Phase separation learning model

Abstract

A phase separation learning model can receive fluid sample images and be operated to determine a presence or absence of phase separation in the fluid sample images. To do so, a computer vision learning model can be trained to identify objects in images via transfer learning utilizing a first plurality of fluid sample images to result in the phase separation learning model. A second plurality of fluid sample images can be input into the phase separation learning model, and the phase separation learning model can make a recommendation regarding the presence or the absence of phase separation within the second plurality of fluid sample images as a result of operation of the phase separation learning model.

Description

TDCC#86233-WO-PCTPHASE SEPARATION LEARNING MODELTechnical Field

[0001] The present disclosure relates to a phase separation learning model that can use a combination of computer- vision models capable of distinguishing phase separation from contaminations and artifacts in formulation development and material compatibility fundamental to identifying phase separation within fluid samples. Such techniques can be useful when phase transitions are challenging to detect or when fluid samples include impurities. Such techniques can be useful to provide a recommendation related to determining phase presence, phase clarity, number of phases, and volume fractions of phases, etc.Background

[0002] A convolutional neural network (CNN) is a regularized type of feed-forward network that learns feature engineering by itself via filters or kernel optimization. A CNN uses convolutional layers, which help with efficiently capturing spatial relations. CNNs may be applied to a process and can make predictions from a plurality of different types of data including text, images, and audio. A CNN consists of an input layer, hidden layers, and an output layer. In a CNN, the hidden layers include one or more layers that perform convolutions. As a convolution kernel slides along an input matrix for the layer, the convolution operation generates a feature map, which in turn contributes to the input of the next layer. Put another way, convolutional layers convolve the input and pass its result to the next layer.

[0003] Chemical laboratory samples such as fluid samples in laboratory vials are mixed and observed to report their physical state. Images of the samples can be captured and used for classifications and recordings. A CNN can be utilized in laboratory image sample classification, for instance, because when dealing with high-dimensional inputs such as images, it is impractical to connect neurons to all neurons in the previous volume. Such a network architecture does not take the spatial structure of the data into account. CNNs exploit spatially local correlation by enforcing a sparse local connectivity pattern between neurons of adjacent layers; each neuron is connected to only a small region of the input volume.TDCC#86233-WO-PCTSummary of the Disclosure

[0004] The present disclosure relates to a machine learning model for identifying phase separation within laboratory image samples, referred to herein as a phase separation learning model. The phase separation learning model allows for detecting phases and artifacts, anomalies, and contaminations in laboratory samples. The phase separation learning model can allow for identifying the presence of macroscopic phase separations, generally visible by the trained human eye, present in a fluid (e.g., liquid) mixture formulation from high-resolution imaging of samples (e.g., 1000-1100 dpi, 1300-1360 dpi, etc.). The phase separation learning model can be built utilizing CNNs originally trained to identify everyday objects (e.g., dogs, cars, trees, etc.) in millions of collected images (e.g., collected via the Internet). A transfer learning approach can be used to retrofit the CNNs to detect phase changes by making architecture adjustments and updating parameters in the CNNs to result in a phase separation learning model through performing re-training exercises using sample images labeled by subject matter experts (SMEs). Contrast enhancement and dynamic cropping may be utilized as preprocessing steps to the CNN model to improve accuracy of phase separation detection. In contrast, some previous approaches do not use machine learning, and specifically they do not use CNNs trained using known phase separation image data to result in a phase separation learning model. Further, some previous approaches do not utilize pre-processing dynamic cropping and / or image enhancement.

[0005] Automated phase and interface analysis based on sample images are important in high throughput research when hundreds to thousands of samples can be processed daily. The traditional analysis based on algorithms around curves and slopes of intensity profiles fails frequently when the phase transitions are challenging to detect or there are impurities in the images. The phase separation learning model using the images (not profiles) can be used to improve the detection of phases and contaminations. The phase separation learning model can include an ensemble of CNN-based computer vision learning models capable of distinguishing phase separation from contamination artifacts with increased accuracy as compared to previous approaches.

