Identifying neurological conditions using image analysis

Abstract

Identifying a neurological condition of a patient using image analysis is described. An image of patient tissue displays a plurality of nerves and a plurality of layers of the patient tissue. Each nerve of the plurality of nerves and each layer of the plurality of layers of the patient tissue in the image are marked using a trained model. An amount of nerve area relative to tissue area is determined based on the markers assigned to the plurality of nerves and the plurality of layers of the patient tissue. A neural condition is determined based on the amount of the neural area relative to the tissue area.

Description

Background Art

[0001] This application claims the benefit of U.S. Provisional Patent Application No. 63 / 369,739, filed on July 28, 2022, the disclosure of which is hereby incorporated by reference in its entirety. Background Art

[0003] Cancer treatments such as chemotherapy often damage the nerves of patients. In particular, chemotherapy can cause peripheral neuropathy. Peripheral neuropathy can be a debilitating side effect of cancer treatment that can occur when chemotherapy drugs damage the peripheral nerves outside the brain and spinal cord. Symptoms of peripheral neuropathy can include pain, burning, tingling, numbness, electric shock sensations, pins and needles sensations, temperature sensitivity, etc. Peripheral neuropathy can start in the extremities (e.g., hands and feet) and move upwards.

[0004] Diagnosing peripheral neuropathy may require determining the condition of the nerves. Traditionally, healthcare professionals use nerve conduction to determine the nerve condition. Nerve conduction involves measuring the speed at which an electrical impulse travels through a nerve. In particular, one electrode stimulates the nerve with a mild electrical impulse, and another electrode located downstream records the result. This can be used to determine the nerve conduction velocity (NCV). The NCV can be used to determine nerve damage and disruption. However, this can be a costly, invasive, time-consuming, and error-prone process. Summary of the Invention

[0005] Systems, devices, apparatuses, methods, and / or computer program product embodiments for using image analysis to identify neuropathy and / or combinations and sub-combinations thereof are provided herein.

[0006] A given embodiment includes a diagnostic method for determining a nerve condition. The method includes receiving an image of a patient's skin. The image shows multiple nerves and multiple layers of the patient's skin. The method also includes using a trained model to label each of the multiple nerves and each of the multiple layers of the patient's skin in the image. The trained model has been trained to identify differences between the nerves and the multiple layers of skin tissue. The method also includes determining the amount of the nerve area relative to the area of the skin tissue layer based on the labels assigned to the multiple nerves and multiple layers of the patient's skin, and determining the nerve condition based on the amount of the nerve area relative to the tissue area layer.

[0007] In some embodiments, the amount of the nerve area relative to the tissue area (e.g., epidermal area) corresponds to the nerve density, and determining the nerve condition includes determining the presence of peripheral neuropathy when the nerve density is below a threshold amount.

[0008] In some embodiments, each nerve and each layer among a plurality of nerves and a plurality of layers in an image of a patient's skin includes: using a model to determine whether a pixel among a plurality of pixels in an input image corresponds to a nerve among the plurality of nerves of the patient's skin or a corresponding layer among the plurality of layers; and assigning a corresponding label to each pixel among the plurality of pixels, the corresponding label indicating whether each corresponding pixel corresponds to a nerve of the patient's skin or a corresponding layer among the plurality of layers.

[0009] In some embodiments, the method further includes training the model. Training of the model includes assigning training labels to each nerve and each layer among a plurality of layers of tissues in each of a plurality of training images. Training of the model further includes inputting the plurality of training images into a convolutional neural network to label each nerve and each layer among a plurality of layers of tissues in each of the plurality of training images, and validating each corresponding label assigned to each nerve and each layer among a plurality of layers of tissues in each of the plurality of training images by the convolutional neural network with the corresponding training label corresponding to each nerve and each layer of the plurality of labels. Assigning training labels to each nerve and each layer among a plurality of layers of tissues in each of the plurality of training images may include applying one or more filters to each of the plurality of training images to identify nerves in each of the plurality of training images. Training of the model may further include inputting a mirrored or rotated version of each of the plurality of training images into the convolutional neural network to label each nerve and each layer among a plurality of layers of tissues in the mirrored or rotated version of each of the plurality of training images.

[0010] In some embodiments, a set of images among the plurality of images includes copies of a single image, and each copy includes a different contrast, brightness, cropping, distortion, or rotation from the single image.

[0011] In some embodiments, each image includes a z-stack image.

[0012] Another given embodiment includes a diagnostic system for determining a nerve condition. The system includes a memory storing instructions and a processor coupled to the memory. The instructions, when executed by the processor, cause the processor to receive an image of a patient's skin. The image shows a plurality of nerves and a plurality of layers of the patient's skin. The instructions, when executed by the processor, further cause the processor to use a trained model to label each nerve and each layer among the plurality of nerves and the plurality of layers of the patient's skin in the image. The trained model has been trained to identify differences between nerves and a plurality of layers of skin tissue. The instructions, when executed by the processor, further cause the processor to determine the amount of a layer of nerve area relative to the skin tissue area based on the labels assigned to the plurality of nerves and the plurality of layers of the patient's skin, and to determine the nerve condition based on the amount of the layer of nerve area relative to the tissue area.

[0013] In some embodiments, the amount of nerve area relative to tissue area (e.g., epidermis) corresponds to nerve density, and determining a nerve condition includes determining the presence of peripheral neuropathy when the nerve density is below a threshold amount.

[0014] In some embodiments, labeling each nerve of a plurality of nerves and each layer of a plurality of layers in an image of a patient's skin includes: using a model to determine whether a pixel among a plurality of pixels in an input image corresponds to a nerve among the plurality of nerves of the patient's skin or a corresponding layer among the plurality of layers; and assigning a corresponding label to each pixel among the plurality of pixels, the corresponding label indicating whether each corresponding pixel corresponds to a nerve of the patient's skin or a corresponding layer among the plurality of layers.

[0015] In some embodiments, the instructions, when executed, further cause the processor to train the model. Training of the model includes assigning training labels to each nerve and each layer of a plurality of layers of tissue in each of a plurality of training images. Training of the model further includes inputting the plurality of training images into a convolutional neural network to label each nerve and each layer of a plurality of layers of tissue in each of the plurality of training images, and validating each corresponding label assigned by the convolutional neural network to each nerve and each layer of a plurality of layers of tissue in each of the plurality of training images with the corresponding training label corresponding to the nerve and each layer of the plurality of labels. Assigning training labels to each nerve and each layer of a plurality of layers of tissue in each of the plurality of training images may include applying one or more filters to each of the plurality of training images to identify nerves in each of the plurality of training images. Training of the model may further include inputting mirror or rotated versions of each of the plurality of training images into the convolutional neural network to label each nerve and each layer of a plurality of layers of tissue in the mirror or rotated versions of each of the plurality of training images.

[0016] In some embodiments, a set of images among the plurality of images includes copies of a single image, and each copy includes a different contrast, brightness, cropping, distortion, or rotation from the single image.

[0017] In some embodiments, each image includes a z-stack image.

[0018] Another given embodiment includes a non-transitory machine-readable medium having instructions stored thereon that, when executed by at least one computing device, cause the at least one computing device to perform operations including receiving an image of a patient's skin. The image shows multiple nerves and multiple layers of the patient's skin. The operations further include using a trained model to label each of the multiple nerves and each of the multiple layers in the image of the patient's skin. The trained model has been trained to recognize differences between nerves and the multiple layers of skin tissue. The operations further include determining, based on the labels assigned to the multiple nerves and multiple layers of the patient's skin, the amount of the layer of nerve area relative to the skin tissue area, and determining a nerve condition based on the amount of the layer of nerve area relative to the tissue area.

[0019] In some embodiments, the amount of nerve area relative to the tissue area (e.g., epidermis) corresponds to nerve density, and determining the nerve condition includes determining the presence of peripheral neuropathy when the nerve density is below a threshold amount.

[0020] In some embodiments, labeling each of the multiple nerves and each of the multiple layers in the image of the patient's skin includes: using the model to determine whether a pixel among the multiple pixels in the input image corresponds to a nerve among the multiple nerves of the patient's skin or a corresponding layer among the multiple layers; and assigning a corresponding label to each of the multiple pixels, the corresponding label indicating whether each corresponding pixel corresponds to a nerve of the patient's skin or a corresponding layer among the multiple layers.

[0021] In some embodiments, the operations further include training the model. Training the model includes assigning training labels to each of the multiple nerves and each of the multiple layers of tissue in each of the multiple training images. Training the model further includes inputting the multiple training images into a convolutional neural network to label each of the multiple nerves and each of the multiple layers of tissue in each of the multiple training images, and validating each corresponding label assigned to each of the multiple nerves and each of the multiple layers of tissue in each of the multiple training images by the convolutional neural network with the corresponding training label corresponding to each nerve and each layer of the multiple labels. Assigning training labels to each of the multiple nerves and each of the multiple layers of tissue in each of the multiple training images may include applying one or more filters to each of the multiple training images to identify the nerves in each of the multiple training images. Training the model may further include inputting a mirrored or rotated version of each of the multiple training images into the convolutional neural network to label each of the multiple nerves and each of the multiple layers of tissue in the mirrored or rotated version of each of the multiple training images.

[0022] In some embodiments, a set of images among the multiple images includes copies of a single image, and each copy includes a different contrast, brightness, cropping, distortion, or rotation from the single image.

[0023] In some embodiments, each image includes a z-stack image. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] The drawings are incorporated herein and constitute a part of the specification, illustrate the present disclosure, and, together with the description, further explain the principles of the present disclosure and enable those skilled in the relevant art to make and use the present disclosure.

[0025] Figure 1 is a block diagram of a system for using image analysis to determine a neurological condition according to some embodiments.

[0026] Figure 2 is a block diagram showing preprocessing of an input image before the input image is transmitted to a model according to some embodiments.

