Methods and systems for training clinical diagnostic models

Abstract

This disclosure provides a novel approach to training clinical diagnostic models using synthetic data generated by an artificial intelligence (Al) model. The disclosed models demonstrate several advantages, including: reducing time, cost, and complexity associated with traditional data collection and labeling. Additionally, the disclosed models address privacy and data management issues and allow for the inclusion of rare or novel cases in the training set.

Description

[0001] METHODS AND SYSTEMS FOR TRAINING CLINICAL DIAGNOSTIC MODELS

[0002] CROSS REFERENCE TO RELATED APPLICATION

[0003] This application claims priority under 35 U.S.C. § 119(e) to U.S. Provisional Patent Application No. 63 / 626,986, filed on January 30, 2024, the disclosure of which is incorporated herein by reference in its entirety.

[0004] FIELD OF THE DISCLOSURE

[0005] This disclosure relates generally to methods and systems fortraining clinical diagnostic models.

[0006] BACKGROUND OF THE DISCLOSURE

[0007] Development of machine learning algorithms lays a foundation for successful application of artificial intelligence technologies in many application scenarios. However, all current machine learning algorithms, especially a deep learning method, depend on a large amount of training data for training a parameter in a model. However, in many scenarios, how to obtain training data becomes a difficult problem. For example, test result data in large quantities from instruments in the field or laboratory has been required to train such a model. Harvesting test result data from clinical diagnostic instrumentation is expensive, timeconsuming, and fraught with patient data privacy restraints that limit downloading or uploading this data over the internet.

[0008] Therefore, there is a current need to develop new methods and systems for training clinical diagnostic models.

[0009] SUMMARY OF THE DISCLOSURE

[0010] This disclosure addresses the current need mentioned above in a number of aspects. In one aspect, this disclosure provides a method for immunohematology analysis using a machine learning model trained by synthetic image data.

[0011] In some embodiments, the method comprises: (a) generating a training dataset comprising synthetic images by a first machine learning model, wherein the synthetic images contain no personally identifiable information of any person; (b) training a second machine learning model based on the training dataset; (c) determining, based at least on a classification result of the training dataset by the second machine learning, that the second machine learning model is able to classify a threshold quantity of images in the training dataset; (d) inputting into the second machine learning model image data comprising one or more images obtained from a sample of a patient; and (e) classifying by the second machine learning model the image data based at least on one or more characteristics of the sample of the patient, wherein one or more characteristics comprise blood type, crossmatching, rhesus (Rh) factor D, or a combination thereof.

[0012] In some embodiments, the first machine learning model comprises Stable Diffusion.

[0013] In some embodiments, the method comprises generating the training dataset by an image2image Stable Diffusion process.

[0014] In some embodiments, a ratio of seed images to the synthetic images ranges from zero to one.

[0015] In some embodiments, an average ratio of the seed images to the synthetic images is about 0.014.

[0016] In some embodiments, the method comprises denoising the synthetic images with a denoising parameter ranging from zero to one.

[0017] In some embodiments, the method comprises denoising the synthetic images with a denoising parameter of about 0.17.

[0018] In some embodiments, the second machine learning model comprises a convolutional neural network model (e.g., deep convolutional neural network model).

[0019] In some embodiments, the step of classifying comprises a sequential model architecture comprising layers for data augmentation, rescaling, convolution, max pooling, dropout, and dense layers.

[0020] In some embodiments, the one or more images are obtained from column agglutinated gel cards.

[0021] In some embodiments, the method comprises training the second machine learning model with agglutination reaction gradings comprising: 1+ agglutination, 2+ agglutination, 3+ agglutination, and 4+ agglutination.

[0022] In some embodiments, the immunohematology analysis comprises forward-blood typing, reverse-blood typing, or antibody screening.

[0023] In some embodiments, the sample comprises a blood sample. In another aspect, this disclosure provides a system for immunohematology analysis using a machine learning model trained by synthetic image data.

[0024] In some embodiments, the system comprises one or more processors configured to: (i) generate a training dataset comprising synthetic images by a first machine learning model, wherein the synthetic images contain no personally identifiable information of any person; (ii) train a second machine learning model based on the training dataset; (iii) determine, based at least on a classification result of the training dataset by the second machine learning, that the second machine learning model is able to classify a threshold quantity of images in the training dataset; (iv) input into the second machine learning model image data comprising one or more images obtained from a sample of a patient; and (v) classify by the second machine learning model the image data based at least on one or more characteristics of the sample of the patient, wherein one or more characteristics comprise blood type, crossmatching, rhesus (Rh) factor D, or a combination thereof.

[0025] In some embodiments, the first machine learning model comprises Stable Diffusion. In some embodiments, the one or more processors are further configured to generate the training dataset by an image2image Stable Diffusion process.

[0026] In some embodiments, wherein an average ratio of seed images to the synthetic images ranges from zero to one.

[0027] In some embodiments, wherein an average ratio of seed images to the synthetic images is about 0.014.

[0028] In some embodiments, the one or more processors are further configured to denoise the synthetic images with a denoising parameter ranging from zero to one.

[0029] In some embodiments, the one or more processors are further configured to denoise the synthetic images with a denoising parameter of about 0.17.

[0030] In some embodiments, the second machine learning model comprises a convolutional neural network model (e.g., deep convolutional neural network model).

[0031] In some embodiments, the step of classifying comprises a sequential model architecture comprising layers for data augmentation, rescaling, convolution, max pooling, dropout, and dense layers.

