second value of a second clinical parameter associated with the skin graft site; executing a skin rejection prediction model using the first value and the second value to generate a prediction associated with skin rejection at the skin graft site, wherein the skin rejection prediction model is generated by performing operations comprising: receiving first raw sensor data from a plurality of first imaging devices; receiving second raw sensor data from a plurality of second imaging devices; determining, from the first raw sensor data, first imaging data, the first imaging data comprising intensities associated with a plurality of color channels associated with the first imaging devices; determining, from the second raw sensor data, second imaging data, the second imaging data comprising intensities associated with the second imaging devices;generating a data structure storing a training data, the training data comprising the first imaging data, the second imaging data, and clinical outcomes associated with the first imaging data and the second imaging data; executing a plurality of feature selection algorithms on tire training data to generate a model subset, the model subset comprising a subset of the training data less than a total amount of the training data; training a plurality of candidate prediction models using the model subset; determining a performance metric for each trained candidate prediction model of the plurality of trained candidate prediction models; and selecting, based on the performance metric, a trained candidate prediction model from tire plurality of trained candidate prediction models as the skin rejection prediction model; and outputting, by the skin rejection prediction model based on the first value and the second value, the prediction associated with skin rejection at the skin graft site.

13. The method of claim 12, wherein the prediction associated with skin rejection comprises a likelihood of skin rejection at the skin graft site.

14. The method of claim 12, wherein the first clinical parameter comprises: total intensity associated with light scattered by the skin graft site; an intensity associated with green light scattered by the skin graft site; a difference of an absorption of blue light by the skin graft site and a red light by the skin graft site; or a difference of an intensity of blue light scattered by the skin graft site and an intensity of red light scattered by the skin graft site.

15. The method of claim 12, wherein the second clinical parameter comprises: center intensity associated with infrared light scattered by the skin graft site; minimum intensity associated with infrared light scattered by the skin graft site; and / or maximum intensity associated with infrared light scattered by the skin graft site.

16. The method of claim 12, further comprising: receiving third value associated with immunosuppressant data, the immunosuppressant data being determined from a body fluid sample associated with the subject: andwherein the prediction associated with the skin rejection is further determined based on the third value.

17. The method of claim 16, wherein outputting the prediction associated with skin rejection comprises: determining, based on the immunosuppressant data, overimmunosuppression or insufficient immunosuppression associated with the subject.

18. The method of claim 17, wherein outputting the prediction associated with skin rejection further comprises: determining, based on determining overimmunosuppression and the immunosuppressant data, a first amount to decrease a dose of an immunosuppressant; determining, based on determining insufficient immunosuppression and the immunosuppressant data, a second amount to increase tire dose of the immunosuppressant; and outputting a recommendation associated with increasing a first amount of the dose or decreasing a second amount of the dose.

19. A system for generating a skin rejection prediction model for a subject comprising: one or more processors; and one or more non-transitory computer-readable media storing computer-executable instructions that, when executed, cause the one or more processors to perform operations comprising: receiving first raw sensor data from a plurality of first imaging devices: receiving second raw sensor data from a plurality of second imaging devices: determining, from the first raw sensor data, first imaging data, the first imaging data comprising intensities associated with a plurality of color channels associated with the first imaging devices; determining, from the second raw sensor data, second imaging data, tire second imaging data comprising intensities associated with the second imaging devices; generating a data structure storing training data, the training data comprising the first imaging data, the second imaging data, and clinical outcomes associated with the first imaging data and the second imaging data;executing a plurality of feature selection algorithms on the training data to generate a model subset, the model subset comprising a subset of the training data less than a total amount of the training data; training a plurality of candidate prediction models using the model subset; determining a performance metric for each trained candidate prediction model of the plurality of trained candidate prediction models; and selecting, based on the performance metric, a trained candidate prediction model from the plurality of trained candidate prediction models as the skin rejection prediction model, wherein the skin rejection prediction model is configured to determine a likelihood of skin rejection at a skin graft site of the subject.

20. A system for predicting skin rejection at a skin graft site of a subject comprising: one or more processors; an output component; and one or more computer-readable media storing computer-executable instructions that, when executed, cause the one or more processors to perform operations comprising: receiving a first value of a first clinical parameter associated with the skin graft site and a second value of a second clinical parameter associated with the skin graft site; executing a skin rejection prediction model using the first value and the second value to generate a prediction associated with skin rejection at the skin graft site, wherein the skin rejection prediction model is generated by performing operations comprising: receiving first raw sensor data from a plurality of first imaging devices: receiving second raw sensor data from a plurality of second imaging devices; determining, from the first raw sensor data, first imaging data, the first imaging data comprising intensities associated with a plurality of color channels associated with the first imaging device; determining, from the second raw sensor data, second imaging data, the second imaging data comprising intensities associated with the second imaging device; generating a data structure storing a training data, the training data comprising the first imaging data, the second imaging data, and clinical outcomes associated with the first imaging data and the second imaging data;executing a plurality of feature selection algorithms on the training data to generate a model subset, the model subset comprising a subset of the training data less than a total amount of the training data; training a plurality of candidate prediction models using the model subset; determining a performance metric for each trained candidate prediction model of the plurality of trained candidate prediction models; and selecting, based on the performance metric, a candidate model from the plurality of candidate models as the skin rejection prediction model; and outputting, by the skin rejection prediction model and the output component and based on the first value and tire second value, the prediction associated with skin rejection at the skin graft site.

Images