[0027] Figure 3 is a block diagram showing generation of various versions of an image according to some embodiments.

[0028] Figure 4 is a block diagram showing generation of various versions of an image according to some embodiments.

[0029] Figure 5 is a block diagram showing pre-assigned labels according to some embodiments.

[0030] Figure 6 is a block diagram showing identification of nerves in an image of tissue according to some embodiments.

[0031] Figure 7 is a block diagram showing a neural network according to some embodiments.

[0032] Figure 8 is a block diagram of a feature map generated by a neural network according to some embodiments.

[0033] Figures 9 to 12 Shows an example image of tissue labeled by a pathologist and a trained model according to some embodiments.

[0034] Figure 13 Shows an example image of skin tissue having nerve fiber crossings from the dermis to the epidermis determined using intraepidermal nerve fiber (IENF) counting rules according to some embodiments.

[0035] Figure 14 Shows an example image labeled by a training model for determining nerve density and nerve fiber crossings according to some embodiments.

[0036] Figure 15An example chart showing test data indicative of nerve density associated with mice undergoing various treatments, according to some embodiments.

[0037] Figure 16 Shows test correlations of nerve conduction and histopathology, according to some embodiments.

[0038] Figure 17 Is a flowchart showing a process for using image analysis to determine a nerve condition, according to some embodiments.

[0039] Figure 18 Is a block diagram of an example component of a device, according to one embodiment.

[0040] Figure 19 Shows a block diagram of example components of a computer system, according to aspects of the present disclosure.

[0041] The figure in which an element first appears is generally indicated by the leftmost one or more digits in the corresponding reference numeral. In the figures, the same reference numerals may indicate identical or functionally similar elements. Detailed Description

[0042] Systems, devices, apparatuses, methods, and / or computer program product embodiments and / or combinations and sub - combinations thereof for using image analysis to identify nerve conditions are provided herein.

[0043] As noted above, traditional methods for identifying nerve conditions can be costly, invasive, time - consuming, and error - prone. Specifically, manual evaluation of images showing tissue by a trained pathologist can be a time - consuming process. Human vision / interpretation is limited, can be inconsistent, and can contain significant bias. For example, if a given treatment is being used to treat a patient's neuropathy, a pathologist may interpret an image of the patient's tissue as indicating that the treatment is working. Additionally, different pathologists may use different methods to interpret images, and thus, the interpretations can vary. Further, many pathologists only consider a subset of the nerves or nerve types in a given image. All of these can lead to inconsistent diagnoses and treatments.

[0044] The embodiments described herein address these problems by using image analysis to identify a patient's nerve condition. In a given embodiment, a processor receives an image of a patient's skin tissue. The image shows a plurality of nerves and a plurality of layers of the patient's skin tissue. Using a trained model, the processor labels each nerve of the plurality of nerves and each layer of the plurality of layers of the patient's skin tissue in the image. The trained model has been trained to identify differences between the nerves and the plurality of layers of the skin tissue. Additionally, the processor determines the amount of the layer of nerve area relative to the skin tissue area based on the labels assigned to the plurality of nerves and the plurality of layers of the patient's skin tissue. The processor determines the nerve condition based on the amount of the layer of nerve area relative to the skin tissue area.

[0045] The embodiments described herein use a trained model configured to implement a neural network to determine a patient's nerve condition. Specifically, the neural network implemented by the trained model provides fast, accurate, and consistent identification of the nerves and layers of a patient's skin tissue. This may involve determining the patient's nerve density or determining the number of nerve fiber crossings between the layers of the patient's tissue. The nerve condition may indicate peripheral neuropathy. Additionally, the nerve condition may indicate whether a drug treatment induces neuropathy, or alternatively whether an ongoing neuropathy treatment is effective.

[0046] Figure 1 is a block diagram of a system for using image analysis to identify a nerve condition. The system may include a server 100, a client device 110, and a database 120. The devices of the system may be connected via a network. For example, the devices of the system may be connected via a wired connection, a wireless connection, or a combination of a wired connection and a wireless connection. In one example embodiment, one or more portions of the network may be an ad hoc network, an intranet, an extranet, a virtual private network (VPN), a local area network (LAN), a wireless LAN (WLAN), a wide area network (WAN), a wireless wide area network (WWAN), a metropolitan area network (MAN), a portion of the Internet, a portion of the public switched telephone network (PSTN), a cellular telephone network, a wireless network, a WiFi network, a WiMax network, any other type of network, or a combination of two or more such networks. Alternatively, the server 100, the client device 110, and the database 120 may be located on a single physical machine or virtual machine.

[0047] In some embodiments, the server 100 and the database 120 may reside in a cloud computing environment. In other embodiments, the server 100 may reside in a cloud computing environment while the database 120 resides outside the cloud computing environment. Additionally, in other embodiments, the server 100 may reside outside the cloud computing environment while the database 120 resides in the cloud computing environment.

[0048] Server 100 may include an analysis engine 102 and a model 104. The analysis engine 102 may train the model 104 to label nerves and tissue layers in images of skin tissue. The layers of the skin tissue may include the dermis, epidermis, and cornea. The model 104 may implement deep learning algorithms such as neural networks, convolutional neural networks (CNNs), artificial neural networks (ANNs), recurrent neural networks (RNNs), deep reinforcement learning, etc. Additionally, the analysis engine 102 may use the output of the model 104 to determine the nerve condition of a given patient.

[0049] The database 120 may be one or more data storage devices configured to store structured and unstructured data. The database 120 may store files for training the model 104. Additionally, the database 120 may store data associated with the nerves and tissue layers labeled by the model 104. In one example, the label may be a mask overlaid on the tissue image. The data may be an image of the tissue including different masks for the nerves and tissue layers. The database 120 may also store data associated with the nerve condition of the patient determined by the analysis engine 102. The data may be, for example, the nerve density within the epidermis. In particular, the nerve density may be determined based on the amount of nerve area / coverage over the tissue area identified in the image of the tissue. Alternatively or additionally, the data may be the number of nerve fibers across multiple layers (e.g., from the dermis to the epidermis).

[0050] The client device 110 may be configured to connect to the server 100 and the database 120. To this end, the client device 110 may transmit a request to the server 100 to train the model 104. The request may include the location of the file for training the model 104 in the database 120. Alternatively, the client device 110 may retrieve the file and include the file in the request. The file may be an image of the tissue, and may be received in one or more batches. As a non-limiting example, the image may be an image of mouse skin tissue. In this example, the file includes two batches. The first batch includes 81 images of 12 mice in two treatment groups. The second batch includes 99 images of 19 mice in four treatment groups. The treatment groups in this example indicate the current nerve condition of each mouse.

[0051] The server 100 may receive the request from the client device 110. The analysis engine 102 may retrieve the file indicated in the request from the database 120. As described above, the file may be an image of the tissue (e.g., mouse skin tissue). The file may be in any of various formats, such as PDF, JPEG, TIF, GIF, etc.

[0052] To increase the amount of available training data, the analysis engine 102 can generate multiple copy versions of each image retrieved from the database 120. For example, the analysis engine 102 can generate rotated or mirrored versions of the corresponding images. Additionally, the analysis engine 102 can generate one or more versions of the corresponding images with different contrast, brightness, cropping, distortion, edge, shape, and / or texture filters.

[0053] The analysis engine 102 can instruct the model 104 to identify and label the nerves and tissue layers in each image (including the different versions of each image). The model 104 can implement a deep learning algorithm to classify and label each image pixel as part of a tissue layer, nerve, or background. As a non-limiting example, the deep learning algorithm can be a CNN.

[0054] The training of the CNN can be divided into two phases: a forward phase and a backward phase. The forward phase includes convolutional layers, pooling layers, and fully connected layers. The analysis engine 102 can instruct the CNN to classify and label each input image pixel as part of a layer, nerve, or background to train the CNN. The input images can be from a training file, including different versions of each image.

[0055] The convolutional layer can apply filters to the input image to generate a feature map. In particular, in the convolutional layer, the CNN can perform feature extraction on the input image. Features can include parts of the input image. For example, a feature can be a different edge or shape of the input image. The CNN can extract different types of features to generate different types of feature maps. For example, the CNN can apply a digital array (e.g., a kernel) across different parts of the input image. The kernel can also be referred to as a filter. As described above, different types of filters can be applied to the input image to generate different feature maps. For example, a filter for identifying shapes in the input image can be different from a filter for edge detection. Thus, different kernels can be applied to identify shapes in the input image compared to edge detection. Each kernel can include a different digital array. The values of the filter or kernel can be randomly assigned and optimized over time (e.g., using a gradient descent algorithm). The kernel can be applied as a sliding window across different parts of the input image. The kernel can be added to a given part of the input image to generate an output value. The output value can be included in the feature map. The feature map can include output values from different kernels applied to each part of the input image. The generated feature map can be a two-dimensional array.

[0056] The pooling layer can generate a simplified feature map. In particular, in the pooling layer, the CNN can reduce the dimension of each feature map generated in the convolutional layer. The CNN can extract parts of a given feature map and discard the rest. The pooled image retains the important features. For example, a feature map may include active regions and non-active regions. The active regions may include the detected features, while the non-active regions may indicate that this part of the image does not include features. Pooling can remove the non-activated regions. In this way, the size of the image is reduced. The CNN can use max pooling or average pooling in the pooling layer to perform these operations. Max pooling keeps the higher values of parts of the feature map while discarding the remaining values. Average pooling keeps the average value of different parts of the feature map. Thus, the CNN can generate a simplified feature map for each feature map in the feature maps generated in the convolutional layer.

[0057] The CNN may include additional convolutional layers. In the additional convolutional layers, the CNN can generate additional feature maps based on the simplified feature maps generated in the pooling layer. In addition, the CNN may include additional pooling layers. In the additional pooling layers, the CNN can generate further simplified feature maps based on the feature maps generated in the additional convolutional layers.