[0032] In some embodiments, the one or more images are obtained from column agglutinated gel cards. In some embodiments, wherein the one or more processors are configured to train the second machine learning model with agglutination reaction gradings comprising: 1+ agglutination, 2+ agglutination, 3+ agglutination, and 4+ agglutination.

[0033] In some embodiments, the immunohematology analysis comprises forward-blood typing, reverse-blood typing, or antibody screening. In some embodiments, the sample comprises a blood sample.

[0034] The foregoing summary is not intended to define every aspect of the disclosure, and additional aspects are described in other sections, such as the following detailed description. The entire document is intended to be related as a unified disclosure, and it should be understood that all combinations of features described herein are contemplated, even if the combination of features are not found together in the same sentence, paragraph, or section of this document. Other features and advantages of the disclosure will become apparent from the following detailed description. It should be understood, however, that the detailed description and the specific examples, while indicating specific embodiments of the disclosure, are given by way of illustration only, because various changes and modifications within the spirit and scope of the disclosure will become apparent to those skilled in the art from this detailed description.

[0035] BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 illustrates an example process to train an IH-500 Results Model using synthetic images generated by an artificial intelligence (Al) model.

[0037] Figure 2 illustrates a comparison between the conventional model training and the model training using synthetic image data according to various embodiments of the present disclosure.

[0038] Figure 3 illustrates sample synthetic images in the Al-generated training set.

[0039] Figure 4 illustrates sample training outputs.

[0040] Figures 5A and 5B illustrate the performance of the classification model according to various embodiments of the present disclosure. Figure 5 A shows that the model achieved 1.0 accuracy across all 4 classes with 8 images of real Gel Card Agglutinations. Figure 5B shows a comparison of the accuracy and efficiency of the existing models and the model according to various embodiments of the present disclosure.

[0041] Figure 6 illustrates an example computing system for implementing the disclosed methods. DETAILED DESCRIPTION OF THE DISCLOSURE

[0042] This disclosure describes a novel approach to training clinical diagnostic models using synthetic data generated by an artificial intelligence (Al) model. The disclosed models demonstrate several advantages, including reducing time, cost, and complexity associated with traditional data collection and labeling. Additionally, the disclosed models address privacy and data management issues and allow for the inclusion of rare or novel cases in the training set.

[0043] In one aspect, this disclosure provides a novel method for immunohematology analysis using a machine learning model trained by synthetic image data. In some embodiments, the method comprises: (a) generating a training dataset comprising synthetic images by a first machine learning model, wherein the synthetic images contain no personally identifiable information of any person; (b) training a second machine learning model based on the training dataset; (c) determining, based at least on a classification result of the training dataset by the second machine learning, that the second machine learning model is able to classify a threshold quantity of images in the training dataset; (d) inputting into the second machine learning model image data comprising one or more images obtained from a sample of a patient; and (e) classifying by the second machine learning model the image data based at least on one or more characteristics of the sample of the patient, such as blood type, crossmatching, rhesus (Rh) factor D, or a combination thereof.

[0044] As used herein, a “machine learning model,” a “model,” or a “classifier” refers to a set of algorithmic routines and parameters that can predict an output(s) for a process input based on a set of input features, with or without being explicitly programmed. The structure of software routines (e.g., the number of subroutines and the relation between them) and / or the values of the parameters can be determined in a training process, which can use actual results of the process that is being modeled. Such systems or models are understood to be necessarily rooted in computer technology and cannot be implemented or even exist in the absence of computing technology. While machine learning systems utilize various types of statistical analyses, machine learning systems are distinguished from statistical analyses by virtue of the ability to learn without explicit programming and being rooted in computer technology. A neural network or an artificial neural network is one set of algorithms used in machine learning for modeling the data using graphs of neurons. Any network structure may be used. Any number of layers, nodes within layers, types of nodes (activations), types of layers, interconnections, learnable parameters, and / or other network architectures may be used. Machine training uses the defined architecture, training data, and optimization to learn values of the learnable parameters of the architecture based on the samples and ground truth of training data.

[0045] A typical machine learning pipeline may include building a machine learning model from a sample dataset (referred to as a “training set”), evaluating the model against one or more additional sample datasets (referred to as a “validation set” and / or a “test set”) to decide whether to keep the model and to benchmark how good the model is, and using the model in “production” to make predictions or decisions against live input data captured by an application service. For training the model to be applied as a machine-learned model, training data is acquired and stored in a database or memory. The training data is acquired by aggregation, mining, and loading from a publicly or privately formed collection, transfer, and / or access. Tens, hundreds, or thousands of samples of training data are acquired. The samples are from scans of different patients and / or phantoms. Simulation may be used to form the training data. The training data includes the desired output (ground truth), such as segmentation, and the input, such as protocol data and imaging data.

[0046] In some embodiments, the training set is used to create a single classifier using any now or hereafter-known methods. In other embodiments, a plurality of training sets are created to generate a plurality of corresponding classifiers. Each of the plurality of classifiers can be generated based on the same or different learning algorithm(s) that utilize(s) the same or different features in the corresponding one of the pluralities of training sets.

[0047] Once trained, the machine-learned or trained classifier is stored for later application. The training determines the values of the learnable parameters of the network. The network architecture, values of non-learnable parameters, and values of the learnable parameters are stored as the machine-learned network. Once stored, the machine-learned network may be fixed. The same machine-learned network may be applied to different patients, different scanners, and / or with different imaging protocols for the scanning. The machine-learned network may be updated. As additional training data is acquired, such as through application of the network for patients and corrections by experts to that output, the additional training data may be used to re-train or update the training.

