Attorney Docket No.46567-1742WO1 NEURAL NETWORK SYSTEM FOR PREDICTING AND LOCALIZING SKIN CONDITIONS CROSS-REFERENCE TO RELATED APPLICATIONS The present application claims priority to U.S. Provisional Application No. 63/659,914, filed on June 14, 2024, the disclosure of which is incorporated herein by reference. FIELD The specification describes systems, methods, and apparatus for using machine- learning techniques to assess the severity of skin conditions. BACKGROUND Skin conditions, such as atopic eczema, may be characterized by inflammation, itchiness, and/or chronic skin lesions, and can significantly impact the quality of life of a subject. Traditional assessment methods for skin conditions rely heavily on visual inspections and subjective judgments. Severity scores, such as the Eczema Area and Severity Index (EASI), can provide quantitative tools that provide a systematic and reproducible way to evaluate the severity and extent of a skin condition. However, manual severity scoring is time-consuming and prone to variability. SUMMARY According to a first aspect of this specification, there is described a computer implemented method for assessing the severity of a skin condition, the method comprising: receiving one or more images of respective sections of skin of a subject; generating, using a segmentation neural network, one or more image segments from the one or more images, the one or image segments each identifying a potential location of skin in an image in the one or more images affected by the skin condition; and processing, using one or more severity models, the one or more image segments to generate one or more severity scores for the one or more images of respective sections of skin of the subject. The segmentation neural network may comprise a pre-trained large-scale model that has been fine-tuned on a training dataset comprising a plurality of images of the skin of one or more subjects and respective ground truth image segments of skin patches affected by the skin condition. The training dataset may comprise a plurality of original training examples taken from real-world images of the skin of one or more subjects and a plurality of synthetic
Attorney Docket No.46567-1742WO1 training examples generated from one or more of the original training examples using an image-to-image translation model. The segmentation neural network may comprise an image transformer model. The segmentation neural network may comprise a multi-layer perceptron model subsequent to the image transformer model. The one or more severity models may comprise one or more convolutional neural network models. The one or more convolutional neural network models may comprise one or more residual networks. The one or more severity models may comprise one or more vision transformer models. The one or more severity models may comprise one or more hybrid models. The method may further comprise: generating, using a large language model, a natural language description of a severity of the skin condition from the one or more severity scores, comprising: retrieving one or more document sections from one or more documents in a database based at least in part on the one or more severity scores, wherein the one or more documents relate to the skin condition; inputting, into the large language model, input data comprising the one or more severity scores and the one or more document sections; and generating, by the large language model, the natural language description of a severity of the skin condition based at least in part on the one or more severity scores and the one or more document sections. Retrieving one or more document sections from one or more documents in the database based at least in part on the one or more severity scores may comprise: retrieving a plurality of document sections from a plurality of documents in a database based at least in part on the one or more severity scores, wherein one or more of the documents relate to the skin condition and one or more of the documents relate to the subject. The skin condition may be a skin lesion condition. The skin condition may be eczema, and the one or more severity scores may comprise one or more Eczema Area and Severity Index, EASI, scores. According to a further aspect of this specification, there is described a system comprising: one or more processors; and a memory, the memory storing computer readable instructions that, when executed by the one or more processors, causes the system to perform operations comprising any one or more of the methods described herein.
Attorney Docket No.46567-1742WO1 According to a further aspect of this specification, there is described a computer program product comprising computer readable instructions that, when executed by data processing apparatus, cause the data processing apparatus to perform operations comprising any one or more of the methods described herein. The subject matter described in this specification can be implemented in particular embodiments so as to realize one or more of the following advantages. The described neural network system for predicting and localizing skin conditions offers significant improvements in automated dermatological diagnostics. By automating the prediction and localization of conditions like atopic eczema using neural networks, the system can significantly reduce the subjectivity and labor involved in traditional assessment methods. The described technology addresses limitations of manual scoring, which is time- consuming and prone to variability. The ability to provide rapid and accurate assessments supports timely and tailored treatment decisions, ultimately improving patient outcomes. Furthermore, the scalable and adaptable nature of the neural network models allows for continuous improvement and integration into diverse clinical environments, making it a valuable tool for healthcare providers. The utilization of generative AI components can pave the way for the rapid delivery of personalized treatments. By understanding the severity of eczema lesions, healthcare providers can recommend targeted therapies that optimize patient outcomes and minimize potential side effects. In regions with limited access to dermatologists or specialized healthcare facilities, such systems can serve as a valuable resource, enabling primary care providers to obtain accurate eczema assessments and facilitating early intervention and timely referrals for patients requiring specialist care. In some implementations, the system achieves enhanced accuracy through the use of neural network architectures such as Vision Transformer (ViT) and hybrid models, which leverage their ability to capture global context and balance local and global features. These models demonstrate improved performance compared to existing models due to their ability to extract more relevant and discriminative features from complex dermatological images. This improvement enhances computer systems by improving computational efficiency. ViT and hybrid models reduce the reliance on deep convolutional stacks, which often require large numbers of parameters and multiple layers to model long-range dependencies. By using attention mechanisms to directly model relationships across the entire image, these models can achieve higher accuracy with fewer layers and reduced inference
Attorney Docket No.46567-1742WO1 time, thereby reducing the overall computational burden. Lightweight variants of ViT and hybrid models can achieve accuracy levels comparable to or better than heavier CNN architectures, enabling deployment on edge devices with limited compute power, such as smartphones or tablets. This improves the overall utility and scalability of the system in real- world settings, particularly in resource-constrained environments. As a result, the use of these architectures allows computer systems to perform complex image analysis tasks more accurately and responsively, while consuming less computational power and facilitating broader accessibility. BRIEF DESCRIPTION OF THE DRAWINGS Example implementations will be described by way of reference to the accompanying drawings, in which: FIG.1 shows an overview of an example system/method for assessing the severity of a skin condition using machine-learning; FIG.2A shows an example model structure for a machine-learning segmentation model; FIG.2B shows an example model structure for a machine-learning segmentation model; FIG.3 shows an overview of an example system/method for assessing the severity of a skin condition using machine-learning that utilises a large language model; FIG.4 shows an example of a method for fine-tuning a segmentation model for localising skin conditions in images; FIG.5 shows a flow diagram of an example method for assessing the severity of a skin condition using machine-learning; FIG.6 shows a flow diagram of an example method for assessing the severity of a skin condition using machine-learning; and FIG.7 shows a schematic overview of a computing system/apparatus. FIG.8 shows a comparison of the segmentation accuracy of different neural network models used for identifying Atopic Dermatitis (AD) regions. FIG.9 shows the performance of several machine learning models in predicting the Eczema Area and Severity Index (EASI) scores.