[0058] The convolutional layer can also apply the rectified linear unit (ReLU) function to the input image. The ReLU function is applied to the input image to remove linearity from the input image. For example, the ReLU function can remove all black elements from the input image and only retain gray and white. This makes the color changes in the input image more abrupt, thus eliminating the linearity in the input image.

[0059] The convolutional layer and the pooling layer can be used for feature learning. Feature learning allows the CNN to identify the desired features in the input image and thus accurately classify the input image. Therefore, by optimizing the convolutional layer and the pooling layer, the CNN can apply the correct filters to the input image to extract the necessary features required for classifying the input image.

[0060] Then, the fully connected layer can use weights and biases to classify the features of the image to generate an output. The CNN can classify each pixel in the input image and label it as part of a layer of tissue, nerve, or background. The CNN can output the input image with the labeled layers, nerves, and background. The label can be a mask applied to the image, indicating whether the corresponding pixel is part of a specific layer, nerve, or tissue. A mask will be described in more detail with reference to Figure 5 describe the mask in more detail.

[0061] Specifically, in the fully connected layer, the CNN can flatten the simplified feature maps generated in the pooling layer into a one-dimensional array (or vector). The fully connected layer is a neural network. The CNN can perform a linear transformation on the one-dimensional array. The CNN can perform the linear transformation by applying weights and biases to the one-dimensional array. Initially, the weights and biases are randomly initialized and can be optimized over time. The CNN can perform a non-linear transformation such as an activation layer function (e.g., softmax or sigmoid) to classify and label each input image pixel as part of a layer, nerve, or background of the tissue.

[0062] In the backward phase, the CNN can use backpropagation to determine whether the CNN can correctly label each pixel of the input image. Backpropagation includes optimizing the input parameters so that the CNN can classify the document more accurately. The input parameters can include values for kernels, weights, biases, etc. Gradient descent can be used to optimize the parameters. Specifically, gradient descent can be used to optimize the classification and labeling of each pixel in the input image by the CNN.

[0063] Gradient descent is an iterative process for optimizing the CNN. Gradient descent can update the parameters of the CNN and cause the CNN to classify and label each pixel in the input image based on the updated parameters and verify the classification and labeling of each pixel in the input image.

[0064] To this end, the CNN can use backpropagation to verify its classification of the input image. Specifically, a subject matter expert (e.g., a pathologist) can determine whether the model 104 has correctly labeled the layers, nerves, and background. The subject matter expert can provide their feedback on the accuracy of the labeling from the client device 110 to the analysis engine 102. Alternatively or additionally, the analysis engine 102 can use the metadata associated with each image to verify the labeling of the layers, nerves, and background. For example, the metadata can include pre-assigned labels for the layers, nerves, and background assigned to each image. A pathologist or a different third party can generate the pre-assigned labels for each image. The pre-assigned labels can be masks assigned to different layers, nerves, and backgrounds.

[0065] The CNN can compare the classification and labeling assigned by the CNN to each pixel in the input image with the information included in the metadata or the feedback provided by the subject matter expert. Based on this comparison, the CNN can use the gradient descent algorithm to update the values of the filters, weights, and biases and re-perform the forward phase on the input image.

[0066] The analysis engine 102 may instruct the CNN to label and classify each pixel of each image in the training file, including different versions of each image. The CNN iteratively optimizes its parameters, reclassifying and labeling each pixel of each image in the training file during the forward phase, and validating the classification and labeling of each pixel during the backward phase until a desired accuracy threshold is reached.

[0067] Once the CNN reaches the desired accuracy threshold, the CNN can be considered fully trained. To this end, the model 104 can be considered fully trained.

[0068] The client device 110 may transmit a request to classify and label a new image of a patient's tissue to the server 100. The request may be to determine the patient's nerve condition based on an image of the patient's tissue. The request may include the new image. Alternatively, the request may include the location of the new image in the database 120. The new image may not be part of the images used to train the model 104.

[0069] The server 100 may receive the request. If the request includes the location of the new image in the database 120, the analysis engine 102 may retrieve the new image from the database 120. The analysis engine 102 may instruct the model 104 to classify and label each pixel of the new image as part of a particular layer, nerve, or background. The model 104 may implement a deep learning algorithm (e.g., CNN) to classify and label each pixel of the new image. The model 104 may output an image including labels for the different layers, nerves, and background of the tissue in the image.

[0070] The analysis engine 102 may use the labels to determine the nerve density within the epidermis. The nerve density may indicate nerve damage. In this regard, the analysis engine 102 may determine the patient's nerve density based on the image output by the model 104, according to the amount / percentage of the patient's nerve area to the total area of the patient's epidermis. Specifically, the analysis engine 102 may identify the nerves labeled in the image output by the model 104 and determine how much tissue area the nerves cover. The analysis engine 102 may generate a determination regarding the patient's nerve condition based on the patient's nerve density. For example, the analysis engine 102 may diagnose the patient with chemotherapy-induced peripheral neuropathy based on the nerve density being below a threshold amount. Alternatively, the analysis engine 102 may diagnose the patient with chemotherapy-induced peripheral neuropathy based on the nerve density being below a second (possibly earlier or pre-treatment) determination of the nerve density by the analysis engine 102 (e.g., by a threshold or percentage). Additionally, the analysis engine 102 may determine the treatment efficacy for peripheral neuropathy by comparing a first (e.g., pre-treatment) determination and a second (e.g., post-treatment) determination of the nerve density by the analysis engine 102. Furthermore, a pathologist may determine peripheral neuropathy by manually analyzing or comparing the nerve density determined by the analysis engine 102.

[0071] Alternatively or additionally, the analysis engine 102 can also use the markings in the image output by the model 104 to determine the number of nerve fiber crossings from the dermis to the epidermis. A smaller number of nerve fibers crossing from the dermis and the epidermis can indicate nerve damage. In this way, the analysis engine 102 can identify the epidermis in the patient tissue (as marked in the image output by the model 104) and the dermis in the patient tissue (as marked in the image output by the model 104). The analysis engine 102 can generate a boundary between the epidermis and the dermis in the image output by the model 104. The analysis engine 102 can determine the number of nerve fiber crossings based on pixel groupings that identify nerves marked as extending from the dermis to the epidermis. This determination of nerve fiber crossings by the analysis engine 102 can be closely related to the counting of nerve fiber crossings according to the IENF counting rules. The IENF will be described with reference to Figure 13 for description.

[0072] The analysis engine 102 can generate a determination regarding the nerve condition of the patient based on the number of nerve fiber crossings. For example, the analysis engine 102 can diagnose that the patient has chemotherapy-induced peripheral neuropathy based on the number of nerve fiber crossings being below a threshold amount. In one example, the number of nerve crossings can include a normalized number of nerve crossings determined by dividing the determined number of nerve crossings by a certain pixel length of the epidermis / dermis boundary. Alternatively, the analysis engine 102 can determine whether the treatment being administered for the diagnosed neuropathy in the patient is effective, or whether a drug has induced the development of neuropathy.

[0073] Figure 2 is a block diagram showing preprocessing of an input image before the input image is transmitted to the model 104. The preprocessing will be described with reference to Figure 1 for description. Figure 2 of it.

[0074] In some embodiments, the input image can be an image of skin tissue. The input image can be part of a training file or a new image of patient tissue. An image of skin tissue can be obtained using a skin biopsy. Images of the tissue can be captured at different focal lengths. The focal plane is the distance between the lens of the camera capturing the image and the focal point.

[0075] The analysis engine 102 can perform preprocessing on the input image before providing the input image to the model 104 for classification and labeling. The analysis engine 102 can receive or retrieve images 200 and 202 of the same tissue at different focal planes. The analysis engine 102 can compile a z-stack image 204 by combining images 200 and 202. To this end, the z-stack image 204 can be a composite image with a greater depth than images 200 and 202. The z-stack image 204 can be the input image.

[0076] As shown by element 206, 2 to n images can be captured at respective focal planes. The analysis engine 102 can combine the 2 to n images to generate a z-stack image 208. The z-stack image 208 can be a composite of the 2 to n images.

[0077] Figure 3 is a block diagram showing various versions of generating images according to some embodiments. Reference will be made to Figure 1 to describe Figure 3 .

[0078] In some embodiments, the analysis engine 102 can generate different versions of each image in the training file by changing the contrast, brightness, cropping, distortion, rotation, etc. of each image. For example, image 300 can be included in the training file. The analysis engine 102 can change one or more of the contrast, brightness, cropping, distortion, rotation, etc. of image 300 to generate image 302. Similarly, image 304 can be included in the training file. The analysis engine 102 can change one or more of the contrast, brightness, cropping, distortion, rotation, etc. of image 304 to generate image 306.

[0079] Image 302 and image 306 can be included in the training file together with image 300 and image 304. This allows an increase in the number of training files, thereby improving the model 104. Thus, each of images 300 to 306 can be used to train the model 104, as described above with reference to Figure 1 . This allows the model 104 to be trained to label and classify each image pixel with different contrast, brightness, cropping, distortion, rotation, etc.

[0080] Figure 4 is a block diagram showing various versions of generating images according to some embodiments. Reference will be made to Figure 1 to describe Figure 4 .

[0081] In some embodiments, the analysis engine 102 can generate different versions of each image in the training file by rotating and mirroring each initial image. For example, image 400 can be included in the training file as an initial image. The analysis engine 102 can flip image 400 to generate image 402. The analysis engine 102 can mirror image 400 to generate image 404. The analysis engine 102 can mirror image 402 to generate image 406. The analysis engine 102 can rotate image 404 to generate image 408. The analysis engine 102 can flip image 408 to generate image 410. The analysis engine 102 can mirror image 408 to generate image 412. The analysis engine 102 can mirror image 410 to generate image 414.