[0048] In some embodiments, the machine learning model may include a supervised learning model. Supervised learning models may include different approaches and algorithms, including analytical learning, artificial neural network, backpropagation, boosting (meta- algorithm), Bayesian statistics, case-based reasoning, decision tree learning, inductive logic programming, Gaussian process regression, genetic programming, group method of data handling, kernel estimators, learning automata, learning classifier systems, minimum message length (decision trees, decision graphs, etc.), multilinear subspace learning, naive Bayes classifier, maximum entropy classifier, conditional random field, Nearest Neighbor Algorithm, probably approximately correct learning (PAC) learning, ripple down rules, a knowledge acquisition methodology, symbolic machine learning algorithms, subsymbolic machine learning algorithms, support vector machines, Minimum Complexity Machines (MCM), random forests, ensembles of classifiers, ordinal classification, data pre-processing, handling imbalanced datasets, statistical relational learning, or Proaftn, a multicriteria classification algorithm, linear regression, logistic regression, deep recurrent neural network (e.g., long short term memory, LSTM), Bayes classifier, hidden Markov model (HMM), linear discriminant analysis (LDA), k-means clustering, density -based spatial clustering of applications with noise (DBSCAN), random forest algorithm, support vector machine (SVM), and / or any model described herein.

[0049] In some embodiments, the classifier may include a supervised or unsupervised Machine Learning or Deep Learning algorithm, Logistic Regression, Naive Bayes, Support Vector Machine, Decision Tree, Random Forest, Gradient Boosting, Regularizing Gradient Boosting, K-Nearest Neighbors, a continuous regression approach, Ridge Regression, Kernel Ridge Regression, Support Vector Regression, deep learning approach, Neural Networks, Convolutional Neural Network (CNNs), Recurrent Neural Networks (RNNs), Gated Recurrent Units (GRUs), Long Short Term Memory Networks (LSTMs), Generative Models, Generative Adversarial Networks (GANs), Deep Belief Networks (DBNs), Feedforward Neural Networks, Autoencoders, Variational Autoencoders, Normalizing Flow Models, Deniosing Diffusion Probabilistic Models (DDPMs), Score Based Generative Models (SGMs), Radial Basis Function Networks (RBFNs), Multilayer Perceptrons (MLPs), Stochastic Neural Networks, and / or any combination thereof.

[0050] In some embodiments, the model may include a convolutional neural network (CNN). The CNN may include a set of convolutional filters configured to filter the first plurality of data structures and, optionally, the second plurality of data structures. The filter may be any filter described herein. The number of filters for each layer may be from 10 to 20, 20 to 30, 30 to 40, 40 to 50, 50 to 60, 60 to 70, 70 to 80, 80 to 90, 90 to 100, 100 to 150, 150 to 200, or more. The kernel size for the filters can be 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, from 15 to 20, from 20 to 30, from 30 to 40, or more. The CNN may include an input layer configured to receive the filtered first plurality of data structures and, optionally, the filtered second plurality of data structures. The CNN may also include a plurality of hidden layers, including a plurality of nodes. The first layer of the plurality of hidden layers is coupled to the input layer. The CNN may further include an output layer coupled to a last layer of the plurality of hidden layers and configured to output an output data structure. The output data structure may include the properties.

[0051] In some embodiments, the first machine learning model comprises Stable Diffusion. Stable Diffusion is a deep learning, text-to-image model based on diffusion techniques. It can be used to generate detailed images conditioned on text descriptions, though it can also be applied to other tasks such as inpainting, outpainting, and generating image-to-image translations guided by a text prompt. Stable Diffusion is a latent diffusion model, a kind of deep generative artificial neural network. Its code and model weights have been open-sourced, and it can run on most consumer hardware equipped with a modest GPU with at least 4 GB of VRAM.

[0052] In some embodiments, the method comprises generating the training dataset by an image2image Stable Diffusion process. Stable Diffusion’s image-to-image (Img2Img) feature can create new images based on existing ones. The process begins when the encoder transforms input images into latent representations. The colors and composition of the input image act as a guiding force.

[0053] In some embodiments, the average ratio of seed images to the synthetic images ranges from zero to one, encompassing any real number with arbitrary precision as constrained by computer resources. In some embodiments, the average ratio of seed images to synthetic images is about 0.014.

[0054] In some embodiments, the method comprises denoising the synthetic images with a denoising parameter ranging from zero to one, encompassing any real number with arbitrary precision as constrained by computer resources. In some embodiments, the method comprises denoising the synthetic images with a denoising parameter of about 0.17.

[0055] In some embodiments, the second machine learning model comprises a convolutional neural network model (e.g., deep convolutional neural network model).

[0056] As referenced herein, the term “neural network” or “feed-forward neural network” refers to a mathematical framework to model a complicated function using a network of simpler functions represented by nodes, where the nodes are arranged in layers. Such a framework has an input layer, at least one hidden layer, and an output layer. Each node has a single output, which acts as an input to the nodes in subsequent layers away from the input and towards the output. When the output of node A acts as an input to node B, it is said that A is connected to B. Each connection between nodes has an associated multiplicative factor called weight, which is tuned using an optimization algorithm known as a learning or training algorithm to iteratively move the output of the neural network closer to that of a desired output. The neural network is trained using training data, while the optimization performance during training is measured by computing the errors between the neural network output and the desired output using an error function. The architecture of the neural network, that is, the number of hidden layers and nodes in each layer, is selected by evaluating the performance of multiple trained neural network architectures on validation data.

[0057] As referenced herein, the term “convolutional neural network” refers to a type of neural network having an input layer, at least one convolutional layer, a plurality of densely connected layers, and an output layer with each layer having one or more nodes (input, dense, and output layer) or filters (in convolutional layers) with or without connections that skip layers or feed the same layers.