[0006] The above summary of the present disclosure is not intended to describe each disclosed embodiment or every implementation of the present disclosure. The description that follows more particularly exemplifies illustrative embodiments. In several places throughout theTDCC#86233-WO-PCTapplication, guidance is provided through lists of examples, which examples can be used in various combinations. In each instance, the recited list serves only as a representative group and should not be interpreted as an exclusive list.Brief Description of the Drawings

[0007] Figure 1 is a block diagram illustrating an example sample deck, mixing, and observation laboratory configuration.

[0008] Figure 2A illustrates example fluid sample images.

[0009] Figure 2B illustrates the example fluid sample images of Figure 2A with the coloring inverted.

[0010] Figure 3 is a method flow diagram for operating a phase separation learning model.

[0011] Figure 4 illustrates an example of a machine readable medium to operate a phase separation learning model.

[0012] Figure 5 is an example machine within which a set of instructions, for causing the machine to perform various methodologies discussed herein, can be executed.

[0013] Figure 6 is another method flow diagram for operating a phase separation learning model.

[0014] Figure 7 illustrates an example machine within which a set of instructions, for causing the machine to perform various methodologies discussed herein, can be executed.Detailed Description

[0015] Phase separation is the creation of two or more distinct phases from a single homogeneous mixture. The mixture may spontaneously separate into two distinct immiscible liquids, or “phases” which may be referred to as a dense phase and a dilute phase. Phase separation can mediate the compartmentalization of proteins in cells, among other mechanisms. Phase separation, also referred to as liquid-liquid phase separation (LLPS) includes at least one of the liquid components having a tendency to interact with itself more than with the other components, becoming a better solvent for itself than a buffer it is dissolved in. As a result, this component also becomes more concentrated in the dense phase.TDCC#86233-WO-PCT

[0016] In fluid sample vials, phase separation can be used to determine if certain materials are compatible or incompatible. Identifying the presence or absence of a phase separation, particularly in fluid sample images, can be challenging, as the images may not be clear, clean pictures due to dust, artifacts, fingerprints, anomalies, etc. In order to classify these fluid sample images, a phase separation learning model can be utilized to determine whether the image has one phase or two phases (or more than two phases). While these images can be analyzed by SMEs, tens or hundreds of thousands of the images may be created annually, meaning individual analysis is impractical.

[0017] While several fluid sample images are created, even greater numbers are needed to create a computer vision learning model from scratch. Because of the relatively small amount of data as compared to data collected by other computer vision image classification models, utilizing those computer vision learning models while integrating fluid sample images having known phase separation (or known lack thereof) can be less expensive and time consuming, while accurately and quickly identifying phase separation. SMEs may confirm fluid sample images as having phase separation or not, based on knowledge of artifacts, fingerprints, anomalies, etc. that may be present, as well as knowledge of locations where artifacts commonly occur, and common identifiers to true phase separation. The confirmed fluid sample images can be utilized to train the computer vision learning models previously trained for image recognition to result in a phase separation learning model.

[0018] As used herein, the singular forms “a”, “an”, and “the” include singular and plural referents unless the content clearly dictates otherwise. Furthermore, the word “may” is used throughout this application in a permissive sense (i.e., having the potential to, being able to), not in a mandatory sense (i.e., must). The term “include,” and derivations thereof, mean “including, but not limited to.” The term “coupled” means directly or indirectly connected and, unless stated otherwise, can include a wireless connection.

[0019] As will be appreciated, elements shown in the various embodiments herein can be added, exchanged, and / or eliminated so as to provide a number of additional embodiments of the present disclosure. In addition, as will be appreciated, the proportion and the relative scale of the elements provided in the figures are intended to illustrate certain embodiments of the present invention and should not be taken in a limiting sense.TDCC#86233-WO-PCT

[0020] Figure 1 is a block diagram 120 illustrating an example sample deck, mixing, and observation laboratory configuration, in which the temperature of the samples can be controlled. In formulation development and material compatibility fundamentals, fluid samples such as those in a vial tray 121 on a sample deck 122 can be mixed and observed to report their physical state. Mixing can occur, for instance in a spin mixer 124, a wrist shaker 123, or a combination thereof, among other mixing techniques. Images of the samples can be captured in the imaging portion 125 of the laboratory configuration and used for classifications and recording. The imaging portion 125 may include a plurality of alternate imaging backgrounds, a camera, and a camera housing, among other apparatus. In some examples, the laboratory configuration can include a robotic arm to move the samples between and / or within different apparatuses of the laboratory configuration.