[0082] Images 400 through 414 can be included in the training file. This allows for an increase in the number of training files. As such, images 400 through 414 can be used to train model 104, as described above with reference to Figure 1 what has been described. This allows model 104 to be trained to label and classify each pixel in each image, regardless of the position of the tissue layers, nerves, and background in the image.

[0083] Figure 5 An example image showing exemplary pre-assigned labels according to some embodiments. Reference will be made to Figure 1 for the description of Figure 5 .

[0084] In some embodiments, a subject matter expert can use a software application to pre-assign labels to the layers, nerves, and background in an image of tissue. The subject matter expert can pre-assign labels for each image in the training file. The pre-assigned labels can be part of the metadata of the training file for validating the labeling and classification of model 104.

[0085] For example, image 500 can be an image of skin tissue in the training file. The subject matter expert can use a software application to assign a first label to the dermis layer 504 of the tissue in the image, a second label to the epidermis layer 506 of the tissue in the image, a third label to the corneal layer 508 of the tissue in the image, a fourth label to the nerves 510 of the tissue in the image, and a fifth label to the background 514 of the image.

[0086] Image 502 includes markings. The markings can be masks overlaid on the layers, nerves, and background. The first marking can be a blue mask, the second marking can be a purple mask, the third marking can be a yellow mask, the fourth marking can be a green mask, and the fifth marking can be a white mask. In this way, the blue mask can be overlaid on the pixels corresponding to the dermis layer, the purple mask can be overlaid on the pixels corresponding to the epidermis layer, the yellow mask can be overlaid on the pixels corresponding to the corneal layer, the green mask can be overlaid on the pixels corresponding to the nerves, and the white mask can be overlaid on the pixels corresponding to the background. However, this is one implementation, and different colors can be used for the various masks as needed.

[0087] Figure 6 Illustrates the identification of nerves in an image of tissue according to some embodiments.

[0088] In some embodiments, pre-assigning markings to images in a training file can include identifying nerves in an image of tissue. Specifically, the application can be executed by the client device 110. Alternatively, the application can reside on the server 100, and the client device 110 can access the application. The application can be used to pre-assign markings in the training file.

[0089] The application can initially automatically detect the nerves in image 600 based on the brightness and contrast of image 600. For example, the nerves in image 600 may be darker compared to the background. Thus, the application can detect the darker regions of image 600 based on the brightness and contrast of the pixels in image 600. The darker pixels can be marked as nerves. Such detection can be based on a pre-trained deep learning model trained on previously marked images, or can be based on a given pixel having a threshold percentage of brightness and / or contrast values that are different from at least one adjacent pixel or subset of adjacent pixels, or a given pixel having a brightness value higher or lower than a given threshold.

[0090] For example, a value corresponding to one or a combination of intensity, brightness, or contrast can be assigned to each pixel in the image. Each pixel having a value equal to or higher than (or conversely, lower than) a specific threshold can be identified as corresponding to a nerve. Each pixel having a value lower than (or conversely, higher than) a specific threshold can be identified as corresponding to tissue. The threshold can be a pre-determined absolute threshold. Alternatively, the threshold can be based on the average or mean value of all the pixels in the image. In some aspects, background pixels or other pixels outside the boundary of the tissue sample can be assigned a "null" value.

[0091] Once a value has been assigned to all the pixels in an image, the nerve density can be determined by dividing the number of pixels corresponding to nerves by the number of pixels corresponding to tissue. Alternatively, the ratio of the number of nerve pixels or tissue pixels to the total number of pixels in the image can be used.

[0092] In some aspects, nerves can be identified by applying one or more filters to an image of tissue. For example, nerves can be identified in image 600 by applying one or more filters corresponding to one or more parameters. The parameters can include brightness, contrast, size, color, edge, shape, and texture filters. A label (e.g., a fourth label or a green mask) can be pre-assigned to the identified nerves and then used for further analysis. In some aspects, the filters are manipulable such that they can be optimized for a given image.

[0093] Figure 7 is a block diagram showing a neural network according to some embodiments. Reference will be made to Figure 1 to describe Figure 7 .

[0094] As described above, model 104 can implement a deep learning algorithm. The deep learning algorithm can be a neural network (e.g., a convolutional neural network) 700. The neural network 700 can include an input layer, a plurality of hidden layers, and an output layer.

[0095] If the neural network 700 is a CNN, the input layer can include a pixel matrix of the input image. The input image can be part of a training file or a new image of a patient's tissue. The pixel matrix of the input image can be transmitted to hidden layer 1.

[0096] One hidden layer can be a convolutional layer. As described above, the convolutional layer can apply filters to the pixel matrix of the input image to generate a feature map. The feature map can be sent to another hidden layer, which can be a pooling layer. As described above, the pooling layer can generate a simplified feature map. The neural network 700 can include additional hidden layers (e.g., additional convolutional layers or pooling layers) to generate additional feature maps and simplified feature maps. Although two hidden layers are explicitly shown in Figure 7 , it should be understood that the neural network 700 can include any number of hidden layers (such as 2, 3, 4,... n hidden layers), where each hidden layer can have a different shape and / or size.

[0097] The simplified feature map can be transmitted to the output layer. The output layer can be a fully connected layer. As described above, in a fully connected layer, the CNN can flatten the simplified feature map generated in the pooling layer into a one-dimensional array (or vector). The fully connected layer can use weights and biases to classify the features of the image to generate an output. The output can be the classified and labeled input image. That is, the output can be an image including the label or mask predicted / classified by the neural network 700.

[0098] If the model 104 is being trained and once the output is generated, the neural network 700 can verify the classified and labeled input image based on the metadata of the input image. The metadata can include pre-assigned labels. The neural network 700 can identify any errors in the classification and the labels assigned to the input image, refine the parameters (e.g., biases and weights), and attempt to re-classify and label the input image. The neural network 700 can iteratively classify and label each image in the training file.

[0099] Figure 8 is a diagram of a feature map generated by a neural network according to some embodiments. Reference will be made to Figure 1 and Figure 7 to describe Figure 8 .

[0100] In some embodiments, the hidden layer of the neural network 700 can generate a feature map 804 representing the pixel matrix 800 of the input image. For example, a digital array 802 can be applied across different parts of the pixel matrix 800 to generate the feature map 804.

[0101] The neural network 700 can simplify the feature map. Additionally, the neural network 700 can generate additional feature maps based on the simplified feature map and further simplify the additional feature maps until a fully connected layer is formed. As described above, the fully connected layer can generate an output of the classified and labeled input image.

[0102] As a non-limiting example, the input image 806 can be transformed into a pixel matrix, a feature map, and a simplified feature map as shown in the transformation 808. The fully connected layer can output the classified and labeled image 810.

[0103] Figures 9 to 12 shows an image of tissue labeled by a pathologist and the model 104 according to some embodiments. Specifically, Figures 9 to 12 shows the ability of the model 104 to automatically label images that would otherwise require a great deal of effort and time from a pathologist (e.g., in the case of nerve fiber crossings) or that are practically impossible for a pathologist (e.g., in the case of nerve fiber density). Additionally, the model 104 can improve the accuracy and consistency of the labels provided by the model 104 compared to manually labeling images by a single or multiple pathologists.

[0104] For example, the image 900 can be an image of a patient's tissue. The patient's tissue can include a dermis layer, an epidermis layer, a cornea layer, and nerves. The image 900 can also include a background. A pathologist can use software to label the image 900 to generate the image 902. The image 902 can have labels assigned to the dermis layer, the epidermis layer, the cornea layer, the nerves, and the background.

[0105] In addition, the image 900 can be labeled by the model 104. The model 104 can label each pixel of the image 900 as part of the dermis layer, the epidermis layer, the cornea layer, the nerves, or the background to generate the image 904. As shown by element 906, the model 104 is capable of automatically performing the labeling function routinely performed by a pathologist and can identify the size and shape of the cornea layer more accurately than a pathologist.

[0106] Regarding Figure 10 ,the image 1000 can be an image of a patient's tissue. The patient's tissue can include a dermis layer, an epidermis layer, a cornea layer, and nerves. The image 1000 can also include a background. A pathologist can use software to label the image 1000 to generate the image 1002. The image 1002 can have labels assigned to the dermis layer, the epidermis layer, the cornea layer, the nerves, and the background.

[0107] In addition, the image 1000 can be labeled by the model 104. The model 104 can label each pixel of the image 1000 as part of the dermis layer, the epidermis layer, the cornea layer, the nerves, or the background to generate the image 1004. The model 104 can identify the size and shape of the cornea layer more accurately than a pathologist.

[0108] Regarding Figure 11 ,the image 1100 can be an image of a patient's tissue. The patient's tissue can include a dermis layer, an epidermis layer, a cornea layer, and nerves. The image 1100 can also include a background. A pathologist can use software to label the image 1100 to generate the image 1102. The image 1102 can have labels assigned to the dermis layer, the epidermis layer, the cornea layer, the nerves, and the background.

[0109] In addition, the image 1100 can be labeled by the model 104. The model 104 can label each pixel of the image 1100 as part of the dermis layer, the epidermis layer, the cornea layer, the nerves, or the background to generate the image 1104. The model 104 is more capable of identifying the size and shape of the dermis layer with higher accuracy than a pathologist.

[0110] Regarding Figure 12, Image 1200 can be an image of a patient's tissue. The patient's tissue can include the dermis layer, epidermis layer, corneal layer, and nerves. Image 1200 can also include a background. A pathologist can use software to label Image 1200 to generate Image 1202. Image 1202 can have labels assigned to the dermis layer, epidermis layer, corneal layer, nerves, and background.