[0058] As referenced herein, the term “deep learning” refers to a type of machine learning where the relationship between training input-output pairs is learned in terms of a hierarchical model composed of a plurality of intermediate layers that extract increasingly complex features as the information flows from input to the output. Examples of deep learning models include, but are not limited to, a deep convolutional neural network, a deep belief network, a recurrent neural network, an autoencoder, or the like.

[0059] As referenced herein, the term “parameters” refers to a set of numbers that determine the behavior of a classifier, such as the weights of connections in a neural network, which are determined using an automated tuning or numerical optimization operation known as “training.”

[0060] As referenced herein, the term “training” refers to a process of using training data and validation data to tune or set the parameters of a classifier using numerical optimization such that the classifier gives a desirable level of performance on these data as determined by a set measure of performance such as “average accuracy of classification on validation data.”

[0061] As referenced herein, the term “training process” refers to the process of setting a set of hyper-parameters of a classifier, training the classifier using training data, evaluating the classifier using validation data, and changing the hyper-parameters, if necessary, before training the classifier again.

[0062] As referenced herein, the term “testing data” refers to a set of input data whose ideal output is unknown at the time of using a machine learning framework that is learned using training. The desired output of the input associated with the testing data may be revealed at a later time, at which point it can be included in the training or validation data or used to evaluate the performance of a trained classifier.

[0063] In some embodiments, the step of classifying comprises a sequential model architecture comprising layers for data augmentation, rescaling, convolution, max pooling, dropout, and dense layers.

[0064] In some embodiments, one or more images are obtained from column-agglutinated gel cards. In some embodiments, the method comprises training the second machine learning model with agglutination reaction gradings comprising: 1+ agglutination, 2+ agglutination, 3+ agglutination, and 4+ agglutination.

[0065] Immunohematology is the study of red blood cell (RBC) antigens and antibodies associated with blood transfusions. There are more than 230 types of antigens present on the surface of RBCs that, based on their chemical structure, can be grouped into two major categories — carbohydrates and polypeptides. Immunohematology analysis may include: (i) Blood type: Grouping samples into A, B, AB, or O; (ii) Crossmatching: Mixing donor and recipient serum to check for agglutination; and / or (iii) rhesus factor D (Rh D) type: Determining if a sample is Rh+ or Rh- by checking for the D antigen. Immunohematology also includes molecular testing using DNA arrays. This can help screen donors for antigennegative RBC components and match the antigen-negative status of a transfusion recipient to that of a donor.

[0066] In some embodiments, the immunohematology analysis comprises forward-blood typing, reverse-blood typing, or antibody screening. In some embodiments, the sample comprises a blood sample.

[0067] In another aspect, this disclosure provides a system for immunohematology analysis using a machine learning model trained by synthetic image data. In some embodiments, the system comprises one or more processors configured to: (i) generate a training dataset comprising synthetic images by a first machine learning model, wherein the synthetic images contain no personally identifiable information of any person; (ii) train a second machine learning model based on the training dataset; (iii) determine, based at least on a classification result of the training dataset by the second machine learning, that the second machine learning model is able to classify a threshold quantity of images in the training dataset; (iv) input into the second machine learning model image data comprising one or more images obtained from a sample of a patient; and (v) classify by the second machine learning model the image data based at least on one or more characteristics of the sample of the patient, wherein one or more characteristics comprise blood type, crossmatching, rhesus factor D, or a combination thereof.

[0068] In some embodiments, the first machine learning model comprises Stable Diffusion. In some embodiments, the one or more processors are further configured to generate the training dataset by an image2image Stable Diffusion process.

[0069] In some embodiments, the second machine learning model comprises a convolutional neural network model (e.g., deep convolutional neural network model).

[0070] In some embodiments, the step of classifying comprises a sequential model architecture comprising layers for data augmentation, rescaling, convolution, max pooling, dropout, and dense layers.

[0071] In some embodiments, one or more images are obtained from column-agglutinated gel cards. In some embodiments, wherein the one or more processors are configured to train the second machine learning model with agglutination reaction gradings comprising: 1+ agglutination, 2+ agglutination, 3+ agglutination, and 4+ agglutination. 1+ agglutination: The presence of many small clumps (agglutinates); 2+ agglutination: The red blood cell button breaks into many medium-sized agglutinates; 3+ agglutination: The red blood cell button breaks into several large agglutinates; 4+ agglutination: A solid clump of red cells; and 0 agglutination: The red cell button is entirely resuspended and not visible.

[0072] In some embodiments, the immunohematology analysis comprises forward-blood typing, reverse-blood typing, or antibody screening. In some embodiments, the sample comprises a blood sample.

[0073] In some embodiments, the method may include preprocessing the images (e.g., synthetic images). In some embodiments, preprocessing comprises detecting a face in a facial image, cropping, resizing, gradation conversion, median filtering, histogram equalization, or size-normalized image processing.

[0074] The term “image” or “images,” as used herein, refers to single or multiple frames of still or animated images, video clips, video streams, or the like. Preprocessing may include detecting a facial image in the image of the subject. Preprocessing may also include cropping, resizing, gradation conversion, median filtering, histogram equalization, and / or size- normalized image processing. In some embodiments, the system may resize the photo or the videos according to a threshold value (e.g., maximum size in kilobytes, megabytes or gigabytes, maximum or minimum resolution in dots per inch (DPI) or pixels per inch (PPI)). In some embodiments, the system may resize the photo or the videos based on the transmission rate of the network and the links.