Attorney Docket No.46567-1742WO1 DETAILED DESCRIPTION This specification describes a neural network system designed to predict severity scores for skin conditions, such as the Eczema Area and Severity Index (EASI) score for eczema, a crucial metric for assessing the extent and severity of atopic eczema. The systems and methods described herein integrate lesion localization with accurate disease severity prediction, providing a robust tool for dermatological assessment and treatment planning. The systems, methods and apparatus described herein provide improvements in automated dermatological diagnostics. By automating the prediction and localization of atopic eczema using neural networks, the systems, methods, and apparatus described herein can significantly reduce the subjectivity and labour involved in traditional methods of diagnosing and/or assessing dermatological conditions. The ability to provide rapid and accurate assessments supports timely and tailored treatment decisions, ultimately improving patient outcomes. Furthermore, the scalable and adaptable nature of the neural network models described herein allows for continuous improvement and integration into diverse clinical environments, making it an invaluable tool for healthcare providers worldwide. Furthermore, by utilising generative AI components, the systems methods and apparatus described herein can pave the way for the rapid delivery of personalized treatments. By understanding the severity of eczema lesions, healthcare providers can recommend targeted therapies that optimize patient outcomes and minimize potential side effects. In regions with limited access to dermatologists or specialized healthcare facilities, such systems can serve as a valuable resource. Primary care providers can utilize the system to obtain accurate eczema assessments, facilitating early intervention and timely referrals for patients who require further evaluation or treatment by dermatological specialists. FIG.1 shows an overview of an example method 100 for assessing the severity of a skin condition using machine-learning. The method may be performed by one or more computing devices/systems, such as the device/system described in relation to FIG.7. One or more images 102 of the skin of a subject are input into a segmentation neural network 104. The segmentation neural network 104 processes the input one or more images 102 to generate an image segmentation 106 that identifies one or more segments/locations 106A, 106B in the one or more images 102 that contain skin affected by a skin disorder. The identified image segments 106A, 106B are input into a severity classification model 108 that
Attorney Docket No.46567-1742WO1 process the identified image segments 106A, 106B to generate one or more severity scores 110 for the skin condition present in the one or more images 102. The combination of two stages, segmentation followed by severity assessment, allows the system/method 100 not only to predict the presence of eczema but also to accurately assess its severity. The skin disorder may be any skin disorder with symptoms that present on the skin of a subject, e.g., in the form of lesions, rashes, spots, blisters and/or the like. Examples of such skin conditions include, but are not limited to, eczema (e.g., atopic dermatitis), acne, fungal infections, burns, skin cancer, and/or contact dermatitis. In the following description, eczema will be used as an example skin condition, though it will be appreciated that the systems and method described herein are applicable to other skin conditions too. The one or more images 102 may comprise one or more colour images (e.g., RGB images) of a part/patch of the skin of a subject. The one or more images 102 may have predefined size, e.g., a predefined height (H) and width (W) in pixels. Such an image may be represented/stored/processed as a H x W x C dimensional array of pixel values, where C is the number of colour channels, e.g., 3 in the RGB case. The one or images 102 have been captured by a user (e.g., the subject themselves, a healthcare worker, or some other third party) using an image capturing device, e.g., a camera of a smartphone or tablet device, a digital camera, or the like. In some examples, the one or more images 102 may comprise stereoscopic images, i.e., images captured using a stereoscopic camera. Such images may additionally comprise image depth data relating to the depth of features in the image, e.g., the image may be in an RGB-D format. The segmentation network 104 is a neural network model that has been trained to segment images 102 of skin to identify potential locations that are affected by a dermatological condition, for example using the training method described in relation to FIG. 4. The segmentation network 104 comprises one or more (e.g., a plurality) of neural network layers. The layers of the segmentation network 102 may comprise one or more convolutional layers, e.g., be a convolutional neural network. The segmentation network 102 may comprise one or more skip connections (e.g., residual connections) between layers of the segmentation network 102. Alternatively or additionally, the layers of the segmentation network 102 may comprise one or more transformer blocks. Alternatively or additionally, the layers of the
Attorney Docket No.46567-1742WO1 segmentation network 102 may comprise one or more fully connected layers, such as multi- layer perceptrons. Some examples of structures of the segmentation network 104 are provided in FIGs.2A (a U-net) and 2B (a mix transformer, such as SegFormer). The segmentation network 104 processes the one or more input images 102 based on a set of learned parameters of the segmentation network 104 to generate image segmentation data 106. The image segmentation data 106 comprises data identifying potential locations of areas of skin affected by the skin condition. One or more cropped images/image segments 106A, 106B (also referred to herein as “patches”) may be extracted from the one or more input images 102 based on the image segmentation data 106. Such image segments 106A, 106B may comprise sub-images/patches of the one or more images 102, e.g., images of dimension Hi x Wi, where Hi < H and Wi < W. In some examples, the image segments 106A, 106B have a fixed size, i.e., each segment has the same predefined size. In other examples, the image segments 106A, 106B can have different sizes. In some examples, the segmentation network 104 may directly output the one or more cropped images/image segments 106A, 106B. In some examples, the image segmentation data 106 comprises an image segmentation mask that identifies, for each pixel in the image, whether the pixel is part of an area potentially affected by the skin condition or not. Such a mask may, for example, be a H x W array of zeros and ones, with a one indicating that the pixel belongs to an area potentially affected by the skin condition and a zero indicating that the pixel does not belong to an area potentially affected by the skin condition. In some examples, one or more bounding boxes may be derived from the segmentation mask. In some examples, the image segmentation data 106 comprises data identifying one or more bounding boxes of segments 106A, 106B of the one or more images 102 that are potentially affected by the skin condition. For example, the image segmentation data 106 may comprise, for each bounding box, the locations (e.g., the pixel coordinates) of two pixels that form the opposite corners of a bounding box that is aligned with the image dimensions. In examples where the bounding box is not constrained to be aligned with the image dimensions, the image segmentation data 106 may comprise, for each bounding box, the locations (e.g., the pixel coordinates) or three or four corners of the bounding box. The one or more image segments 106A, 106B derived from the image segmentation data 106 are input into one or more (e.g., a plurality of) classification models 108. The one or
Attorney Docket No.46567-1742WO1 more image segments are, in some examples, scaled to a predefined size before input into the classification model 108 using, e.g., image upsampling (e.g., superresolution) and/or image downsampling techniques. Alternatively, the localized images 106a, 106B are divided into patches of a predetermined size, and the patches input into the one or more classification models 108. The one or more classification models 108 process the input image segments 106A, 106B based on a set of learned parameters of the classification model to generate one or more severity scores 110 indicative of a severity of the skin condition present in the one or more images 102. In some examples, each of the one or more classification models 108 produces a respective severity score 110 for an aspect of the skin condition that is associated with that classification model, e.g., the set of severity scores comprises a score from each of the classification models 108. In other examples, one or more of the classification models 108 may output a plurality of severity scores 110, each severity score associated with a different aspect of the skin condition. The one or more classification models 108 may comprise one or more neural network models. The neural network models may comprise one or more (e.g., a plurality of) neural network layers. For example, the one or more classification models 108 may comprise one or more convolutional neural networks (i.e., neural networks with one or more convolutional layers), e.g., be a multi-model CNN-based classifier. CNNs, with their capacity for hierarchical feature extraction from raw images, are widely used for multi-class image classification tasks. One or more of the neural network models may comprise one or more skip/residual connections. Examples of classification models 108 that may be used include, but are not limited to, ResNet and/or a Visual Geometry Group (VGG) image classification (VIC) model. The one or more classification models 108 may alternatively or additionally comprise one or more transformer models, i.e., models comprising one or more (e.g., a plurality of) transformer layers. For example, the one or more classification models may comprise a Vision Transformer (ViT) model. ViT models represent images as sequences of patches, using self-attention mechanisms to effectively capture global dependencies. In some examples, the one or more classification models 108 may comprise one or more hybrid models. Hybrid models combine convolutional and attention/transformer layers. Hybrid models balance local detail extraction with global context modelling.