[0021] For example, fluid samples for which phase separation determinations are desired can be placed in vial tray 121. A robotic arm may move the vials of fluid sample to either or both of the spin mixer 124 and the wrist shaker 123. Upon completion of mixing, the vials of fluid samples can be transferred by the robotic arm to the imaging portion 125. Alternatively, the robotic arm may move the sample directly to the imaging portion 125. Fluid sample images can be acquired and a phase separation learning model may operate to determine and recommend whether phase separation in each of the vials has occurred. In some examples, a computing device may be present within the imaging portion 125 where the phase separation learning model is operated. In other examples, the phase separation learning model received sample fluid image data outside of the imaging portion 125.

[0022] Figure 2A illustrates first example fluid sample images 226-1, 226-2,..., 226-n and second example fluid sample images 228-1, 228-2,..., 228-m. The first example fluid sample images 226- 1, 226-2,..., 226-n illustrate samples with a single phase, while the second example fluid sample images 228-1, 228-2,..., 228-m illustrate samples with multiple (two) phases. For instance, the darker areas in the first example fluid sample images 226-1, 226-2,..., 226-n and the second example fluid sample images 228-1, 228-2,..., 228-m illustrate a first phase, while the lighter areas (e.g., bands of light) indicate a phase change (e.g., a second phase) within the fluid samples.

[0023] Figure 2B illustrates the first example fluid sample images 226-1, 226-2,..., 226-n and the second example fluid sample images 228-1, 228-2,..., 228-m of Figure 2 A with theTDCC#86233-WO-PCTcoloring inverted. The first example fluid sample images 226-1, 226-2,..., 226-n illustrate samples with a single phase, while the second example fluid sample images 228-1, 228-2,..., 228-m illustrate samples with multiple (two) phases. For instance, the lighter areas in the first example fluid sample images 226-1, 226-2,..., 226-n and the second example fluid sample images 228-1, 228-2,..., 228-m illustrates a first phase, while the darker areas (e g., bands of darkness) indicate a phase change (e.g., a second phase) within the fluid samples.

[0024] In some examples, contrast enhancement can be performed on fluid sample images such as the first fluid sample images 226-1, 226-2,..., 226-n and the second fluid sample images 228-1, 228-2,..., 228-m before feeding the images to the phase separation learning model. In some instances, contrast between phases is easily identifiable, however in other instances, materials and other factors can make the contrast much fainter. By enhancing the contrast, the phase separation learning model can more accurately identify phase separation within fluid sample images.

[0025] Alternatively, or in addition to the contrast enhancement, dynamic cropping can be performed on fluid sample images such as the first fluid sample images 226-1, 226-2,..., 226-n and the second fluid sample images 228-1, 228-2,..., 228-m before feeding the images to the phase separation learning model. For instance, when samples are made in a laboratory the height of the samples is not always the same. The vials may be similar, but how much material is in each vial can vary from image to image. Additionally, near the top of the vials, caps and other elements may cause some light reflections and a bright line due to the air liquid interface. These issues can interfere with these computer vision learning models because the computer vision learning models may identify them as features that can be used for image classification. To address this, a dynamic cropping can be performed on the fluid sample images to focus in on a part of the image that was the actual sample, and to remove parts of the image of those brighter artifacts, which can aid the computer vision learning model identify if there was one phase or more than one phase using the cropped fluid sample images. The first fluid sample images 226-1, 226-2,..., 226-n and the second fluid sample images 228-1, 228-2,..., 228 illustrate dynamically cropped fluid sample images.

[0026] The cropping can be performed automatically and can include calibrating the fluid sample images relative to a reflection of light on a vial or vial housing. A reflection learning analysis can be used to identify the light reflection in fluid sample images and determine whatTDCC#86233-WO-PCTportion of each of the fluid sample images includes the fluid sample on which to focus, rather than a reflection, artifact, etc. The reflection learning analysis represents a traditional (e.g., human-designed) image analysis. It involves integrating pixel intensity of a sample image along a minor axis to create an integrated intensity profile (e.g., a curve). An algorithm can be applied to identify an appropriate image cutline based on the shape of the curve. This may be a same or similar type of analysis used previously to identify phase separations, but because the relative brightness of the meniscus and vial caps are higher, this reflection learning analysis may not suffer from interference from contaminants and artifacts the way a phase separation identification does.