[0075] In some embodiments, the system may perform additional processing steps to the captured images or videos to digitalize the data file and optionally compress it into a convenient compressed file format and send it to a network protocol stack for subsequent conveyance over a local or wide area network. Typical compression schemes include MPEG, JPEG, H.261 or H.263, wavelet, or a variety of proprietary compression schemes. A typical network topology is the popular Ethernet standard, IEEE 802.3, and may operate at speeds from 10 Mb / s to 100 Mb / s. Network protocols are typically TCP / IP and UDP / IP and may be Unicast or Multicast as dictated by the system requirements.

[0076] Figure 6 is a functional diagram illustrating a programmed computer system in accordance with some embodiments. As will be apparent, other computer system architectures and configurations can be used to perform the described methods. Computer system 600, which includes various subsystems as described below, includes at least one microprocessor subsystem (also referred to as a processor or a central processing unit (CPU) 606). For example, processor 606 can be implemented by a single-chip processor or by multiple processors. In some embodiments, processor 606 is a general-purpose digital processor that controls the operation of the computer system 600. In some embodiments, processor 606 also includes one or more coprocessors or special purpose processors (e.g., a graphics processor, a network processor, etc.). Using instructions retrieved from memory 607, processor 606 controls the reception and manipulation of input data received on an input device (e.g., image processing device 603, I / O device interface 602), and the output and display of data on output devices (e.g., display 601).

[0077] Processor 606 is coupled bi-directionally with memory 607, which can include, for example, one or more random access memories (RAM) and / or one or more read-only memories (ROM). As is well known in the art, memory 607 can be used as a general storage area, a temporary (e.g., scratchpad) memory, and / or a cache memory. Memory 607 can also be used to store input data and processed data, as well as to store programming instructions and data, in the form of data objects and text objects, in addition to other data and instructions for processes operating on processor 606. Also, as is well known in the art, memory 607 typically includes basic operating instructions, program code, data, and objects used by the processor 606 to perform its functions (e.g., programmed instructions). For example, memory 607 can include any suitable computer-readable storage media described below, depending on whether, for example, data access needs to be bi-directional or uni-directional. For example, processor 606 can also directly and very rapidly retrieve and store frequently needed data in a cache memory included in memory 607.

[0078] A removable mass storage device 608 provides additional data storage capacity for the computer system 600 and is optionally coupled either bi-directionally (read / write) or unidirectionally (read-only) to processor 606. A fixed mass storage 609 can also, for example, provide additional data storage capacity. For example, storage devices 608 and / or 609 can include computer-readable media such as magnetic tape, flash memory, PC-CARDS, portable mass storage devices such as hard drives (e.g., magnetic, optical, or solid-state drives), holographic storage devices, and other storage devices. Mass storage 608 and / or 609 generally store additional programming instructions, data, and the like that typically are not in active use by the processor 606. It will be appreciated that the information retained within mass storage 608 and 609 can be incorporated, if needed, in a standard fashion as part of memory 607 (e.g., RAM) as virtual memory.

[0079] In addition to providing processor 606 access to storage subsystems, bus 610 can be used to provide access to other subsystems and devices as well. As shown, these can include a display 601, a network interface 604, an input / output (VO) device interface 602, an image processing device 603, as well as other subsystems and devices. For example, image processing device 603 can include a camera, a scanner, etc.; I / O device interface 602 can include a device interface for interacting with a touchscreen (e.g., a capacitive touch-sensitive screen that supports gesture interpretation), a microphone, a sound card, a speaker, a keyboard, a pointing device (e.g., a mouse, a stylus, a human finger), a global positioning system (GPS) receiver, a differential global positioning system (DGPS) receiver, an accelerometer, and / or any other appropriate device interface for interacting with system 600. Multiple I / O device interfaces can be used in conjunction with computer system 600. The I / O device interface can include general and customized interfaces that allow the processor 606 to send and, more typically, receive data from other devices such as keyboards, pointing devices, microphones, touchscreens, transducer card readers, tape readers, voice or handwriting recognizers, biometrics readers, cameras, portable mass storage devices, and other computers.

[0080] The network interface 604 allows processor 606 to be coupled to another computer, computer network, or telecommunications network using a network connection as shown. For example, through the network interface 604, the processor 606 can receive information (e.g., data objects or program instructions) from another network, or output information to another network in the course of performing method / process steps. Information, often represented as a sequence of instructions to be executed on a processor, can be received from and outputted to another network. An interface card or similar device and appropriate software implemented by (e.g., executed / performed on) processor 606 can be used to connect the computer system 600 to an external network and transfer data according to standard protocols. For example, various process embodiments disclosed herein can be executed on processor 606 or can be performed across a network such as the Internet, intranet networks, or local area networks, in conjunction with a remote processor that shares a portion of the processing. Additional mass storage devices (not shown) can also be connected to processor 606 through network interface 604

[0081] In addition, various embodiments disclosed herein further relate to computer storage products with a computer-readable medium that includes program code for performing various computer-implemented operations. The computer-readable medium includes any data storage device that can store data that can thereafter be read by a computer system. Examples of computer-readable media include, but are not limited to: magnetic media such as disks and magnetic tape; optical media such as CD-ROM disks; magneto-optical media such as optical disks; and specially configured hardware devices such as application-specific integrated circuits (ASICs), programmable logic devices (PLDs), and ROM and RAM devices. Examples of program code include both machine code as produced, for example, by a compiler, or files containing higher level code (e.g., script) that can be executed using an interpreter.