Attorney Docket No.46567-1742WO1 The one or more classification models 108 may comprise a mixture of CNNs, transformers and/or hybrid models. For example, for EASI scores, transformer (e.g., ViT) models and hybrid models excel in predicting Erythema, Thickness, and Excoriation scores due to their ability to capture global context and subtle patterns, which are characteristic of these features. Conversely, CNNs perform better in predicting Lichenification scores, attributed to their strong local feature extraction capabilities, which are better suited for capturing the detailed textural nuances of thickened, leathery skin. The one or more classification models 108 have parameters that have been trained using machine-learning techniques on a set of training data comprising a plurality of images of skin patches that have a skin condition, each associated with a corresponding one or more ground truth severity scores for the skin condition. The one or more classification models 108 may be trained using an objective/ loss function that compares candidate severity scores output by the one or more classification models 108 to the ground truth severity scores, e.g., an absolute difference-based loss, such as an L2 loss, or a classification loss (e.g., for categorical scores). An optimisation routine, such as stochastic gradient descent, may be applied to the loss function to determine updates to parameters of the model. The severity scores 110 may comprise a plurality of severity scores that each relate to a different aspect of the skin condition. One or more of the severity scores 110 may be numerical scores in a respective predefined numerical range, e.g., a score out of five, a score out of ten, a percentage score or the like. One or more of the severity scores 110 may be a categorical score from a set of categories, e.g., the set {“mild”, “moderate”, “severe”} or the like. The severity scores 110 may comprise a plurality of individual scores, each associated with a different aspect of the skin condition. For example, the one or more severity scores 110 may comprise one or more of (e.g., all) of: an erythema score; an excoriation score; a Lichenification score; and/or a thickness score. For eczema, the one or severity scores may comprise one or more Eczema Area and Severity Index (EASI) scores, e.g., an Erythema score, an Excoriation score, a Thickness score and/or a Lichenification score. The one or more severity scores 110 may alternatively or additionally comprise an overall severity score for the skin condition. The overall severity score is, in some examples, output by the one or more classification models 108 directly. In other examples, the overall
Attorney Docket No.46567-1742WO1 severity score is derived from a combination of the individual scores output by the one or more classification models 108, e.g., using a set of predefined rules and/or an averaging process. In some examples, one or more severity scores 110 are generated by one or more classification models 108 for each patch/segment 106A, 106B of the input image 102. Each patch 106A, 106B may be individually input into the one or more classification model 108, which output a respective set of one or more severity scores 110 for each input patch 106A, 106B. The sets of severity scores for the patches can be combined/aggregated to generate a final set of one or more severity scores 110 for the image 102, e.g., by taking an average score for each score type, a maximum score or each score type, or the like. For example, two image patches 106A and 106B may be produced by the segmentation model 104. The first image patch 106A is input into a set of classification models, each specialising in a different eczema score type (e.g., an Erythema score, an Excoriation score, a Thickness score and Lichenification score). The set of classification models output a value for each of these score types for the first image patch 106A. The second image patch 106B is input into the set of classification models 108, either subsequently to the first image patch 106A or in parallel. The set of classification models output a value for each of the score types for the second image patch 106B. An overall score for each of the Erythema score, the Excoriation score, the Thickness score and the Lichenification score is generated by combining the respective scores from each image patch, e.g., by averaging the scores or taking the maximum score. The one or more severity scores 110 may be output to a user via a display of a user device, e.g., a smartphone, a tablet, a personal computer or the like. The one or more severity scores 110 may be output in an application performing at least a part of the method 100. The one or more severity scores 110 may be output as an overlay on the image 302. In some examples, the segmentation data 106 may also be used to identify, via the display, the parts 106A, 106B of the input image 102 affected by the skin condition. FIG.2A shows an example model structure for a machine-learning segmentation model. This example is based on a U-net architecture (described, for example, in “U-Net: Convolutional Networks for Biomedical Image Segmentation”, O. Ronnenberger et al., arXiv:1505.04597 (2015) the contents of which are incorporated herein by reference, in particular sections 2 and 3). The model 200 is a fully convolutional network.
Attorney Docket No.46567-1742WO1 The model 200 comprises a first section 204A (a “contracting” part) configured to downsample an input image 202 using a plurality of convolutional layers 206A-C to generate a plurality of feature maps 208. In the example shown, three sets of convolutional layers 206A-C are used, though in general additional convolutional layers may also be used. Each convolutional layer may comprise a plurality of convolutional filters followed by an activation function (e.g., a rectified linear unit) and a maxpooling operation. The model 200 further comprises a second section 204B (an “expansive” part) configured to upsample the plurality of feature maps 208 using a plurality of convolutional layers 210A-C to generate the model output 212, i.e., segmentation data. In the example shown, three sets of convolutional layers 210A-C are used, though in general additional convolutional layers may also be used. Each convolutional layer may comprise a plurality of convolutional filters followed by an upsampling operation. Alternatively, the upsampling operation may be implemented directly by the convolutional filters. The model 200 further comprises one or more skip/residual connections 214 that each connect a respective convolutional layer in the first section 204A to a corresponding convolutional layer in the second section 204B. The input to the corresponding layer in the second section 204B is thus the output of the previous layer of the second section 204B concatenated with the output of the layer in the first section 204A that is connected via the skip connection. This structure, i.e., a contracting path followed by an expansive path connected with skip connections, gives the model 200 a characteristic “u” shape, and hence models of this type may be referred to as “U-nets”. In medical applications, U-Net has demonstrated outstanding performance in tasks such as tumour detection and organ delineation. U-Net architectures, while effective for many segmentation tasks, can be limited by their reliance on local receptive fields, which is one inherent limitation of CNN based models. The use of transformer-based models, which incorporate multi-scale features, can address this limitation by allowing the model to capture both fine details and global contextual information. FIG.2B shows an example transformer-based structure for a machine-learning segmentation model 220. This example is based on a mixed transformer model, such as SegFormer (described, for example, in “SegFormer: Simple and Efficient Design for
Attorney Docket No.46567-1742WO1 Semantic Segmentation with Transformers”, E. Xie et al., arXiv: 2105.15203 (2021), the contents of which are incorporated herein by reference, particularly sections 3.1 and 3.2). The model 220 comprises an encoder section 222A and a decoder section 222B. The encoder section 222A comprises a plurality of transformer blocks 224A-N, e.g., four blocks, configured to extract feature maps 226A-N from an input image at a plurality of length scales, i.e., each transformer block is associated with a respective length scale. The input image may be divided into a plurality of patches 226 before input into the encoder section 222A, e.g., patches of size 4 x 4. Each transformer block 224A-N may comprise a self-attention mechanism and a feedforward and/or convolutional layer. The transfer blocks may further comprise an overlapped patch merging operation, for example as described in “SegFormer: Simple and Efficient Design for Semantic Segmentation with Transformers”. The set of multi-scale feature maps 226A-N is input into the decoder section 222B, which processes them to generate image segmentation data 228. The decoder section 222B may comprise one or more (e.g., a plurality) of multi-layer perceptron (MLP) layers 230A-M. The decoder section 222B may further comprise one or more upsampling operations. Examples of decoder structures are described in “SegFormer: Simple and Efficient Design for Semantic Segmentation with Transformers”. The use of transformer models can improve the segmentation performance over CNN-based approaches. Table I shows a comparison of the segmentation accuracy on a validation dataset for a CNN-based model and a Transformer-based model. FIG.3 shows an overview of an example system/method 300 for assessing the severity of a skin condition using machine-learning that utilises a large language model (LLM) 312 to generate a natural language description 318 of the severity scores 310. The method may be performed by one or more computing devices/systems, such as the device/system described in relation to FIG.7.