[0027] The reflection learning analysis can be used to determine likely locations of a reflection of light on the vial or the vial house. The reflection learning analysis can be updated as new data is received regarding light reflection on the vials or the vial houses. The reflection learning analysis can identify where that bright reflection of light would be, and the information can be used as a reference point of where to determine where the sample is in the fluid sample image relative to the bright light. Put another way, the reflection learning analysis can be used to determine what part of the fluid sample image should be included in a cropped image, and what part should not be included. The phase separation learning model can then focus on only the cropped image from then on.

[0028] Figure 3 is a method flow diagram 330 for operating a phase separation learning model. As illustrated at 331, the method can include receiving raw image data, and as illustrated at 332, the method can include dynamic cropping of the images and / or contrast enhancement as described herein. The dashed line originating at the raw images at 331 that bypasses the dynamic cropping and / or contrast enhancement at 332 indicates that the dynamic cropping and / or contrast enhancement are optional. As illustrated at 333, the method can include the phase separation learning model receiving the cropped and / or enhanced images. The phase separation learning model can include a plurality of convolution layers, in combination with other mechanisms that iteratively refine and produce a result as a sample image progresses through the model. As the model training begins, a plurality of iterations can be performed to determine which filters to use to emphasize particular features, etc. The phase separation learning model can adapt parameters to alter the filter until the phase separation learning model determines which filters best identify and distinguish fluid sample images having phaseTDCC#86233-WO-PCTseparation from fluid sample images that do not have phase separation, accounting for the presence of interloping image artifacts. Put another way, as illustrated at 334, as the phase separation learning model further progresses, a series of convolutional layers of the phase separation learning model can learn features that best distinguish whether or not phase separation is present in a sample image. A pentultimate fully-connected layer aggregates the feature information and provides probability scales that can be used to decide “yes” or “no” regarding whether the sample image has a phase separation.

[0029] Figure 4 illustrates an example of a machine readable medium 440 to operate a phase separation learning model. The machine readable medium 440 can be communicatively connected to a processor resource 471 by a communication path 472. In some examples, a communication path 472 can include a wired or wireless connection that can allow communication between devices and / or components within a single device. As used herein, the processor resource 471 can include, but is not limited to: a central processing unit (CPU), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA), a metal-programmable cell array (MPCA), a semiconductor-based microprocessor, or other combination of circuitry and / or logic to orchestrate execution of instructions 473, 474, 475, 476. In some examples, a graphics processing unit (GPU) may be used as part of the hardware for training the CNN model. In a specific example, the processor resource 471 utilizes a non-transitory computer-readable medium 440 storing instructions 473, 474, 475, 476, that, when executed, cause the processor resource 471 to perform corresponding functions.

[0030] The machine readable medium 440 may be electronic, magnetic, optical, or other physical storage device that stores executable instructions. Thus, a non-transitory machine-readable medium (MRM) (e.g., machine readable medium 440) may be, for example, a non-transitory MRM comprising Random-Access Memory (RAM), read-only memory (ROM), an Electrically Erasable Programmable ROM (EEPROM), a storage drive, an optical disc, and the like. The machine readable medium 440 may be disposed within a controller and / or computing device. In this example, the executable instructions 473, 474, 475, 476, can be “installed” on the device. Additionally, and / or alternatively, the machine readable medium 440 can be a portable, external, or remote storage medium, for example, which allows a computing system to download the instructions 473, 474, 475, 476, from the portable / external / remote storage medium. In this situation, the executable instructions may be part of an “installation package”.TDCC#86233-WO-PCT

[0031] The machine readable medium 440 includes instructions 473 to update a computer vision learning model trained to identify objects in images via transfer learning utilizing a first plurality of fluid sample images to result in a phase separation learning model. For example, the computer vision learning model may be a CNN learning model originally trained to identify everyday objects such as cats, vehicles, and plants, among others, in millions of images from the Internet. In some examples, a plurality of computer vision learning models can be used. Utilizing a transfer learning approach, these computer vision learning models can be retrofit to result in a CNN phase separation learning model to identify phase separation by making architectural adjustments to the computer vision learning models (e.g., adding / adjusting convolution layers) and updating parameters in the computer vision learning model through training exercises using fluid sample images identified by SME operators.