[0082] The computer system as shown in Figure 6 is an example of a computer system suitable for use with the various embodiments disclosed herein. Other computer systems suitable for such use can include additional or fewer subsystems. In some computer systems, subsystems can share components (e.g., for touchscreen-based devices such as smartphones, tablets, etc., I / O device interface 602 and display 601 share the touch-sensitive screen component, which both detect user inputs and displays outputs to the user). In addition, bus 610 is illustrative of any interconnection scheme serving to link the subsystems. Other computer architectures having different configurations of subsystems can also be utilized.

[0083] Additional Definitions

[0084] To aid in understanding the detailed description of the compositions and methods according to the disclosure, a few express definitions are provided to facilitate an unambiguous disclosure of the various aspects of the disclosure. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure belongs.

[0085] Aspects of the present disclosure are described herein with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the disclosure. In some embodiments, the flowchart and block diagrams in the Figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present disclosure. In this regard, each block in the flowchart or block diagrams may represent a module, a segment, or a portion of instructions, which may include one or more executable instructions for implementing the specified logical function(s). In some alternative implementations, the functions noted in the blocks may occur out of the order noted in the Figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and / or flowchart illustration, and combinations of blocks in the block diagrams and / or flowchart illustration, can be implemented by special purpose hardware-based systems that perform the specified functions or acts or carry out combinations of special purpose hardware and computer instructions.

[0086] These computer readable program instructions may be provided to a processor of a general-purpose computer, a special purpose computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions / acts specified in the flowchart and / or block diagram block or blocks. These computer readable program instructions may also be stored in a computer readable storage medium that can direct a computer, a programmable data processing apparatus, and / or other devices to function in a particular manner, such that the computer readable storage medium having instructions stored therein comprises an article of manufacture including instructions which implement aspects of the function / act specified in the flowchart and / or block diagram block or blocks.

[0087] The computer readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer, other programmable apparatus or other device to produce a computer implemented process, such that the instructions which execute on the computer, other programmable apparatus, or other device implement the functions / acts specified in the flowchart and / or block diagram block or blocks.

[0088] It will be understood that, although the terms “first,” “second,” etc., may be used herein to describe various elements, components, regions, layers and / or sections, these elements, components, regions, layers and / or sections should not be limited by these terms. These terms are only used to distinguish one element, component, region, layer or section from another element, component, region, layer or section. Thus, a first element, component, region, layer or section discussed below could be termed a second element, component, region, layer or section without departing from the teachings of example embodiments.

[0089] Unless specifically stated otherwise, as apparent from the above discussion, it is appreciated that throughout the description, discussions utilizing terms such as “processing,” “performing,” “receiving,” “computing,” “calculating,” “determining,” “identifying,” “displaying,” “providing,” “merging,” “combining,” “running,” “transmitting,” “obtaining,” or the like, refer to the action and processes of a computer system, or similar electronic computing device, that manipulates and transforms data represented as physical (or electronic) quantities within the computer system memories or registers or other such information storage, transmission or display devices.

[0090] As used herein, the term “if’ may be construed to mean “when,” “upon,” “in response to determining,” or “in response to detecting,” depending on the context. Similarly, the phrase “if it is determined” or “if [a stated condition or event] is detected” may be construed to mean “upon determining” or “in response to determining” or “upon detecting [the stated condition or event]” or “in response to detecting [the stated condition or event],” depending on the context.

[0091] As used herein, the term “classification” refers to any number or other characters that are associated with a particular property of a sample. The classification can be binary (e.g., positive or negative) or have more levels of classification (e.g., a scale from 1 to 10 or 0 to 1). The term “cutoff’ or “threshold” refers to a predetermined number used in an operation. For example, a cutoff value can refer to a classification score as used above. A threshold value may be a value above or below which a particular classification applies. Either of these terms can be used in either of these contexts.

[0092] The term “machine learning,” as used herein, refers to a computer algorithm used to extract useful information from a database by building probabilistic models in an automated way.

[0093] The term “regression tree,” as used herein, refers to a decision tree that predicts values of continuous variables.

[0094] The term “supervised learning,” as used herein, refers to a data analysis using a well- defined (known) dependent variable. All regression and classification algorithms are supervised. In contrast, “unsupervised learning” refers to the collection of algorithms where groupings of the data are defined without the use of a dependent variable.

[0095] The term “test data” refers to a data set independent of the training data set used to evaluate the estimates of the model parameters (z.e., weights).

[0096] As used herein, the term “clustering tree” refers to a hierarchical tree structure in which observations, such as organisms, genes, and polynucleotides, are separated into one or more clusters. The root node of a clustering tree consists of a single cluster containing all observations, and the leaf nodes correspond to individual observations. A clustering tree can be constructed based on a variety of characteristics of the observations. Many techniques known in the art, e.g., hierarchical clustering analysis, can be used to construct a clustering tree. A non-limiting example of a clustering tree is a phylogenetic, taxonomic or evolutionary tree.

[0097] As used herein, the term “ / / / vitro" refers to events that occur in an artificial environment, e.g., in a test tube or reaction vessel, in cell culture, etc., rather than within a multi-cellular organism.

[0098] As used herein, the term “ / / / vivo" refers to events that occur within a multi-cellular organism, such as a non-human animal.

[0099] It is noted here that, as used in this specification and the appended claims, the singular forms “a,” “an,” and “the” include plural reference unless the context clearly dictates otherwise. The terms “including,” “comprising,” “containing,” or “having” and variations thereof are meant to encompass the items listed thereafter and equivalents thereof as well as additional subject matter unless otherwise noted.

[0100] The phrases “in one embodiment,” “in various embodiments,” “in some embodiments,” and the like are used repeatedly. Such phrases do not necessarily refer to the same embodiment, but they may unless the context dictates otherwise.