Attorney Docket No.46567-1742WO1 One or more images 302 of the skin of a subject are input into one or more machine- learning models 304. The one or more machine-learning models 304 processes the input one or more images 302 to generate one or more severity scores 310 for the skin condition present in the one or more images 302. The method 100 described in relation to FIG.1 may be used to generates the severity scores 310. In some alternative examples, the one or more machine- learning models 304 generate the severity scores 310 directly from the one or more images 302, i.e. the one or more machine-learning models 304 have been trained to generate the severity scores 310 directly from the one or more images 302. Other methods of generating the severity scores 310 using machine learning models 304 may alternatively be used. The one or more severity scores are input into a LLM 312. The LLM 312 uses the one or more severity scores 310 to retrieve one or more relevant document sections from one or more documents 314 relating to the skin condition that are stored in a database/document repository 316 accessible by the LLM 312. The LLM uses the severity scores 310 and the one or more retrieved document sections to generate a natural linage summary 318 of the severity of the skin condition. While numerical scores and segmented images provide quantitative data, integrating descriptive textual information can enhance the interpretability of diagnostic results for clinicians and patients. By leveraging the Retrieval-Augmented Generation (RAG) framework with a database of dermatological documents and, in some examples, patient data, the method 300 goes beyond mere description, providing actionable suggestions reflected by each patient's unique condition and needs. In some examples, an LLM is a machine-learning model with at least hundreds of millions of parameters. In some of those examples, the LLM includes at least billions of parameters, such as one hundred billion or more parameters. In some additional or alternative examples, an LLM is a sequence-to-sequence model, is Transformer-based, and/or can include an encoder and/or a decoder. An example of an LLM model is GPT4 from open AI. To retrieve the documents, the LLM 312 may, in some examples, generate a search request based on the severity scores 310. The search request may be a natural language search request and/or a database search request. In some examples, additional data input into the LLM 312, such as images or patches of images, may also be used to generate the search request. The search request may, in some examples, further be based on the data relating to
Attorney Docket No.46567-1742WO1 the subject to retrieve documents specific to the subject, e.g., patient histories, test results, biometric information, or the like. The search request is used by the database 316 to search the database 316 for relevant sections of documents in the set of documents 314 stored in the database 316. The relevant sections may comprise one or more snippets of a document in the database, e.g., relevant parts of one or more documents 314 in the database 316. The relevant sections may comprise one or more full documents in the database. The database 316 returns the retrieved one or more sections of the documents of text to the LLM 312. Input data comprising the one or more severity scores 310 and the one or more document sections is input into the LLM 312. The input data may be in the form of one or more LLM prompts. The LLM prompts may have a set format, with slots that are filled by the severity scores and the retrieved one or more document sections. such as an instruction to prepare an output based followed by the severity scores 310 and the retrieved document sections. For example, the input data may comprise the prompt “Please summarise the results of this severity assessment based on the following severity scores: [severity scores]; using the following documents: [list of text of document snippets]”. The LLM 312 processes the input data to generate a natural language summary 318 of the severity scores 310 that is conditioned on the retrieved document sections. The natural language summary 318 may comprise a description of what the score values mean based on descriptions of the score meanings in documents 314 stored in the database 316. In some examples, the natural language summary 318 may suggest a treatment based on treatments that are associated with the particular severity scores 310 that described in the set of documents 314. In some examples, the natural language summary 318 may suggest that the subject contact a health professional. The natural language summary 318 may be output to a user via a display of a user device, e.g., a smartphone, a tablet, a personal computer or the like. The natural language summary 318 may be output in an application performing at least a part of the method. The natural language summary 318 may be output as an overlay on the image 302. FIG.4 shows an example of a method for training a segmentation model 404 for localising skin conditions in images. The method may be performed by one or more computing devices/systems, such as the device/system described in relation to FIG.7.
Attorney Docket No.46567-1742WO1 The method 400 may be used to fine tune a pretrained image localisation model to specialise on segmenting images of skin conditions. Pre-trained models have demonstrated exceptional effectiveness across various domains. The utilization of these pre-existing models accelerates the development process and enhances model robustness by providing a solid foundation of learned features. Alternatively, the method 400 may be used to train a segmentation model 404 from scratch. The segmentation model 404 may have any of the structures described herein, for example those described in respect of FIGs.2A and 2B. The method 400 uses a training dataset 450 comprising a plurality of training examples 452. Each training example 452 comprises a training image 454 comprising an image of the skin of a subject and a corresponding set of ground truth segmentation data 456 indicating locations in the training image 454 affected by the skin condition (e.g., patches of skin affected by the skin condition). The ground truth segmentation data 456 may, for example, comprise a ground truth segmentation mask. Alternatively, the ground truth segmentation data 456 may, for example, comprise ground truth locations of bounding boxes for patches of skin affected by the skin condition. The training dataset 450 may comprise a plurality of “real” examples, e.g., images of skin captured from a subject. The training dataset 450 may further comprise a plurality of synthetic examples generated from one or more (e.g., a plurality) of the real training examples. Obtaining diverse and comprehensive training datasets 450 for dermatological conditions can be challenging due to privacy issues, variability in disease presentation, and the rarity of certain eczema types or stages. The use of synthetic samples can address this issue. Image-to-image translation models, such as generative adversarial networks (GANs) and Stable Diffusion models, can be used for data augmentation to synthetically expand the training dataset with realistic variations of skin condition (e.g., eczema) images, which can enhance robustness and generalization ability of the model 404 when trained on the data. For example, given a realistic skin image with certain eczema symptoms, a stable diffusion model can be prompted to generate one or more variants of the input image, such as an image of similar severity while appearing on different body areas, different skin types, or patients from different demographic groups.