[0032] Training the computer vision learning model, for example, can include near-continuously training the computer vision learning model as new phase separation image data is received at the computer vision learning model. For instance, as new fluid sample images with known presence or absence of phase separation are created and collected, the computer vision learning model can be updated.

[0033] The machine readable medium 440 includes instructions 474 to input a second plurality of fluid sample images into the phase separation learning model. For instance, the second plurality of fluid sample images may be images needing phase separation determinations (e g., newly created samples). The machine readable medium 440 includes instructions 475 to operate the phase separation learning model to determine a presence or absence of a phase separation within the second plurality of fluid sample images. The phase separation learning model may be a CNN model and can include a plurality of convolutional layers such as a first learning layer to determine a plurality of features appearing in the second plurality of fluid sample images and a second learning layer to determine which of the plurality of features indicates phase separation in the second plurality of fluid sample images. The first learning layer, for instance, may be the penultimate layer described with respect to Figure 3, while the second learning layer may be the final probability learning layer. More than two layers may be present, and the first and the second layers are described herein as example layers. As additional fluid sample images are run through the phase separation learning model, parameters within theTDCC#86233-WO-PCTphase separation learning model can be controlled and adjusted to further enhance the phase separation learning model.

[0034] The machine readable medium 440 includes instructions 476 to provide a recommendation regarding the presence or the absence of phase separation within the second plurality of fluid sample images. The recommendation may include a probability of a particular fluid sample image containing or not containing a phase separation. The recommendation, in some examples, may also be indeterminate, or may include a recommendation to re-run the fluid sample image.

[0035] In some embodiments, the machine readable medium 440 can include instructions to perform dynamic cropping on the second plurality of fluid sample images prior to inputting the second plurality of fluid sample images into the phase separation learning model, such that only a portion of each fluid sample image is viewed by the phase separation learning model. Dynamically cropping the second plurality of fluid sample images can include calibrating each of the second plurality of fluid sample images relative to a reflection of light on a vial housing a respective fluid sample. For instance, dynamic cropping can include operating a reflection learning analysis to identify light reflection in the first plurality of fluid sample images and determine what portion of each of the first plurality of fluid sample images includes the fluid sample.

[0036] In some embodiments, the machine readable medium 440 can include instructions to perform contrast enhancement on the second plurality of fluid sample images prior to inputting the second plurality of fluid sample images into the phase separation learning model. By increasing the contrast between an object in an image (e.g., dust particle, artifact, etc.) and their background (e.g., fluid), the visibility of the object can be improved, as can the accuracy of the phase separation learning model.

[0037] Figure 5 illustrates an example of a device 550 to operate a phase separation learning model. In some examples, the device 550 is a computing device that includes a processor resource 571 and a machine readable medium 540 to store instructions 582, 583, 584, 585, 586, 587 that are executed by the processor resource 571 to perform particular functions. Figure 5 illustrates how a computing device can execute instructions to perform functions described herein.TDCC#86233-WO-PCT

[0038] The device 550 includes instructions 582 stored by the machine readable medium 540 that are executed by the processor resource 571 to retrain a previously trained first computer vision learning model using identified phase separation image data of a first plurality of fluid sample images to result in a phase separation learning model. The previously trained first computer vision learning model may be a model that does not determine phase separation within the first plurality of fluid sample images. For example, the previously trained first computer vision learning model may be a CNN computer vision learning model that applies convolutions to images by applying a filter to the images that can amplify different patterns, shapes, colors, etc. that may be seen in the image using the CNN computer vision learning model but may be difficult to detect with the human eye. The CNN computer vision learning model may have originally been trained on millions of images on the Internet for image classification purposes (e g., what the animal or plant in an image is), and it can be retrained using the identified phase separation image data. More than one CNN computer vision learning model may be used in some embodiments.