[0101] The terms “and / or” or “ / ” means any one of the items, any combination of the items, or all the items with which this term is associated.

[0102] As used herein, the term “approximately” or “about,” as applied to one or more values of interest, refers to a value that is similar to a stated reference value. In some embodiments, the term “approximately” or “about” refers to a range of values that fall within 10%, 9%, 8%, 7%, 6%, 5%, 4%, 3%, 2%, 1%, or less in either direction (greater than or less than) of the stated reference value unless otherwise stated or otherwise evident from the context (except where such number would exceed 100% of a possible value). Unless indicated otherwise herein, the term “about” is intended to include values, e.g., weight percents, proximate to the recited range that are equivalent in terms of the functionality of the individual ingredient, the composition, or the embodiment.

[0103] It is to be understood that wherever values and ranges are provided herein, all values and ranges encompassed by these values and ranges are meant to be encompassed within the scope of the present disclosure. Moreover, all values that fall within these ranges, as well as the upper or lower limits of a range of values, are also contemplated by the present application.

[0104] As used herein, the term “each,” when used in reference to a collection of items, is intended to identify an individual item in the collection but does not necessarily refer to every item in the collection. Exceptions can occur if explicit disclosure or context clearly dictates otherwise.

[0105] The use of any and all examples, or exemplary language (e.g., “such as”) provided herein, is intended merely to better illuminate the disclosure and does not pose a limitation on the scope of this disclosure unless otherwise claimed. No language in the specification should be construed as indicating any non-claimed element as essential to the practice of this disclosure.

[0106] All methods described herein are performed in any suitable order unless otherwise indicated herein or otherwise clearly contradicted by context. Regarding any of the methods provided, the steps of the method may occur simultaneously or sequentially. When the steps of the method occur sequentially, the steps may occur in any order, unless noted otherwise.

[0107] In cases in which a method comprises a combination of steps, each and every combination or sub-combination of the steps is encompassed within the scope of the disclosure, unless otherwise noted herein.

[0108] Each publication, patent application, patent, and other reference cited herein is incorporated by reference in its entirety to the extent that it is not inconsistent with the present disclosure. Publications disclosed herein are provided solely for their disclosure prior to the filing date of the present disclosure. Nothing herein is to be construed as an admission that the present disclosure is not entitled to antedate such publication by virtue of prior invention. Further, the dates of publication provided may be different from the actual publication dates, which may need to be independently confirmed.

[0109] It is understood that the examples and embodiments described herein are for illustrative purposes only and that various modifications or changes in light thereof will be suggested to persons skilled in the art and are to be included within the spirit and purview of this application and scope of the appended claims.

[0110] EXAMPLE

[0111] This example describes a novel approach to training clinical diagnostic models using synthetic data generated by Stable Diffusion. The goal was to demonstrate the feasibility of this technique by looking at the accuracy obtained by running real agglutinated Gel Cards against the synthetically trained model.

[0112] The method employs Stable Diffusion vl.5 in ‘image2image’ mode to create training images from real results of Column Agglutinated Gel Cards obtained from the Bio-Rad IH-500 Instrument (Figures 1 and 2). These Al-generated images are used to train an image classification model (Convolutional Neural Network), focusing on four classes of reaction gradings (Figure 3). The model is trained exclusively with synthetic imagery, with no real instrument-harvested images included in the training dataset.

[0113] The model achieved accuracy = 1.0 across all four classes when tested with real Gel Card Agglutinations (Figures 5A-5B). This demonstrates the effectiveness of synthetic data in training medical diagnostic models, potentially reducing the time, cost, and complexity associated with traditional data collection and labeling. Additionally, this approach addresses privacy and data management issues and allows for the inclusion of rare or novel cases in the training set. The findings suggest that Al-generated synthetic data can significantly enhance the development of clinical diagnostic tools, offering a more efficient pathway to market entry.

[0114] Methodology

[0115] Stable Diffusion vl.5 was used to generate training images from real Gel Cards. The real Gel Card images were harvested from test results generated on the IH-500 Instrument manufactured by Bio-Rad.

[0116] An image classification model in Keras was then constructed and trained exclusively on the output images from the image2image stable diffusion process. Sample synthetic images from the Al-generated training set are shown in Figure 3.

[0117] Model Architecture and Training

[0118] The process involved training an image classification model using Al-generated images. Image generation was performed using the image-to-image feature of Stable Diffusion 1.5, with the source images constituting images of reacted gel cards harvested from the IH-500 instrument in the field. The average ratio of seed images to Stable Diffusion generated images was 0.014. Denoising was kept at 0.17, which was experimentally found to generate realistic variants of the seed image.

[0119] A sequential model architecture was employed, consisting of layers for data augmentation, rescaling, convolution (Conv2D), max pooling, 0.35 dropout, and dense layers. The model was trained over 32 epochs.

[0120] Training data augmentation was performed using the built-in Keras functions of Zoom, Rotation (0.2), and Horizontal flipping. Training data was split 80 / 20 train to validation. Finally, images were randomly order- shuffled before training.

[0121] The model was trained across four classes of reaction gradings (‘+,’ ‘++,’ ‘+++,’ and ‘ ++++’), with an average number of 274.5 synthetic images / class. The training data did not include any images harvested from an instrument - the training set was completely generated by Stable Diffusion.