Attorney Docket No.46567-1742WO1 Furthermore, in some examples, a series of image transformations may be applied to further augment the training dataset, e.g., resizing, random rotation, random horizontal flip, and/or centre cropping, followed by normalization. The training/fine-tuning method 400 comprises obtaining a training example 452 from the training dataset 450 and inputting the training image 454 of the training example 452 into a segmentation model 404. The segmentation model 404 processes the training image 454 based on current values of parameters (e.g., weights and/or biases) of the segmentation model 404 to generate candidate segmentation data 458. The candidate segmentation data 458 is indicative of one or more candidate locations 458A, 458B of skin in the training image 454 that is affected by the skin condition. For example, the candidate segmentation data 458 may, for example, be a candidate segmentation mask. Alternatively, the candidate segmentation data 458 may, for example, be candidate locations of bounding boxes for patches of skin affected by the skin condition. The candidate segmentation data 458 is compared to the corresponding ground truth segmentation data 456 from the training example 452 using a loss/objective function 460. The loss/objective function 460 may be any appropriate segmentation loss/objective function known in the art. For example, the loss/objective function 460 may compare the locations of corners of candidate bounding boxes to corresponding ground truth bounding box corner locations, e.g., using an L2 loss or Manhattan distance. The loss/objective function 460 may compare candidate pixel classifications in a candidate segmentation map to ground truth pixel classifications in a ground truth segmentation map using a classification loss. Based on the value of the loss/objective function 460, updates to the parameters of the model 404 are determined. An optimisation procedure may be applied to the loss/objective function 460 to determine the parameter updates, e.g., stochastic gradient descent/ascent. The method 400 may be iterated over the training dataset 450 until one or more threshold conditions are satisfied, e.g., a threshold number of training epochs or a threshold performance on a test/evaluation dataset. The severity models described herein may be trained in a similar manner, with the training dataset comprising a plurality of training examples, each comprising an image of a patch of skin affected by the skin condition and a corresponding set of one or more ground truth severity scores. Candidate severity scores output by the one or more severity models may be compares to corresponding ground truth severity scores in the set of ground truth
Attorney Docket No.46567-1742WO1 severity scores, e.g., using a loss/objective function, such as an L2 loss. The values of the loss/objective function are used to determine parameter updates for the one or more severity models, e.g., using an optimisation routine. For example, for both image segmentation model training in stage one and score prediction model training in stage two, the Adam optimizer may be used, with a learning rate of 0.00001 and weight decay of 0.00001 for model training. FIG.5 shows a flow diagram of an example method 500 for assessing the severity of a skin condition using machine-learning. The method 500 may be performed by one or more computers operating in one or more locations. For example, the method may be performed by the system/apparatus described in relation to FIG.7. At operation 502, one or more images of respective sections/patches of skin of a subject are received. The images may be colour images, such as RGB images, captured by an imaging device, such as digital camera, smartphone camera, tablet camera, or the like. At operation 504, a segmentation neural network is used to generate one or more image segments from the one or more images. The one or image segments each identify a potential location of skin in an image in the one or more images affected by the skin condition. The one or more images are input into the segmentation neural network, which processes them based on a set of learned parameters to determine image segmentation data. The segmentation neural network may comprise one or more of: one or more convolutional layers; one or more skip/residual connections; one or more fully connected layers; one or more transformer layers; and/or one or more multi-layer perceptrons. The segmentation neural network may be U-net. Alternatively, the segmentation neural network may be a vision transformer network. The segmentation neural network may output image segmentation data that identifies one or more sub-images/segments in the images that contain areas of skin potentially affected by the skin condition, e.g., areas of skin containing lesions, rashes, spots, blisters, moles or the like. The image segmentation data may comprise a segmentation mask. Alternatively, the image segmentation data may comprise data indicative of the locations of one or more bounding boxes of the sub-images/segments. The method may further comprise generating one or more cropped images from the image segmentation data and the one or more input images that correspond to the sub- images/segments identified by the segmentation neural network. The one or more cropped
Attorney Docket No.46567-1742WO1 images may be scaled to a predetermined size (e.g., a predetermined pixel height and/or width). At operation 506, the one or more image segments are processed using one or more severity models to generate one or more severity scores of the skin condition for the one or more images of respective sections of skin of the subject. The one or more severity scores may comprise an overall severity score. The one or more severity scores may comprise a plurality of severity scores, each associated with a different aspect of the skin condition. An overall severity score may be derived from the plurality of severity scores. In some examples, the skin condition is eczema. In such examples, the severity score may, for example, comprise an EASI score. Other eczema severity scores may alternatively or additionally be used. In some examples, the skin condition is a skin cancer. In such examples, the severity score may, for example, comprise a Glasgow 7-point checklist score. Other skin cancer severity scores may alternatively or additionally be used. In some examples, the skin condition is acne. In such examples, the severity score may, for example, comprise a Global Acne Grading System (GAGS) score. Other acne severity scores may alternatively or additionally be used. Many other examples of skin conditions and respective severity scores are possible. FIG.6 shows a flow diagram of an example method 600 for assessing the severity of a skin condition using machine-learning. The method 600 may be performed by one or more computers operating in one or more locations. For example, the method may be performed by the system/apparatus described in relation to FIG.7. At operation 602, one or more severity scores for a skin condition are generated from one or more images of sections of skin of a subject. The one or more severity scores may, for example, be generated using the method described in relation to FIGs.1 and/or 4. At operation 604, a natural language description of a severity of the skin condition from the one or more severity scores is generated using a large language model (LLM). In some examples, at least a part of the one or more images of sections of skin of a subject are also input into the LLM, e.g., segments of the one or more images generated, for example, as described in relation to operation 504 of FIG.5. In some examples, data relating to the subject is also input into the LLM, e.g.., a subject name and/or identifier.