[0111] In addition, Image 1200 can be labeled by Model 104. Model 104 can label each pixel of Image 1200 as part of the dermis layer, epidermis layer, corneal layer, nerves, or background to generate Image 1204. Model 104 is more capable than a pathologist of identifying the size and shape of the epidermis layer and dermis layer with higher accuracy.

[0112] Figure 13 An image of skin tissue is shown, which shows the crossing of nerve fibers from the dermis layer to the epidermis layer. In Figure 13 the IENF counting rule is used to determine the number of nerve fiber crossings.

[0113] The IENF counting rule is a standard counting rule for counting nerve fibers. The IENF counting rule counts the nerves that cross from the dermis layer to the epidermis layer. The total IENF count can be normalized by dividing the IENF count by the boundary length between the dermis layer and the epidermis layer. The normalized IENF count can be used to determine the nerve condition.

[0114] Image 1300 shows an image of skin tissue including the epidermis layer, dermis layer, and nerves. Image 1302 shows a diagram of the nerves of the skin tissue shown in Image 1300 and the boundary (BM) between the epidermis layer and the dermis layer. Image 1300 includes nerve fibers 1304 to 1310.

[0115] In Image 1302, the nerves in the dermis layer are labeled as nerve fibers a to i. Based on the IENF counting rule, if a nerve fiber originates in the dermis layer and crosses the epidermis layer, it is counted as a nerve fiber crossing. If a nerve fiber branches into multiple different fibers in the dermis layer or on the boundary (BM) before crossing the epidermis layer, it is counted as two nerve fiber crossings. If a nerve fiber branches into multiple nerve fibers in the epidermis layer after the boundary (BM), it is counted as a single nerve fiber crossing. If a nerve fiber in the epidermis layer has a disconnected part from the original nerve fiber in the dermis layer, it is not counted as a nerve fiber crossing. If a nerve fiber does not cross the epidermis layer, it is not counted as a nerve fiber crossing.

[0116] In image 1302, nerve fiber a branches into two nerve fibers after passing through the epidermal layer. Thus, nerve fiber a is counted as one nerve fiber crossing. Nerve fiber b branches into two nerve fibers after the boundary (BM) and within the epidermal layer. Thus, nerve fiber b is counted as one nerve fiber crossing. Nerve fiber c branches into two nerve fibers before passing through the epidermal layer (at the boundary (BM)). Thus, nerve fiber c is counted as two nerve fiber crossings. Nerve fiber d branches into two nerve fibers before passing through the epidermal layer (within the dermal layer). Thus, nerve fiber d is counted as two nerve fiber crossings. Nerve fiber e is a single nerve fiber that passes through the dermal layer into the epidermal layer. Thus, nerve fiber e is counted as one nerve fiber crossing. Nerve fiber f does not pass through the epidermal layer. Thus, nerve fiber f is not counted as a nerve fiber crossing. Nerve fiber g is a single nerve fiber that passes through the dermal layer into the epidermal layer. Thus, nerve fiber g is counted as one nerve fiber crossing. Nerve fiber h branches into two nerve fibers before passing through the epidermal layer (at the boundary (BM)). Thus, nerve fiber h is counted as two nerve fiber crossings. The branch of nerve fiber i disconnects before entering the epidermal layer. Thus, nerve fiber i is not counted as a nerve fiber crossing. Based on the IENF counting rule, image 1302 indicates 10 nerve fiber crossings.

[0117] Figure 14 An image showing determination of nerve density and nerve fiber crossings marked by model 104 according to some embodiments. Reference will be made to Figure 1 for description Figure 14 .

[0118] Model 104 can label each pixel of an image of patient tissue to generate an image, such as image 1400. Analysis engine 102 can use the labels to determine nerve density. For example, analysis engine 102 can determine the number of pixels marked as nerves in image 1400. Analysis engine 102 can also determine the number of pixels marked as a type of tissue in image 1400. Analysis engine 102 can ignore pixels marked as belonging to the background.

[0119] Analysis engine 102 can calculate nerve density based on the ratio between the number of pixels marked as nerves in the epidermis and the number of pixels marked as tissue in the epidermis. For example, analysis engine 102 can divide the number of pixels marked as nerves by the number of pixels marked as tissue to determine nerve density. In Figure 14 a specific example, analysis engine 102 can determine that 8.76% of the tissue area constituting the epidermis determined by analysis engine 102 is covered by nerves in image 1400. Analysis engine 102 can use nerve density to determine the patient's nerve condition. The lower the nerve density, the greater the likelihood that the patient is diagnosed with peripheral neuropathy.

[0120] Additionally or alternatively, the analysis engine 102 can assess the likelihood of peripheral neuropathy by counting the nerve fiber crossings from the dermis to the epidermis, as shown in image 1402. For example, the analysis engine 102 can define the boundary 1404 between the epidermal layer and the dermal layer based on image 1402, as marked by the model 104. The analysis engine 102 can identify a set of pixels of the nerve that are marked as crossing the boundary 1404 between the epidermal layer and the dermal layer. The number of identified crossings can be normalized by dividing the number of crossings identified along a specific boundary length by the number of pixels within that specific boundary length.

[0121] As a non-limiting example, the analysis engine 102 can determine that the length of the crossing boundary 1404 includes 4401 pixels. Although shown in orange in Figure 14 for visibility, the crossing boundary 1404 will include a line with a width of a single pixel. Additionally, the analysis engine 102 can determine that the image 1402 can include 36 crossings along the length of the crossing boundary 1404. The number of crossings can be normalized by dividing the number of crossings (e.g., 36) by the number of pixels in the length of the crossing boundary 1404 (4401 in this example). The lower the number of nerve fiber crossings, the more likely the patient is to be diagnosed with peripheral neuropathy.

[0122] Figure 15 A graph including exemplary test data showing the epidermal nerve density of mice treated with paclitaxel determined using the analysis engine 102 and the model 104, according to some embodiments. Specifically, the graph 1500 indicates the differences in epidermal nerve density between mice treated with various treatments as determined using the analysis engine 102 and the model 104. For example, the analysis engine 102 can use such information to determine the presence or development of peripheral neuropathy caused by a treatment regimen such as cancer treatment. This information can also be used, for example, by the analysis engine 102 to determine the effectiveness of various neuropathy treatments such as topical injections of corticosteroids, lidocaine, or botulinum toxin, using the data reflected in the graph 1500. Although the example of Figure 15 specifically refers to paclitaxel, it is contemplated that the nerve analysis system and method of the present application, particularly the system described with reference to Figure 1 can be used to measure the effects of a variety of drugs that may induce neuropathy, such as chemotherapy, cardiac and blood pressure treatment drugs (e.g., amiodarone), infectious disease drugs (e.g., chloroquine), autoimmune disease drugs (e.g., infliximab), anti-epileptic drugs (e.g., phenytoin), and so on.

[0123] In this example test, the analysis engine 102 uses an image of mouse tissue labeled as model 104 to determine the nerve density of mice administered with a neuropathy-inducing drug. The mice are grouped based on the treatment provided to each mouse. The treatments include: a control group injected with saline, and a test group injected with a neuropathy-inducing drug, in this case, paclitaxel (PCTX) at 25 mg / kg. PCTX 25 mg / kg indicates the treatment of injecting paclitaxel intravenously into the mice. Paclitaxel is an anti-tumor drug commonly used to treat cancer, and its side effect is that it can induce neuropathy. As shown, two independent batches of the control group and the PCTX group were analyzed.

[0124] In this example test, the analysis engine 102 determines the nerve density by dividing the number of nerve pixels by the number of epidermal layer pixels in the image of the mouse tissue.

[0125] Chart 1500 indicates the ratio of nerve area to epidermal area (nerve density) of mice undergoing neuropathy treatment, as determined by the analysis engine using an image of mouse tissue labeled as model 104. In these two batches, the mice are grouped based on the treatment provided to each mouse. The treatments include the control group and PCTX 25 mg / kg treatment. In this example test, the analysis engine 102 determines the nerve density by dividing the number of nerve pixels by the number of epidermal layer pixels in the image of the mouse tissue.

[0126] Chart 1500 includes two batches, Batch 1 and Batch 2. Each batch includes the corresponding control group and PCTX 25 mg / kg IV treatment.

[0127] Chart 1502 indicates the nerve density of mice undergoing neuropathy treatment in Batch 1 and Batch 2 determined by an automated system (such as the aforementioned analysis engine 102). Chart 1502 reflects the data in Chart 1500 in a bar chart format. The comparison between Chart 1500 and Chart 1502 shows that an automated system trained to determine the nerve fiber ratio can adequately distinguish the efficacy of different treatments, even if their differences are small. This is because the automated analysis results in a tightly packed distribution of values, that is, the within-group variance is small.

[0128] Chart 1504 indicates the epidermal nerve density of the mice in Batch 1 determined by a pathologist. From the comparison, it can be seen that the comparison output of the automated system shown in Chart 1502 is confirmed by the manual evaluation of the pathologist shown in Chart 1504. Although the comparison results between the control group and paclitaxel in Batch 1 are similar in both charts, the nerve fiber ratio reported in the manual evaluation in Chart 1504 is higher within a larger distribution range compared to the automated system in Chart 1502. This may be due to the bias of the pathologist. Nevertheless, the comparison results support the following findings: the automated system can accurately identify and distinguish treatment outcomes and has lower within-group variability compared to manual evaluation.

[0129] Figure 16 A significant correlation between nerve conduction and determination of nerve fiber density according to the above test model is shown. Nerve response amplitude and latency are conventional methods for detecting nerve damage. For example, stimulation and recording electrodes can be applied to the skin, and the nerve response amplitude and latency can be determined based on calibrated electrical stimulation. Therefore, to further validate the method and system for determining nerve fiber density as described herein, the Figure 15 nerve response amplitudes and latencies of certain members of the control groups and the PCTX groups of Batch 1 and Batch 2 cited therein were analyzed. Chart 1600 shows a highly positive correlation between the nerve response amplitude and nerve fiber density in the control group and during PCTX treatment as measured according to the present disclosure, while Chart 1602 shows a highly negative correlation between the nerve response latency and nerve fiber density in the control group and during PCTX treatment as measured according to the present disclosure, thus confirming the authenticity of the neuropathy detection method described herein.