[0039] In some examples, retraining can include retrofitting the previously trained first computer vision learning model by applying the identified phase separation image data to determine whether data already encoded into parameters of the previously trained first computer vision learning model are applicable to a phase separation determination. The identified phase separation image data can include image data with known phase separation or image data that is known to not contain phase separation. For instance, the identified phase separation image data may be fluid sample images that a plurality of SMEs has reviewed and identified as either having phase separation or not having phase separation. In some examples, the identified phase separation image data is cropped identified phase separation image data.

[0040] The device 550 includes instructions 583 stored by the machine readable medium 540 that are executed by the processor resource 571 to perform contrast enhancement on a second plurality of fluid sample images. Contrast enhancement can include increasing a difference in brightness between fluid in each of the plurality of fluid samples within the cropped second plurality of fluid sample images and other features visible in the cropped second plurality of fluid sample images.

[0041] The device 550 includes instructions 584 stored by the machine readable medium 540 that are executed by the processor resource 571 to perform dynamic cropping on the secondTDCC#86233-WO-PCTplurality of fluid sample images. The cropped second plurality of fluid sample images are cropped such that features that do not affect phase separation determinations are removed, or “cropped” out of the image. For instance, the cropped second plurality of fluid sample images illustrate a portion of each of the second plurality of fluid sample images having image features outside a fluid sample within each of the second plurality of fluid sample images removed. For instance, reflections that are not part of the fluid sample, may be cropped out of the fluid sample image. The dynamic cropping can include calibrating each of the second plurality of fluid sample images relative to a reflection of light or other artifact on a vial housing a respective fluid sample. The dynamic cropping can be performed before or after the contrast enhancement.

[0042] The device 550 includes instructions 585 stored by the machine readable medium 540 that are executed by the processor resource 571 to input the cropped and the enhanced first plurality of fluid sample images into the phase separation learning model. The device 550 includes instructions 586 stored by the machine readable medium 540 that are executed by the processor resource 571 to operate the phase separation learning model to determine a presence or absence of a phase separation within the second plurality of fluid sample images.

[0043] In some embodiments, the phase separation learning model can include multiple layers such as a first learning layer to determine a plurality of features appearing in the second plurality of fluid sample images and a second learning layer to determine which of the plurality of features indicates phase separation in the second plurality of fluid sample images. The “first layer” and / or the “second layer” can each include multiple respective layers, as will be appreciated by one of skill in the art. The phase separation learning model and the previously trained first computer vision learning model can be CNN models, with the first and the second learning layers being different convolution layers. More than two layers may be present in the phase separation learning model.

[0044] The device 550 includes instructions 587 stored by the machine readable medium 540 that are executed by the processor resource 571 to provide a recommendation regarding the presence or the absence of phase separation within the second plurality of fluid sample images. For instance, a recommendation may be sent via a network device, as a report, or in some other manner and can include a probability of a particular fluid sample containing or not containing a phase separation based on the associate fluid sample image run through the phase separation learning model.TDCC#86233-WO-PCT

[0045] Figure 6 is another method flow diagram 660 for operating a phase separation learning model. As illustrated at 661, the method can include updating a computer vision learning model trained to identify objects in images via transfer learning utilizing a first plurality of fluid sample images to result in a phase separation learning model. Transfer learning is a supervised machine learning technique that uses a pre-trained model to improve the performance of a new task. Supervised learning involves training with labeled data, in this instance, known phase separation in particular fluid sample images. Transfer learning allows for training deep neural networks with less data. For example, for image classification, knowledge gained while learning to recognize dogs, cats, trees, etc. can be applied when trying to recognize phase separation. Trained computer vision learning models have already learned patterns, features, colors, etc. to look for, so by utilizing the first plurality of fluid sample images that have known presence or absence of phase separation, a phase separation learning model can result.

[0046] For instance, a first portion of the first plurality of fluid sample images can have a known presence of phase separation, and a second portion of the first plurality of fluid sample images has a known absence of phase separation. The known presence or absence of phase separation may be based on determinations made by SMEs having studied the first plurality of fluid sample images.