Attorney Docket No.46567-1742WO1 Generating the natural language description of the severity of the skin condition using the LLM comprises, at operation 606, retrieving one or more document sections from one or more documents in a database based at least in part on the one or more severity scores. The one or more documents comprises one or more documents relating to the skin condition. The LLM may, in some examples, generate a search request, such as a natural language search request and/or a database search request, based on the severity scores and, if present, parts of the one or more images. The search request is used to search the database for relevant documents/snippets of documents. The search request may further be based on the data relating to the subject to retrieve documents specific to the subject, e.g., patient histories, test results, biometric information, or the like. At operation 608, input data comprising the one or more severity scores and the one or more document sections is input into the LLM. The input data may be in the form of one or more LLM prompts, such as an instruction to prepare an output based followed by the severity scores and the retrieved document sections. For example, the input data may comprise the prompt “Please summarise the results of this severity assessment based on the following severity scores and documents; [severity scores]; [text of document snippets]”. At operation 610, the natural language description of a severity of the skin condition is generated by the LLM based at least in part on the one or more severity scores and the one or more document sections. The natural language description is then output to a user (e.g., the subject and/or a healthcare professional) via a user interface, e.g., via an application on a user device, a program on a personal computer, or the like. The natural language description may include a recommended treatment for the skin condition. The natural language description may include a recommendation to contact a health professional. FIG.7 shows a schematic example of a system/apparatus 700 for performing any of the methods described herein. The system/apparatus shown is an example of a computing device. The apparatus 700 may be in the form of a personal computer, mobile computer device (such as a smart phone, tablet and/or laptop), or distributed computing system. The apparatus (or system) 700 comprises one or more processors 702. The one or more processors control operation of other components of the system/apparatus 700. The one or more processors 702 may, for example, comprise a general-purpose processor. The one or more processors 702 may be a single core device or a multiple core device. The one or more processors 702 may comprise a Central Processing Unit (CPU) or a graphical processing unit
Attorney Docket No.46567-1742WO1 (GPU). Alternatively, the one or more processors 702 may comprise specialised processing hardware, for instance a RISC processor or programmable hardware with embedded firmware. Multiple processors may be included. The system/apparatus comprises a working or volatile memory 704. The one or more processors may access the volatile memory 704 in order to process data and may control the storage of data in memory. The volatile memory 704 may comprise RAM of any type, for example, Static RAM (SRAM) or Dynamic RAM (DRAM), or it may comprise Flash memory, such as an SD-Card. The system/apparatus comprises a non-volatile memory 706. The non-volatile memory 706 stores a set of operation instructions 708 for controlling the operation of the processors 702 in the form of computer readable instructions. The non-volatile memory 706 may be a memory of any kind such as a Read Only Memory (ROM), a Flash memory or a magnetic drive memory. The one or more processors 702 are configured to execute operating instructions 708 to cause the system/apparatus to perform any of the methods described herein. The operating instructions 708 may comprise code (i.e. drivers) relating to the hardware components of the system/apparatus 700, as well as code relating to the basic operation of the system/apparatus 700. Generally speaking, the one or more processors 702 execute one or more instructions of the operating instructions 708, which are stored permanently or semi- permanently in the non-volatile memory 706, using the volatile memory 704 to store temporarily data generated during execution of said operating instructions 708. FIG.8 shows a comparison of the segmentation accuracy of different neural network models used for identifying Atopic Dermatitis (AD) regions. In particular, U-Net and SegFormer architectures are evaluated for identifying AD regions. The performance is measured across three metrics: Accuracy (%), Mean Intersection over Union (MIOC), and Dice Similarity Coefficient (GDS). The results show that SegFormer achieves higher accuracy across all metrics when compared to U-Net models of similar scales. For instance, SegFormer's mit-b0 model, despite being the lightest with only 3M parameters, achieves the highest accuracy at 91.50%, and also shows strong MIOC (0.49) and GDS (0.63) scores. This comparison suggests that SegFormer, particularly its lighter variants, provides an effective and efficient solution for identifying AD regions.
Attorney Docket No.46567-1742WO1 FIG.9 shows the performance of several machine learning models in predicting the Eczema Area and Severity Index (EASI) scores, including the accuracy in predicting the Erythema, Thickness, Excoriation, and Lichenification scores. The plots demonstrate distinct performance trends across the different EASI scores and model types. For Erythema, Thickness, and Excoriation, ViT-based models and hybrid models generally show higher accuracy compared to CNN-based models, particularly when comparing models of similar scales. This suggests that the capacity of ViT and hybrid models to capture global context and subtle patterns is beneficial for predicting these specific EASI features. For the Lichenification score, CNN-based models tend to exhibit better accuracy. This is attributed to CNNs' strong local feature extraction capabilities, which are more suited for the detailed textural nuances characteristic of lichenified skin. Any mentioned apparatus and/or other features of particular mentioned apparatus may be provided by apparatus arranged such that they become configured to carry out the desired operations only when enabled, e.g. switched on, or the like. In such cases, they may not necessarily have the appropriate software loaded into the active memory in the non- enabled (e.g. switched off state) and only load the appropriate software in the enabled (e.g. on state). The apparatus may comprise hardware circuitry and/or firmware. The apparatus may comprise software loaded onto memory. Such software/computer programs may be recorded on the same memory/processor/functional units and/or on one or more memories/processors/ functional units. Any mentioned apparatus/circuitry/elements/processor may have other functions in addition to the mentioned functions, and that these functions may be performed by the same apparatus/circuitry/elements/processor. One or more disclosed aspects may encompass the electronic distribution of associated computer programs and computer programs (which may be source/transport encoded) recorded on an appropriate carrier (e.g. memory, signal). Any “computer” described herein can comprise a collection of one or more individual processors/processing elements that may or may not be located on the same circuit board, or the same region/position of a circuit board or even the same device. In some examples one or more of any mentioned processors may be distributed over a plurality of devices. The same or different processor/processing elements may perform one or more functions described herein.
Attorney Docket No.46567-1742WO1 The term “signalling” may refer to one or more signals transmitted as a series of transmitted and/or received electrical/optical signals. The series of signals may comprise one, two, three, four or even more individual signal components or distinct signals to make up said signalling. Some or all of these individual signals may be transmitted/received by wireless or wired communication simultaneously, in sequence, and/or such that they temporally overlap one another. With reference to any discussion of any mentioned computer and/or processor and memory (e.g. including ROM, CD-ROM etc.), these may comprise a computer processor, Application Specific Integrated Circuit (ASIC), field-programmable gate array (FPGA), and/or other hardware components that have been programmed in such a way to carry out the inventive function. The applicant hereby discloses in isolation each individual feature described herein and any combination of two or more such features, to the extent that such features or combinations are capable of being carried out based on the present specification as a whole, in the light of the common general knowledge of a person skilled in the art, irrespective of whether such features or combinations of features solve any problems disclosed herein, and without limitation to the scope of the claims. The applicant indicates that the disclosed aspects/examples may consist of any such individual feature or combination of features. In view of the foregoing description, it will be evident to a person skilled in the art that various modifications may be made within the scope of the disclosure. While there have been shown and described and pointed out fundamental novel features as applied to examples thereof, it will be understood that various omissions and substitutions and changes in the form and details of the devices and methods described may be made by those skilled in the art without departing from the scope of the disclosure. For example, it is expressly intended that all combinations of those elements and/or method steps which perform substantially the same function in substantially the same way to achieve the same results are within the scope of the disclosure. Moreover, it should be recognized that structures and/or elements and/or method steps shown and/or described in connection with any disclosed form or examples may be incorporated in any other disclosed or described or suggested form or example as a general matter of design choice. Furthermore, in the claims means-plus-function clauses are intended to cover the structures described herein as performing the recited function and not only structural equivalents, but also equivalent structures.