[0130] Figure 17 is a flowchart showing an example process for determining a nerve condition using image analysis according to an embodiment. Method 1700 can be executed by processing logic, which may include hardware (e.g., circuits, dedicated logic, programmable logic, microcode, etc.), software (e.g., instructions executed on a processing device), or a combination thereof. It should be understood that not all steps are necessary to implement the disclosure provided herein. In addition, those of ordinary skill in the art can understand that some of these steps can be executed simultaneously or in an order different from the Figure 17 order shown therein.

[0131] Reference will be made to Figure 1 to describe Method 1700. However, Method 1700 is not limited to this example embodiment.

[0132] In operation 1702, an image of patient tissue is received. For example, server 100 may receive an image of patient tissue. The image shows the nerves and layers of the patient tissue. The patient tissue may be skin tissue. The image may be received together with a request to determine the nerve condition of the patient tissue.

[0133] In operation 1704, each nerve and layer of the patient tissue in the image is labeled. For example, server 100 may use model 104 to label each nerve and each layer of the patient tissue in the image. Model 104 is trained to be able to recognize the difference between nerves and tissue. Specifically, model 104 is trained to label each pixel as corresponding to a given tissue layer, nerve, or image background.

[0134] In operation 1706, the amount of nerve area relative to the tissue area is determined based on the labeling. For example, server 100 may use analysis engine 102 to determine the amount of nerve area relative to the layer (e.g., epidermis) of the tissue area of the patient tissue in the image, based on the labels assigned to the nerves and layers of the patient tissue in the image. Analysis engine 102 may isolate the pixels associated with the nerves based on the nerves labeled in the image. In addition, analysis engine 102 may identify the total number of pixels labeled as corresponding to the layer of the tissue. That is, analysis engine 102 may exclude the pixels labeled as background. Analysis engine 102 may determine the number of pixels covering the tissue area.

[0135] In operation 1708, the nerve condition is determined. For example, server 100 may (using analysis engine 102) determine the nerve condition based on the amount of nerve area relative to the tissue area (e.g., epidermis area). The amount of nerve area relative to the tissue area may be referred to as nerve density. In some embodiments, a nerve density below a threshold amount may indicate peripheral neuropathy. Alternatively, a nerve density lower than a previously determined nerve density may indicate peripheral neuropathy. Additionally, analysis engine 102 may use the amount of nerve area relative to the tissue area to determine the effectiveness of an ongoing treatment for a patient diagnosed with neuropathy.

[0136] In some embodiments, analysis engine 102 may generate a boundary between the epidermis layer and the dermis layer in the image based on the labels assigned by model 104. Analysis engine 102 may determine the number of nerve fiber crossings across the boundary. A number of nerve fiber crossings below a threshold amount may indicate peripheral neuropathy. Additionally, analysis engine 102 may use the number of nerve fiber crossings to determine the effectiveness of an ongoing treatment for a patient diagnosed with neuropathy.

[0137] Various embodiments may be implemented, for example, using one or more computer systems (such as Figure 18 the computer system 1800 shown inFigure 17 Method 1700 therein. Additionally, computer system 1800 can be at least a portion of server 100, client device 110, and data storage device 120, such as Figure 1 shown. For example, computer system 1800 routes communications to various applications. Computer system 1800 can be any computer capable of performing the functions described herein.

[0138] Computer system 1800 can be any well-known computer capable of performing the functions described herein.

[0139] Computer system 1800 includes one or more processors (also referred to as central processing units or CPUs), such as processor 1804. Processor 1804 is connected to a communication infrastructure or bus 1806.

[0140] One or more of processors 1804 can each be a graphics processing unit (GPU). In one embodiment, a GPU is a type of processor, an electronic circuit specifically designed to handle math-intensive applications. A GPU can have a parallel architecture capable of efficiently processing large blocks of data in parallel (e.g., math-intensive data common in computer graphics applications, images, videos, etc.).

[0141] Computer system 1800 also includes user input / output devices 1803 that communicate with communication infrastructure 1806 via user input / output interface 1802, such as monitors, keyboards, pointing devices, etc.

[0142] Computer system 1800 also includes main or primary memory 1908, such as random access memory (RAM). Main memory 1808 can include one or more levels of cache. Control logic (i.e., computer software) and / or data is stored in main memory 1808.

[0143] Computer system 1800 can also include one or more secondary storage devices or memories 1810. Secondary memory 1810 can include, for example, hard disk drive 1812 and / or removable storage device or drive 1814. Removable storage drive 1814 can be a floppy disk drive, tape drive, optical disk drive, optical storage device, tape backup device, and / or any other storage device / drive.

[0144] The removable storage drive 1814 can interact with a removable storage unit 1818. The removable storage unit 1818 includes a computer-usable or readable storage device storing computer software (control logic) and / or data. The removable storage unit 1818 can be a floppy disk, magnetic tape, compact disk, DVD, optical storage disk, and / or any other computer data storage device. The removable storage drive 1814 reads from and / or writes to the removable storage unit 1818 in a well-known manner.

[0145] According to an exemplary embodiment, the secondary memory 1810 can include other components, tools, or other methods for allowing the computer system 1800 to access computer programs and / or other instructions and / or data. Such components, tools, or other methods can include, for example, a removable storage unit 1822 and an interface 1820. Examples of the removable storage unit 1822 and the interface 1820 can include a program cartridge and a cartridge interface (such as found in video game devices), a removable memory chip (such as an EPROM or PROM) and an associated socket, a memory stick and a USB port, a memory card and an associated memory card slot, and / or any other removable storage unit and an associated interface.

[0146] The computer system 1800 can also include a communication or network interface 1824. The communication interface 1824 enables the computer system 1800 to communicate and interact with any combination of remote devices, remote networks, remote entities, etc. (collectively and individually referred to by the reference numeral 1828). For example, the communication interface 1824 can allow the computer system 1900 to communicate with the remote device 1828 via a communication path 1826, which can be wired and / or wireless and can include any combination of a LAN, a WAN, the Internet, etc. Control logic and / or data can be transmitted to and from the computer system 1800 via the communication path 1826.

[0147] In one embodiment, a tangible, non-transitory device or article including a tangible, non-transitory computer-usable or readable medium storing control logic (software) is also referred to herein as a computer program product or a program storage device. This includes, but is not limited to, the computer system 1800, the main memory 1808, the secondary memory 1810, and the removable storage units 1818 and 1822, as well as any tangible article embodying any combination of the foregoing. When such control logic is executed by one or more data processing devices (such as the computer system 1800), the data processing devices are caused to operate as described herein.

[0148] Based on the teachings contained in this disclosure, for those skilled in the relevant art, how to use in addition to Figure 18It will be apparent to those of ordinary skill in the art that embodiments of the present disclosure can be made and used by a data processing device, computer system, and / or computer architecture other than those shown in the figures. In particular, the embodiments can operate in conjunction with software, hardware, and / or operating systems other than those described herein.

[0149] It should be understood that the detailed description section (and not any other section) is intended to interpret the claims. The other sections may set forth one or more, but not all, of the exemplary embodiments contemplated by the inventors and are thus not intended to limit the present disclosure or the appended claims in any way.

[0150] Although the present disclosure describes exemplary embodiments for exemplary fields and applications, it should be understood that the present disclosure is not limited thereto. Other embodiments and modifications thereof are possible and within the scope and spirit of the present disclosure. For example, and without limiting the generality of this paragraph, the embodiments are not limited to the software, hardware, firmware, and / or entities shown in the figures and / or described herein. Further, the embodiments (whether explicitly described herein or not) have significant utility for fields and applications other than those described herein.

[0151] The embodiments have been described herein in terms of functional building blocks that perform specific functions and relationships thereof. For ease of description, the boundaries of these functional building blocks have been arbitrarily defined herein. Alternate boundaries can be defined so long as the specified functions and relationships (or their equivalents) are appropriately performed. In addition, alternative embodiments can perform functional blocks, steps, operations, methods, etc. in an order different from that described herein.

[0152] References herein to "one embodiment", "an embodiment", "an example embodiment", or similar phrases indicate that the embodiment described may include a particular feature, structure, or characteristic, but every embodiment may not necessarily include the particular feature, structure, or characteristic. Moreover, such phrases do not necessarily refer to the same embodiment. Further, when a particular feature, structure, or characteristic is described in connection with an embodiment, whether or not explicitly recited or described herein, incorporating such feature, structure, or characteristic into other embodiments will be within the knowledge of those of ordinary skill in the relevant art. Additionally, some embodiments may use the terms "coupled" and "connected" and their derivatives to describe. These terms are not necessarily intended as synonyms for each other. For example, some embodiments may be described using the terms "connected" and / or "coupled" to indicate that two or more elements are in direct physical or electrical contact with each other. However, the term "coupled" can also mean that two or more elements are not in direct contact with each other but still cooperate or interact with each other.

[0153] The breadth and scope of the present disclosure should not be limited by any of the above exemplary embodiments, but should be defined only in accordance with the appended claims and their equivalents.

Claims

1. A diagnostic method for determining a nerve condition, the method comprising: Receiving, by a processor, an image of a patient's skin, the image showing a plurality of nerves and a plurality of layers of the patient's skin; Using a trained model executed by the processor to label each nerve of the plurality of nerves and each layer of the plurality of layers of the patient's skin in the image, wherein the trained model has been trained to recognize differences between the nerves and the plurality of layers of skin tissue; Determining, by the processor, an amount of a layer of nerve area relative to an area of the patient's skin tissue based on the labels assigned to the plurality of nerves and the plurality of layers of the patient's skin; And Determining a nerve condition based on the amount of the layer of nerve area relative to the area of the patient's skin tissue.