[0122] No data exists from the field on the expected distribution of the four results classes, so it was decided not to attempt to significantly imbalance any result class in the training data. The same reasoning was applied to Color and Pixel Depth distributions - field results were chosen randomly without regard to color histogram spread or pixel depth. Sample training outputs are shown in Figure 4. Results

[0123] Test image Criteria:

[0124] A) They are identical or otherwise similar visually to the image2image inputs to the Stable Diffusion generator.

[0125] B) They are not identical or otherwise similar visually to the output of the image from the Stable Diffusion generator.

[0126] C) They are harvested from the first-class labeled Google image search results for the term ‘Gel Card (n)+’ (where n is in (1..4)).

[0127] Predicted Class Accuracy:

[0128] The model achieved 1.0 accuracy across all 4 classes with 8 images of real Gel Card Agglutinations, as shown in Figures 5A and 5B.

[0129] Discussion

[0130] Model training is an expensive and time-consuming venture, especially when data needs to be labeled. For medical instrumentation, this effort is even more pronounced as training data must be collected and labeled from the field. The technique demonstrated here can shift the burden from field collection onto low-cost hardware, reducing the time and effort needed to collect and label images by degrees of magnitude. Access to instruments is not easy or cheap - access to Stable Diffusion, on the other hand, is. Additionally, with Synthetic data, privacy and data management complexities are virtually eliminated.

[0131] Generative Al models can also be leveraged to provide special cases rarely observed in the field or to provide training models on novel or otherwise difficult-to-produce training data. Techniques like Low-Rank Adaptation models (LORA) can provide even more specificity as needs arise without the computation and time cost of adjusting billions of weights in a generative model.

[0132] The present disclosure is not to be limited in scope by the specific embodiments described herein. Indeed, various modifications of the disclosure, in addition to those described herein, will become apparent to those skilled in the art from the foregoing description and the accompanying figures. Such modifications are intended to fall within the scope of the appended claims.

Claims

CLAIMSWhat is claimed is:

1. A method for immunohematology analysis using a machine learning model trained by synthetic image data, comprising: generating a training dataset comprising synthetic images by a first machine learning model, wherein the synthetic images contain no personally identifiable information of any person; training a second machine learning model based on the training dataset; determining, based at least on a classification result of the training dataset by the second machine learning, that the second machine learning model is able to classify a threshold quantity of images in the training dataset; inputting into the second machine learning model image data comprising one or more images obtained from a sample of a patient; and classifying by the second machine learning model the image data based at least on one or more characteristics of the sample of the patient, wherein one or more characteristics comprise blood type, crossmatching, rhesus factor D, or a combination thereof.

2. The method of claim 1, wherein the first machine learning model comprises Stable Diffusion.

3. The method of claim 2, comprising generating the training dataset by an image2image Stable Diffusion process.

4. The method of claim 2, wherein an average ratio of seed images to the synthetic images ranges from zero to one.

5. The method of any one of the preceding claims, comprising denoising the synthetic images with a denoising parameter ranging from zero to one.

6. The method of any one of the preceding claims, wherein the second machine learning model comprises a convolutional neural network model.

7. The method of any one of the preceding claims, wherein the step of classifying comprises a sequential model architecture comprising layers for data augmentation, rescaling, convolution, max pooling, dropout, and dense layers.

8. The method of any one of the preceding claims, wherein the one or more images are obtained from column agglutinated gel cards.

9. The method of any one of the preceding claims, comprising training the second machine learning model with agglutination reaction gradings comprising: 1+ agglutination, 2+ agglutination, 3+ agglutination, and 4+ agglutination.

10. The method of any one of the preceding claims, wherein the immunohematology analysis comprises forward-blood typing, reverse-blood typing, or antibody screening.

11. The method of any one of the preceding claims, wherein the sample comprises a blood sample.

12. A system for immunohematology analysis using a machine learning model trained by synthetic image data, comprising one or more processors configured to: generate a training dataset comprising synthetic images by a first machine learning model, wherein the synthetic images contain no personally identifiable information of any person; train a second machine learning model based on the training dataset; determine, based at least on a classification result of the training dataset by the second machine learning, that the second machine learning model is able to classify a threshold quantity of images in the training dataset; input into the second machine learning model image data comprising one or more images obtained from a sample of a patient; andclassify by the second machine learning model the image data based at least on one or more characteristics of the sample of the patient, wherein one or more characteristics comprise blood type, crossmatching, rhesus factor D, or a combination thereof.

13. The system of claim 12, wherein the first machine learning model comprises Stable Diffusion.

14. The system of claim 13, wherein the one or more processors are further configured to generate the training dataset by an image2image Stable Diffusion process.

15. The system of claim 13, wherein an average ratio of seed images to the synthetic images ranges from zero to one.

16. The system of any one of claims 11-15, wherein the one or more processors are further configured to denoise the synthetic images with a denoising parameter ranging from zero to one.

17. The system of any one of claims 11-16, wherein the second machine learning model comprises a convolutional neural network model.

18. The system of any one of claims 11-17, wherein the step of classifying comprises a sequential model architecture comprising layers for data augmentation, rescaling, convolution, max pooling, dropout, and dense layers.

19. The system of any one of claims 11-18, wherein the one or more images are obtained from column agglutinated gel cards.

20. The system of any one of claims 11-19, wherein the one or more processors are configured to train the second machine learning model with agglutination reaction gradings comprising: 1+ agglutination, 2+ agglutination, 3+ agglutination, and 4+ agglutination.21 . The system of any one of claims 11-20, wherein the immunohematology analysis comprises forward-blood typing, reverse-blood typing, or antibody screening.

22. The system of any one of claims 11-21, wherein the sample comprises a blood sample.

23. A method or system substantially as shown and described in the specification and appended drawings.

Images