Attorney Docket No.46567-1742WO1 Implementations of the methods described herein may be realised as in digital electronic circuitry, integrated circuitry, specially designed ASICs (application specific integrated circuits), computer hardware, firmware, software, and/or combinations thereof. These may include computer program products (such as software stored on e.g. magnetic discs, optical disks, memory, Programmable Logic Devices) comprising computer readable instructions that, when executed by a computer, such as that described in relation to Figure 7, cause the computer to perform one or more of the methods described herein. Any system feature as described herein may also be provided as a method feature, and vice versa. As used herein, means plus function features may be expressed alternatively in terms of their corresponding structure. In particular, method aspects may be applied to system aspects, and vice versa. Furthermore, any, some and/or all features in one aspect can be applied to any, some and/or all features in any other aspect, in any appropriate combination. It should also be appreciated that particular combinations of the various features described and defined in any aspects of the invention can be implemented and/or supplied and/or used independently. Although several embodiments have been shown and described, it would be appreciated by those skilled in the art that changes may be made in these embodiments without departing from the principles of this disclosure, the scope of which is defined in the claims and their equivalents. The terms “drug” or “medicament” are used synonymously herein and describe a pharmaceutical formulation containing one or more active pharmaceutical ingredients or pharmaceutically acceptable salts or solvates thereof, and optionally a pharmaceutically acceptable carrier. An active pharmaceutical ingredient (“API”), in the broadest terms, is a chemical structure that has a biological effect on humans or animals. In pharmacology, a drug or medicament is used in the treatment, cure, prevention, or diagnosis of disease or used to otherwise enhance physical or mental well-being. A drug or medicament may be used for a limited duration, or on a regular basis for chronic disorders. As described below, a drug or medicament can include at least one API, or combinations thereof, in various types of formulations, for the treatment of one or more diseases. Examples of API may include small molecules having a molecular weight of 500 Da or less; polypeptides, peptides, and proteins (e.g., hormones, growth factors, antibodies, antibody fragments, and enzymes); carbohydrates and polysaccharides; and nucleic acids,
Attorney Docket No.46567-1742WO1 double or single stranded DNA (including naked and cDNA), RNA, antisense nucleic acids such as antisense DNA and RNA, small interfering RNA (siRNA), ribozymes, genes, and oligonucleotides. Nucleic acids may be incorporated into molecular delivery systems such as vectors, plasmids, or liposomes. Mixtures of one or more drugs are also contemplated. The drug or medicament may be contained in a primary package or “drug container” adapted for use with a drug delivery device. The drug container may be, e.g., a cartridge, syringe, reservoir, or other solid or flexible vessel configured to provide a suitable chamber for storage (e.g., short- or long-term storage) of one or more drugs. For example, in some instances, the chamber may be designed to store a drug for at least one day (e.g., 1 to at least 30 days). In some instances, the chamber may be designed to store a drug for about 1 month to about 2 years. Storage may occur at room temperature (e.g., about 20°C), or refrigerated temperatures (e.g., from about - 4°C to about 4°C). In some instances, the drug container may be or may include a dual-chamber cartridge configured to store two or more components of the pharmaceutical formulation to-be-administered (e.g., an API and a diluent, or two different drugs) separately, one in each chamber. In such instances, the two chambers of the dual-chamber cartridge may be configured to allow mixing between the two or more components prior to and/or during dispensing into the human or animal body. For example, the two chambers may be configured such that they are in fluid communication with each other (e.g., by way of a conduit between the two chambers) and allow mixing of the two components when desired by a user prior to dispensing. Alternatively or in addition, the two chambers may be configured to allow mixing as the components are being dispensed into the human or animal body. The drugs or medicaments contained in the drug delivery devices as described herein can be used for the treatment and/or prophylaxis of many different types of medical disorders. Examples of disorders include, e.g., diabetes mellitus or complications associated with diabetes mellitus such as diabetic retinopathy, thromboembolism disorders such as deep vein or pulmonary thromboembolism. Further examples of disorders are acute coronary syndrome (ACS), angina, myocardial infarction, cancer, macular degeneration, inflammation, hay fever, atherosclerosis and/or rheumatoid arthritis. Examples of APIs and drugs are those as described in handbooks such as Rote Liste 2014, for example, without limitation, main groups 12 (anti-diabetic drugs) or 86 (oncology drugs), and Merck Index, 15th edition. Examples of APIs for the treatment and/or prophylaxis of type 1 or type 2 diabetes mellitus or complications associated with type 1 or type 2 diabetes mellitus include an
Attorney Docket No.46567-1742WO1 insulin, e.g., human insulin, or a human insulin analogue or derivative, a glucagon-like peptide (GLP-1), GLP-1 analogues or GLP-1 receptor agonists, or an analogue or derivative thereof, a dipeptidyl peptidase-4 (DPP4) inhibitor, or a pharmaceutically acceptable salt or solvate thereof, or any mixture thereof. As used herein, the terms “analogue” and “derivative” refers to a polypeptide which has a molecular structure which formally can be derived from the structure of a naturally occurring peptide, for example that of human insulin, by deleting and/or exchanging at least one amino acid residue occurring in the naturally occurring peptide and/or by adding at least one amino acid residue. The added and/or exchanged amino acid residue can either be codeable amino acid residues or other naturally occurring residues or purely synthetic amino acid residues. Insulin analogues are also referred to as "insulin receptor ligands". In particular, the term “derivative” refers to a polypeptide which has a molecular structure which formally can be derived from the structure of a naturally occurring peptide, for example that of human insulin, in which one or more organic substituent (e.g. a fatty acid) is bound to one or more of the amino acids. Optionally, one or more amino acids occurring in the naturally occurring peptide may have been deleted and/or replaced by other amino acids, including non-codeable amino acids, or amino acids, including non-codeable, have been added to the naturally occurring peptide. Examples of insulin analogues are Gly(A21), Arg(B31), Arg(B32) human insulin (insulin glargine); Lys(B3), Glu(B29) human insulin (insulin glulisine); Lys(B28), Pro(B29) human insulin (insulin lispro); Asp(B28) human insulin (insulin aspart); human insulin, wherein proline in position B28 is replaced by Asp, Lys, Leu, Val or Ala and wherein in position B29 Lys may be replaced by Pro; Ala(B26) human insulin; Des(B28-B30) human insulin; Des(B27) human insulin and Des(B30) human insulin. Examples of insulin derivatives are, for example, B29-N-myristoyl-des(B30) human insulin, Lys(B29) (N- tetradecanoyl)-des(B30) human insulin (insulin detemir, Levemir®); B29-N-palmitoyl-des(B30) human insulin; B29-N-myristoyl human insulin; B29-N- palmitoyl human insulin; B28-N-myristoyl LysB28ProB29 human insulin; B28-N-palmitoyl- LysB28ProB29 human insulin; B30-N-myristoyl-ThrB29LysB30 human insulin; B30-N- palmitoyl- ThrB29LysB30 human insulin; B29-N-(N-palmitoyl-gamma-glutamyl)-des(B30) human insulin, B29-N-omega-carboxypentadecanoyl-gamma-L-glutamyl-des(B30) human insulin (insulin degludec, Tresiba®); B29-N-(N-lithocholyl-gamma-glutamyl)-des(B30) human insulin; B29-N-(ω-carboxyheptadecanoyl)-des(B30) human insulin and B29-N-(ω- carboxyheptadecanoyl) human insulin.