2. The method according to claim 1, wherein the amount of nerve area relative to the area of the patient's skin tissue corresponds to nerve density.

3. The method according to claim 2, wherein determining the nerve condition includes determining the presence of peripheral neuropathy when the nerve density is below a threshold amount.

4. The method according to claim 2, further comprising: Receiving, by the processor, a second image of the patient's skin, the second image showing a plurality of nerves and a plurality of layers of the patient's skin; Using the trained model executed by the processor to label each nerve of the plurality of nerves and each layer of the plurality of layers of the patient's skin in the second image; And Determining, by the processor, a second amount of nerve area relative to the skin tissue area based on the labels assigned to the plurality of nerves and the plurality of layers of the patient's skin, wherein the amount of the nerve area is a first amount, and determining the nerve condition includes determining a difference between the first amount and the second amount, the difference corresponding to a difference in nerve density, and determining the presence of peripheral neuropathy when the difference in nerve density is greater than a threshold amount.

5. The method according to claim 1, wherein labeling each nerve of the plurality of nerves and each layer of the plurality of layers in the image of the patient's skin includes: Using the model executed by the processor to determine whether a pixel among a plurality of pixels in the image corresponds to a nerve of the plurality of nerves of the patient's skin or a corresponding layer among the plurality of layers; And Assigning, by the processor, a corresponding label to each pixel of the plurality of pixels, the corresponding label indicating whether each corresponding pixel corresponds to a nerve of the patient's skin or a corresponding layer among the plurality of layers.

6. The method according to claim 1, further comprising training the model, wherein the training of the model includes: Assigning, by the processor, training labels to each nerve and each layer of a plurality of layers of tissue in each of a plurality of training images; Inputting, by the processor, the plurality of training images into a convolutional neural network to label each nerve and each layer of the plurality of layers of tissue in each of the plurality of training images; And The processor validates each respective tag of each layer of the nerve and the tissue in each of the plurality of training images assigned by the convolutional neural network with a respective training tag corresponding to each layer of the nerve and the plurality of tags.

7. The method according to claim 6, wherein assigning the training tags to each layer of the nerve and the tissue in each of the plurality of training images comprises: The processor applies one or more filters to each of the plurality of training images to identify the nerve in each of the plurality of training images.

8. The method according to claim 6, wherein the training of the model comprises: The processor inputs a mirrored or rotated version of each of the plurality of training images into the convolutional neural network to label each layer of the nerve and the tissue in the mirrored or rotated version of each of the plurality of training images.

9. The method according to claim 6, wherein a set of the plurality of training images comprises copies of a single image, and each of the copies comprises a different contrast, brightness, cropping, distortion, or rotation from the single image.

10. The method according to claim 1, wherein the layer corresponds to the epidermis.

11. A diagnostic system for determining a nerve condition, the system comprising: A memory comprising instructions stored thereon; And A processor coupled to the memory, wherein the instructions, when executed by the processor, cause the processor to: Receive an image of a patient's skin, the image showing a plurality of nerves and a plurality of layers of the patient's skin; Use a trained model to label each of the plurality of nerves and each of the plurality of layers of the patient's skin in the image, wherein the trained model has been trained to identify differences between the nerves and the plurality of layers of skin tissue; Determine the amount of the layer of the nerve area relative to the area of the patient's skin tissue based on the labels assigned to the plurality of nerves and the plurality of layers of the patient's skin; And Determine the nerve condition based on the amount of the layer of the nerve area relative to the tissue area.

12. The system according to claim 11, wherein the amount of the nerve area relative to the tissue area corresponds to nerve density.

13. The system according to claim 12, wherein determining the nerve condition comprises determining the presence of peripheral neuropathy when the nerve density is below a threshold amount.

14. The system according to claim 12, wherein the instructions, when executed by the processor, further cause the processor to: Receive a second image of the patient's skin, the second image showing a plurality of nerves and a plurality of layers of the patient's skin; Use the trained model to label each of the plurality of nerves and each of the plurality of layers of the patient's skin in the second image; and Determine a second quantity of the nerve area relative to the skin tissue area based on the markings of the plurality of nerves and the plurality of layers assigned to the patient's skin, wherein the quantity of the nerve area is a first quantity, and determining the nerve condition includes determining the difference between the first quantity and the second quantity, the difference corresponding to a difference in nerve density, and determining the presence of peripheral neuropathy when the difference in nerve density is greater than a threshold quantity.

15. The system according to claim 11, wherein marking each nerve of the plurality of nerves and each layer of the plurality of layers in the image of the patient's skin includes: Using the model to determine whether a pixel among a plurality of pixels in the image corresponds to a nerve of the plurality of nerves of the patient's skin or a corresponding layer of the plurality of layers; And Assigning a corresponding marking to each pixel of the plurality of pixels, the corresponding marking indicating whether each corresponding pixel corresponds to a nerve of the patient's skin or a corresponding layer of the plurality of layers.

16. The system according to claim 16, wherein the instructions, when executed, cause the processor to: Train the model, wherein the training of the model includes: Assigning training markings to each nerve and each layer of a plurality of layers of tissue in each of a plurality of training images; Inputting the plurality of training images into a convolutional neural network to mark each nerve and each layer of the plurality of layers of tissue in each of the plurality of training images; And Validating each corresponding marking of each nerve and each layer of the plurality of layers of tissue in each of the plurality of training images assigned by the convolutional neural network with the corresponding training marking corresponding to each nerve and each layer of the plurality of markings.

17. The system according to claim 16, wherein assigning the training markings to each nerve and each layer of the plurality of layers of tissue in each of the plurality of training images includes: Applying one or more filters to each of the plurality of training images to identify the nerves in each of the plurality of training images.

18. The system according to claim 16, wherein the training of the model includes: Inputting a mirrored or rotated version of each of the plurality of training images into the convolutional neural network to mark each nerve and each layer of the plurality of layers of tissue in the mirrored or rotated version of each of the plurality of training images.

19. The system according to claim 16, wherein a set of the plurality of training images includes copies of a single image, and each copy in the copies includes a different contrast, brightness, cropping, distortion, or rotation from the single image.

20. The system according to claim 11, wherein the layer corresponds to the epidermis.

21. A non-transitory machine-readable medium having instructions stored thereon, the instructions, when executed by at least one computing device, cause the at least one computing device to perform operations, the operations including: Receive an image of a patient's skin, the image showing a plurality of nerves and a plurality of layers of the patient's skin; Use a trained model to label each nerve of the plurality of nerves and each layer of the plurality of layers of the patient's skin in the image, wherein the trained model has been trained to recognize differences between the nerves and the plurality of layers of skin tissue; Determine the amount of nerve area relative to the area of the layer of the patient's skin tissue based on the labels assigned to the plurality of nerves and the plurality of layers of the patient's skin; And Determine a nerve condition based on the amount of the nerve area relative to the layer of the patient's skin tissue.

22. The non-transitory machine-readable medium according to claim 21, wherein the amount of the nerve area relative to the area of the patient's skin tissue corresponds to nerve density.

23. The non-transitory machine-readable medium according to claim 22, wherein determining the nerve condition includes determining the presence of peripheral neuropathy when the nerve density is below a threshold amount.

24. The non-transitory machine-readable medium according to claim 22, wherein the operations further include: Receive a second image of the patient's skin, the second image showing a plurality of nerves and a plurality of layers of the patient's skin; Use the trained model to label each nerve of the plurality of nerves and each layer of the plurality of layers of the patient's skin in the second image; And Determine a second amount of nerve area relative to the skin tissue area based on the labels assigned to the plurality of nerves and the plurality of layers of the patient's skin, wherein the amount of the nerve area is a first amount, and determining the nerve condition includes determining a difference between the first amount and the second amount, the difference corresponding to a difference in nerve density, and determining the presence of peripheral neuropathy when the difference in nerve density is greater than a threshold amount.

25. The non-transitory machine-readable medium according to claim 21, wherein labeling each nerve of the plurality of nerves and each layer of the plurality of layers in the image of the patient's skin includes: Using the model to determine whether a pixel among a plurality of pixels in the image corresponds to a nerve of the plurality of nerves of the patient's skin or a corresponding layer among the plurality of layers; And Assigning a corresponding label to each pixel of the plurality of pixels, the corresponding label indicating whether each corresponding pixel corresponds to a nerve of the patient's skin or a corresponding layer among the plurality of layers.

26. The non-transitory machine-readable medium according to claim 21, wherein the operations further include: Training the model, wherein the training of the model includes: Assigning training labels to each nerve and each layer of a plurality of layers of tissue in each of a plurality of training images; Inputting the plurality of training images into a convolutional neural network to label each nerve and each layer of the plurality of layers of tissue in each of the plurality of training images; and Validating each respective label of each layer of the nerve and the tissue in each of the plurality of training images assigned by the convolutional neural network with a respective training label corresponding to each layer of the nerve and the plurality of labels.

27. The non-transitory machine-readable medium according to claim 26, wherein assigning the training labels to each layer of the nerve and the tissue in each of the plurality of training images includes: Applying one or more filters to each of the plurality of training images to identify the nerve in each of the plurality of training images.

28. The non-transitory machine-readable medium according to claim 26, wherein the training of the model includes: Inputting a mirrored or rotated version of each of the plurality of training images into the convolutional neural network to label each layer of the nerve and the tissue in the mirrored or rotated version of each of the plurality of training images.

29. The non-transitory machine-readable medium according to claim 26, wherein a set of the plurality of training images includes copies of a single image, and each of the copies includes a different contrast, brightness, cropping, distortion, or rotation from the single image.