[0047] Updating the computer vision learning model can include updating the computer vision learning model with a convolution determined during the training to apply to the second plurality of fluid sample images, controlling how many parameters are updated each iteration, and / or controlling by how much parameters are updated each iteration, among others. The second plurality of fluid sample images can include images needing analysis for the presence or absence of phase separation. Through multiple iterations, the phase separation learning model can be near-continuously adjusted as new known or unknown fluid sample images are received. Parameters can be adjusted to improve the training, learning, and image recognition accuracy, as well.

[0048] As illustrated at 662, the method can include dynamically cropping a second plurality of fluid sample images having unknown phase separation. For instance, the second plurality of fluid sample images needing analysis can be dynamically cropped to remove unwanted and / or unnecessary anomalies (e.g., reflections, vial components, etc.). The cropped images can be received at the phase separation learning model, as illustrated at 663, with each ofTDCC#86233-WO-PCTthe plurality of cropped images a cropped version of a respective one of the second plurality of fluid sample images.

[0049] As illustrated at 664, the method can include operating the phase separation learning model to determine a presence or absence of phase separation within the second plurality of fluid sample images and a probability of the presence or the absence of phase separation. This can include determining a probability that a determined presence of phase separation comprises two distinct phases in a respective fluid sample image or an anomaly in the respective fluid sample image. As illustrated at 665, the method can include providing a recommendation regarding the presence or the absence of phase separation within the second plurality of fluid sample images based on results of the operation of the phase separation learning model.

[0050] Figure 7 illustrates an example machine 700 within which a set of instructions, for causing the machine 700 to perform various methodologies discussed herein, can be executed. In various embodiments, the machine 700 can be analogous to a controller. In alternative embodiments, the machine 700 can be connected (e.g., networked) to other machines in a LAN, an intranet, an extranet, and / or the Internet. The machine 700 can operate in the capacity of a server or a client machine in client-server network environment, as a peer machine in a peer-to-peer (or distributed) network environment, or as a server or a client machine in a cloud computing infrastructure or environment.

[0051] The machine 700 can be a personal computer (PC), a tablet PC, a set-top box (STB), a Personal Digital Assistant (PDA), a cellular telephone, a web appliance, a server, a network router, a switch or bridge, or any machine capable of executing a set of instructions (sequential or otherwise) that specify actions to be taken by that machine. Further, while a single machine 700 is illustrated, the term “machine” shall also be taken to include any collection of machines that individually or jointly execute a set (or multiple sets) of instructions to perform any one or more of the methodologies discussed herein.

[0052] The example machine 700 includes a processing device 702, a main memory 704 (e.g., read-only memory (ROM), flash memory, dynamic random access memory (DRAM) such as synchronous DRAM (SDRAM) or Rambus DRAM (RDRAM), etc.), a static memory 706 (e.g., flash memory, static random access memory (SRAM), etc.), and a data storage system 708, which communicate with each other via a bus 710.TDCC#86233-WO-PCT

[0053] The processing device 702 represents one or more general -purpose processing devices such as a microprocessor, a central processing unit (CPU), or the like. More particularly, the processing device can be a complex instruction set computing (CISC) microprocessor, reduced instruction set computing (RISC) microprocessor, very long instruction word (VLIW) microprocessor, or a processor implementing other instruction sets, or processors implementing a combination of instruction sets. The processing device 702 can also be one or more specialpurpose processing devices such as an application specific integrated circuit (ASIC), a field programmable gate array (FPGA), a digital signal processor (DSP), network processor, or the like. The processing device 702 is configured to execute instructions 718 for performing the operations and steps discussed herein. The machine 700 can further include a network interface device 712 to communicate over the network 714.

[0054] The data storage system 708 can include a machine-readable storage medium 716 (also known as a computer-readable medium) on which is stored one or more sets of instructions 718 or software embodying any one or more of the methodologies or functions described herein. The instructions 718 can also reside, completely or at least partially, within the main memory 704 and / or within the processing device 702 during execution thereof by the machine 700, the main memory 704 and the processing device 702 also constituting machine-readable storage media.