Attorney Docket No.46567-1742WO1 Examples of GLP-1, GLP-1 analogues and GLP-1 receptor agonists are, for example, Lixisenatide (Lyxumia®), Exenatide (Exendin-4, Byetta®, Bydureon®, a 39 amino acid peptide which is produced by the salivary glands of the Gila monster), Liraglutide (Victoza®), Semaglutide, Taspoglutide, Albiglutide (Syncria®), Dulaglutide (Trulicity®), rExendin-4, CJC-1134-PC, PB-1023, TTP-054, Langlenatide / HM-11260C (Efpeglenatide), HM-15211, CM-3, GLP-1 Eligen, ORMD-0901, NN-9423, NN-9709, NN-9924, NN-9926, NN-9927, Nodexen, Viador-GLP-1, CVX-096, ZYOG-1, ZYD-1, GSK-2374697, DA-3091, MAR-701, MAR709, ZP-2929, ZP-3022, ZP-DI-70, TT-401 (Pegapamodtide), BHM-034. MOD-6030, CAM-2036, DA-15864, ARI-2651, ARI-2255, Tirzepatide (LY3298176), Bamadutide (SAR425899), Exenatide-XTEN and Glucagon-Xten. An example of an oligonucleotide is, for example: mipomersen sodium (Kynamro®), a cholesterol-reducing antisense therapeutic for the treatment of familial hypercholesterolemia or RG012 for the treatment of Alport syndrome. Examples of DPP4 inhibitors are Linagliptin, Vildagliptin, Sitagliptin, Denagliptin, Saxagliptin, Berberine. Examples of hormones include hypophysis hormones or hypothalamus hormones or regulatory active peptides and their antagonists, such as Gonadotropine (Follitropin, Lutropin, Choriongonadotropin, Menotropin), Somatropine (Somatropin), Desmopressin, Terlipressin, Gonadorelin, Triptorelin, Leuprorelin, Buserelin, Nafarelin, and Goserelin. Examples of polysaccharides include a glucosaminoglycane, a hyaluronic acid, a heparin, a low molecular weight heparin or an ultra-low molecular weight heparin or a derivative thereof, or a sulphated polysaccharide, e.g. a poly-sulphated form of the above- mentioned polysaccharides, and/or a pharmaceutically acceptable salt thereof. An example of a pharmaceutically acceptable salt of a poly-sulphated low molecular weight heparin is enoxaparin sodium. An example of a hyaluronic acid derivative is Hylan G-F 20 (Synvisc®), a sodium hyaluronate. The term “antibody”, as used herein, refers to an immunoglobulin molecule or an antigen-binding portion thereof. Examples of antigen-binding portions of immunoglobulin molecules include F(ab) and F(ab')2 fragments, which retain the ability to bind antigen. The antibody can be polyclonal, monoclonal, recombinant, chimeric, de-immunized or humanized, fully human, non-human, (e.g., murine), or single chain antibody. In some embodiments, the antibody has effector function and can fix complement. In some
Attorney Docket No.46567-1742WO1 embodiments, the antibody has reduced or no ability to bind an Fc receptor. For example, the antibody can be an isotype or subtype, an antibody fragment or mutant, which does not support binding to an Fc receptor, e.g., it has a mutagenized or deleted Fc receptor binding region. The term antibody also includes an antigen-binding molecule based on tetravalent bispecific tandem immunoglobulins (TBTI) and/or a dual variable region antibody-like binding protein having cross-over binding region orientation (CODV). The terms “fragment” or “antibody fragment” refer to a polypeptide derived from an antibody polypeptide molecule (e.g., an antibody heavy and/or light chain polypeptide) that does not comprise a full-length antibody polypeptide, but that still comprises at least a portion of a full-length antibody polypeptide that is capable of binding to an antigen. Antibody fragments can comprise a cleaved portion of a full length antibody polypeptide, although the term is not limited to such cleaved fragments. Antibody fragments that are useful in the present invention include, for example, Fab fragments, F(ab')2 fragments, scFv (single-chain Fv) fragments, linear antibodies, monospecific or multispecific antibody fragments such as bispecific, trispecific, tetraspecific and multispecific antibodies (e.g., diabodies, triabodies, tetrabodies), monovalent or multivalent antibody fragments such as bivalent, trivalent, tetravalent and multivalent antibodies, minibodies, chelating recombinant antibodies, tribodies or bibodies, intrabodies, nanobodies, small modular immunopharmaceuticals (SMIP), binding-domain immunoglobulin fusion proteins, camelized antibodies, and VHH containing antibodies. Additional examples of antigen- binding antibody fragments are known in the art. The terms “Complementarity-determining region” or “CDR” refer to short polypeptide sequences within the variable region of both heavy and light chain polypeptides that are primarily responsible for mediating specific antigen recognition. The term “framework region” refers to amino acid sequences within the variable region of both heavy and light chain polypeptides that are not CDR sequences, and are primarily responsible for maintaining correct positioning of the CDR sequences to permit antigen binding. Although the framework regions themselves typically do not directly participate in antigen binding, as is known in the art, certain residues within the framework regions of certain antibodies can directly participate in antigen binding or can affect the ability of one or more amino acids in CDRs to interact with antigen. Examples of antibodies are anti PCSK-9 mAb (e.g., Alirocumab), anti IL-6 mAb (e.g., Sarilumab), and anti IL-4 mAb (e.g., Dupilumab).
Attorney Docket No.46567-1742WO1 Pharmaceutically acceptable salts of any API described herein are also contemplated for use in a drug or medicament in a drug delivery device. Pharmaceutically acceptable salts are for example acid addition salts and basic salts. Those of skill in the art will understand that modifications (additions and/or removals) of various components of the APIs, formulations, apparatuses, methods, systems, and embodiments described herein may be made without departing from the full scope and spirit of the present invention, which encompass such modifications and any and all equivalents thereof. An example drug delivery device may involve a needle-based injection system as described in Table 1 of section 5.2 of ISO 11608-1:2014(E). As described in ISO 11608- 1:2014(E), needle-based injection systems may be broadly distinguished into multi-dose container systems and single-dose (with partial or full evacuation) container systems. The container may be a replaceable container or an integrated non-replaceable container. As further described in ISO 11608-1:2014(E), a multi-dose container system may involve a needle-based injection device with a replaceable container. In such a system, each container holds multiple doses, the size of which may be fixed or variable (pre-set by the user). Another multi-dose container system may involve a needle-based injection device with an integrated non-replaceable container. In such a system, each container holds multiple doses, the size of which may be fixed or variable (pre-set by the user). As further described in ISO 11608-1:2014(E), a single-dose container system may involve a needle-based injection device with a replaceable container. In one example for such a system, each container holds a single dose, whereby the entire deliverable volume is expelled (full evacuation). In a further example, each container holds a single dose, whereby a portion of the deliverable volume is expelled (partial evacuation). As also described in ISO 11608-1:2014(E), a single-dose container system may involve a needle-based injection device with an integrated non-replaceable container. In one example for such a system, each container holds a single dose, whereby the entire deliverable volume is expelled (full evacuation). In a further example, each container holds a single dose, whereby a portion of the deliverable volume is expelled (partial evacuation).