30. The non-transitory machine-readable medium according to claim 21, wherein the layer corresponds to the epidermis.

31. A diagnostic method for determining a nerve condition, the method comprising: Receiving, by a processor, an image of a patient's skin, the image showing a plurality of nerves and a plurality of layers of the patient's skin; Using a trained model executed by the processor to label each of the plurality of nerves and each of the plurality of layers of the patient's skin in the image, wherein the trained model has been trained to identify differences between the nerves and the plurality of layers of skin tissue; Determining, by the processor, the number of nerve crossings between adjacent layers of skin tissue area based on the labels assigned to the plurality of nerves and the plurality of layers of the patient's skin; And Determining a nerve condition based on the number of nerve crossings.

32. The method according to claim 31, wherein determining the nerve condition includes determining the presence of peripheral neuropathy when the number of nerve crossings is below a threshold amount.

33. The method according to claim 31, further comprising: Receiving, by the processor, a second image of the patient's skin, the second image showing a plurality of nerves and a plurality of layers of the patient's skin; Using the trained model executed by the processor to label each of the plurality of nerves and each of the plurality of layers of the patient's skin in the second image; And Determining, by the processor, a second number of nerve crossings between adjacent layers of skin tissue area in the second image based on the labels assigned to the plurality of nerves and the plurality of layers of the patient's skin, The number of the nerve crossings is a first number, and determining the nerve condition includes determining a difference between the first number and the second number, and determining the presence of peripheral neuropathy when the difference is greater than a threshold amount.

34. The method according to claim 31, wherein marking each nerve of the plurality of nerves and each layer of the plurality of layers in the image of the patient's skin includes: Using the model executed by the processor to determine whether a pixel among a plurality of pixels in the image corresponds to a nerve of the plurality of nerves of the patient's skin or a corresponding layer of the plurality of layers; And Assigning, by the processor, a corresponding label to each pixel of the plurality of pixels, the corresponding label indicating whether each corresponding pixel corresponds to a nerve of the patient's skin or a corresponding layer of the plurality of layers.

35. The method according to claim 31, further comprising training the model, wherein the training of the model includes: Assigning, by the processor, training labels to each nerve and each layer of a plurality of layers of tissue in each of a plurality of training images; Inputting, by the processor, the plurality of training images into a convolutional neural network to label each nerve and each layer of the plurality of layers of tissue in each of the plurality of training images; And Validating, by the processor, each corresponding label of each nerve and each layer of the plurality of layers of tissue in each of the plurality of training images assigned by the convolutional neural network with corresponding training labels corresponding to each nerve and each layer of the plurality of labels.

36. The method according to claim 35, wherein assigning the training labels to each nerve and each layer of the plurality of layers of tissue in each of the plurality of training images includes: Applying, by the processor, one or more filters to each of the plurality of training images to identify the nerves in each of the plurality of training images.

37. The method according to claim 35, wherein the training of the model includes: Inputting, by the processor, a mirrored or rotated version of each of the plurality of training images into the convolutional neural network to label each nerve and each layer of the plurality of layers of tissue in the mirrored or rotated version of each of the plurality of training images.

38. The method according to claim 35, wherein a set of the plurality of training images includes copies of a single image, and each copy in the copies includes a different contrast, brightness, cropping, distortion, or rotation from the single image.

39. The method according to claim 31, wherein the layer corresponds to the epidermis.

40. A diagnostic system for determining a nerve condition, the system comprising: A memory including instructions stored thereon; And A processor coupled to the memory, wherein the instructions, when executed by the processor, cause the processor: Receive an image of a patient's skin, the image showing a plurality of nerves and a plurality of layers of the patient's skin; Use a trained model to label each nerve of the plurality of nerves and each layer of the plurality of layers of the patient's skin in the image, wherein the trained model has been trained to recognize differences between the nerves and the plurality of layers of skin tissue; Determine the number of nerve crossings between adjacent layers of skin tissue area based on the labels assigned to the plurality of nerves and the plurality of layers of the patient's skin; And Determine a nerve condition based on the number of nerve crossings.

41. The system according to claim 40, wherein determining the nerve condition includes determining the presence of peripheral neuropathy when the number of nerve crossings is below a threshold amount.

42. The system according to claim 40, wherein the instructions, when executed by the processor, further cause the processor to: Receive a second image of the patient's skin, the second image showing a plurality of nerves and a plurality of layers of the patient's skin; Use the trained model to label each nerve of the plurality of nerves and each layer of the plurality of layers of the patient's skin in the second image; and Determine a second number of nerve crossings between adjacent layers of skin tissue area in the second image based on the labels assigned to the plurality of nerves and the plurality of layers of the patient's skin, wherein the number of nerve crossings is a first number, and determining the nerve condition includes determining a difference between the first number and the second number, and determining the presence of peripheral neuropathy when the difference is greater than a threshold amount.

43. The system according to claim 40, wherein labeling each nerve of the plurality of nerves and each layer of the plurality of layers in the image of the patient's skin includes: Using the model to determine whether a pixel among a plurality of pixels in the image corresponds to a nerve of the plurality of nerves of the patient's skin or a corresponding layer among the plurality of layers; And Assign a corresponding label to each of the plurality of pixels, the corresponding label indicating whether each corresponding pixel corresponds to a nerve of the patient's skin or a corresponding layer among the plurality of layers.

44. The system according to claim 40, wherein the instructions, when executed, cause the processor to: train the model, wherein the training of the model includes: Assign training labels to each nerve and each layer of a plurality of layers of tissue in each of a plurality of training images; Input the plurality of training images into a convolutional neural network to label each nerve and each layer of the plurality of layers of tissue in each of the plurality of training images; And Validate each corresponding label assigned to each nerve and each layer of the plurality of layers of tissue in each of the plurality of training images by the convolutional neural network with a corresponding training label corresponding to each nerve and each layer of the plurality of labels.

45. The system according to claim 44, wherein assigning the training labels to each of the multiple layers of the nerves and the tissue in each of the multiple training images comprises: Applying one or more filters to each of the multiple training images to identify the nerves in each of the multiple training images.

46. The system according to claim 44, wherein the training of the model comprises: Inputting a mirrored or rotated version of each of the multiple training images into the convolutional neural network to label each of the multiple layers of the nerves and the tissue in the mirrored or rotated version of each of the multiple training images.

47. The system according to claim 44, wherein a set of the multiple training images comprises copies of a single image, and each of the copies comprises a different contrast, brightness, cropping, distortion, or rotation from the single image.

48. The system according to claim 40, wherein the layer corresponds to the epidermis.

49. A non-transitory machine-readable medium having instructions stored thereon, the instructions when executed by at least one computing device cause the at least one computing device to perform operations, the operations comprising: Receiving an image of a patient's skin, the image showing multiple nerves and multiple layers of the patient's skin; Using a trained model to label each of the multiple nerves and each of the multiple layers of the patient's skin in the image, wherein the trained model has been trained to identify differences between the nerves and the multiple layers of the skin tissue; Determining the number of nerve crossings between adjacent layers of the skin tissue area based on the labels assigned to the multiple nerves and the multiple layers of the patient's skin; And Determining a nerve condition based on the number of nerve crossings.

50. The non-transitory machine-readable medium according to claim 49, wherein determining the nerve condition comprises determining the presence of peripheral neuropathy when the number of nerve crossings is below a threshold amount.

51. The non-transitory machine-readable medium according to claim 49, wherein the operations further comprise: Receiving a second image of the patient's skin, the second image showing multiple nerves and multiple layers of the patient's skin; Using the trained model to label each of the multiple nerves and each of the multiple layers of the patient's skin in the second image; And Determining a second number of nerve crossings between adjacent layers of the skin tissue area in the second image based on the labels assigned to the multiple nerves and the multiple layers of the patient's skin, wherein the number of nerve crossings is a first number, and determining the nerve condition comprises determining a difference between the first number and the second number, and determining the presence of peripheral neuropathy when the difference is greater than a threshold amount.

52. The non-transitory machine-readable medium according to claim 49, wherein labeling each of the multiple nerves and each of the multiple layers of the patient's skin in the image comprises: Use the model to determine whether a pixel among a plurality of pixels in the image corresponds to a nerve among the plurality of nerves of the patient's skin or a corresponding layer among the plurality of layers; and Assign a corresponding label to each pixel among the plurality of pixels, the corresponding label indicating whether each corresponding pixel corresponds to a nerve of the patient's skin or a corresponding layer among the plurality of layers.

53. The non-transitory machine-readable medium according to claim 49, wherein the operation further comprises: Training the model, wherein the training of the model comprises: Assigning training labels to each layer among the plurality of nerves and the plurality of layers of tissue in each of the plurality of training images; Inputting the plurality of training images into a convolutional neural network to label each layer among the plurality of nerves and the plurality of layers of tissue in each of the plurality of training images; and Validating each corresponding label of each layer among the plurality of nerves and the plurality of layers of tissue in each of the plurality of training images assigned by the convolutional neural network with a corresponding training label corresponding to each layer of the nerve and the plurality of labels.

54. The non-transitory machine-readable medium according to claim 53, wherein assigning the training labels to each layer among the plurality of nerves and the plurality of layers of tissue in each of the plurality of training images comprises: Applying one or more filters to each of the plurality of training images to identify the nerves in each of the plurality of training images.

55. The non-transitory machine-readable medium according to claim 53, wherein the training of the model comprises: Inputting a mirrored or rotated version of each of the plurality of training images into the convolutional neural network to label each layer among the plurality of nerves and the plurality of layers of tissue in the mirrored or rotated version of each of the plurality of training images.

56. The non-transitory machine-readable medium according to claim 53, wherein a set of the plurality of training images includes copies of a single image, and each copy in the copies includes a different contrast, brightness, cropping, distortion, or rotation from the single image.

57. The non-transitory machine-readable medium according to claim 49, wherein the layer corresponds to the epidermis.

Images