[0055] In one embodiment, the instructions 718 include instructions to implement functionality corresponding to CNNs described herein. While the machine-readable storage medium 716 is shown in an example embodiment to be a single medium, the term “machine-readable storage medium” should be taken to include a single medium or multiple media that store the one or more sets of instructions. The term “machine-readable storage medium” shall also be taken to include any medium that is capable of storing or encoding a set of instructions for execution by the machine and that cause the machine to perform any one or more of the methodologies of the present disclosure. The term “machine-readable storage medium” shall accordingly be taken to include, but not be limited to, solid-state memories, optical media, and magnetic media.

[0056] Although specific embodiments have been described above, these embodiments are not intended to limit the scope of the present disclosure, even where only a single embodiment is described with respect to a particular feature. Examples of features provided inTDCC#86233-WO-PCTthe disclosure are intended to be illustrative rather than restrictive unless stated otherwise. The above description is intended to cover such alternatives, modifications, and equivalents as would be apparent to a person skilled in the art having the benefit of this disclosure.

[0057] The scope of the present disclosure includes any feature or combination of features disclosed herein (either explicitly or implicitly), or any generalization thereof, whether or not it mitigates any or all of the problems addressed herein. Various advantages of the present disclosure have been described herein, but embodiments may provide some, all, or none of such advantages, or may provide other advantages.

[0058] In the foregoing Detailed Description, some features are grouped together in a single embodiment for the purpose of streamlining the disclosure. This method of disclosure is not to be interpreted as reflecting an intention that the disclosed embodiments of the present disclosure have to use more features than are expressly recited in each claim. Rather, as the following claims reflect, inventive subject matter lies in less than all features of a single disclosed embodiment. Thus, the following claims are hereby incorporated into the Detailed Description, with each claim standing on its own as a separate embodiment.

Claims

TDCC#86233-WO-PCTClaimsWhat is claimed is:

1. A method, comprising:updating a computer vision learning model trained to identify objects in images via transfer learning utilizing a first plurality of fluid sample images to result in a phase separation learning model,wherein a first portion of the first plurality of fluid sample images has a known presence of phase separation, andwherein a second portion of the first plurality of fluid sample images has a known absence of phase separation;dynamically cropping a second plurality of fluid sample images having unknown phase separation;receiving a plurality cropped images, each of the plurality of cropped images a cropped version of a respective one of the second plurality of fluid sample images;operating the phase separation learning model to determine a presence or absence of phase separation within the second plurality of fluid sample images and a probability of the presence or the absence of phase separation; andproviding a recommendation regarding the presence or the absence of phase separation within the second plurality of fluid sample images based on results of the operation of the phase separation learning model.

2. The method of claim 1, wherein dynamically cropping the second plurality of fluid sample images comprises calibrating each of the second plurality of fluid sample images relative to a reflection of light on a vial housing a respective fluid sample.

3. The method of claim 1, wherein operating the phase separation learning model to determine the probability of the presence or the absence of phase separation comprises determining a probability that a determined presence of phase separation comprises two or moreTDCC#86233-WO-PCTdistinct phases in a respective fluid sample image or an anomaly in the respective fluid sample image.

4. The method of claim 1, further comprising updating the computer vision learning model with a convolution determined during the training to apply to the second plurality of fluid sample images.

5. The method of claim 1, wherein updating the computer vision learning model comprises controlling how many parameters are updated each iteration.

6. The method of claim 1, wherein updating the computer vision learning model comprises controlling by how much parameters are updated each iteration.

7. The method of claim 1, comprising performing contrast enhancement on the second plurality of fluid sample images prior to inputting the second plurality of fluid sample images into the phase separation learning model.

8. The method of claim 1, wherein performing the dynamic cropping comprises operating a reflection learning analysis to identify light reflection in the first plurality of fluid sample images.

9. The method of claim 1, wherein operating the phase separation learning model comprises:operating a first learning layer to determine a plurality of features appearing in the second plurality of fluid sample images; andoperating a second learning layer to determine which of the plurality of features indicates phase separation in the second plurality of fluid sample images.

10. The method of claim 1, comprising retrofitting a previously trained computer vision learning model by applying identified phase separation image data to determine whether dataTDCC#86233-WO-PCTalready encoded into parameters of the previously trained first computer vision learning model are applicable to a phase separation determination.

Images