System and method for automated evaluation of artificial tissue

Abstract

Systems and methods for geometric coding of tissue phenotypes are described herein. In at least some aspects, the systems and methods include obtaining images of artificial tissue under a first set of conditions. The system and method further includes extracting a first region from the image such that the first region defines the artificial tissue within the image. The systems and methods further include extracting one or more geometric features from the first region, wherein the one or more geometric features provide a shape-based characterization of the artificial tissue. The systems and methods further include generating a first tissue descriptor based on the one or more geometric features. The first tissue descriptor may encode a phenotype of the artificial tissue under the first set of conditions. Still further, in some aspects, the systems and methods may include implementing quality control (QC) of artificial tissue.

Description

[0001] Related Applications

[0002] This application claims the benefit of U.S. Provisional Application No. 63 / 525,451 (filed July 7, 2023), which is incorporated by reference herein. BACKGROUND

[0003] Quality control (QC) of artificial or engineered tissues can be used to track the development of artificial tissues during maturation of the artificial tissues to determine whether to include the artificial tissues for further analysis or to exclude them. Existing QC methods typically rely on manual inspection of the artificial tissues during the maturation process. Such manual inspection results in user-induced bias that can lead to high variability in the quality of the artificial tissues generated. This can have a detrimental effect on any downstream tasks, such as drug discovery and development tasks, that rely on such artificial tissues. Furthermore, manual QC inspection can inhibit large-scale production of artificial tissues. Moreover, even when artificial tissues pass QC inspection, it is often difficult to identify or interpret variability that arises in experimental results when the artificial tissues are used within experimental settings or downstream tasks.

[0004] Accordingly, there is a need for new methods to assess the quality of artificial tissues (e.g., during maturation) and to encode the condition-dependent phenotypes of the artificial tissues. SUMMARY

[0005] According to one aspect of the disclosure, a method for geometric encoding of tissue phenotypes is provided. The method includes obtaining an image of an artificial tissue under a first set of conditions; extracting a first region from the image such that the first region encloses the artificial tissue within the image; and extracting one or more geometric features from the first region. The one or more geometric features provide a shape-based characterization of the artificial tissue. The method further includes generating a first tissue descriptor based on the one or more geometric features, wherein the first tissue descriptor encodes a phenotype of the artificial tissue under the first set of conditions.

[0006] According to further aspects of the present disclosure, a system for quality control (QC) of engineered tissue is provided. The system includes a bioreactor including a device configured for growing tissue and a sensor assembly configured to obtain one or more images of the tissue within the device. The system further includes a QC unit communicatively coupled to the bioreactor. The QC unit includes one or more processors configured to obtain, at a first predetermined time point, a first image of the tissue within the device from the bioreactor and extract one or more image segments from the first image, wherein each of the one or more image segments includes a region of interest of the tissue within the first image. The one or more processors of the QC unit are further configured to determine one or more tissue parameters from the one or more image segments, wherein the one or more tissue parameters are indicative of a physiological state of the tissue at the first predetermined time point. The one or more processors of the QC unit are further configured to output a QC report based on the one or more tissue parameters.

[0007] According to additional aspects of the present disclosure, a system for engineering tissue attachment grading is provided. The system includes a multi-task learning (MTL) network including an input convolutional neural network configured to receive an input image, a plurality of independent output networks each configured to estimate an attachment score associated with the input image, and a shared network coupled between the input convolutional neural network and each of the plurality of independent output networks. The system further includes a control unit communicatively coupled to the MTL network and including one or more processors configured to obtain an image of an engineered tissue grown within a device including a tissue scaffold for attachment to the engineered tissue and extract a first region from the image, wherein the first region of the image includes a first portion of the engineered tissue and a portion of the tissue scaffold. The one or more processors of the control unit are further configured to determine a plurality of attachment scores based on the first region using the MTL network and determine an attachment grade for the first region based on the plurality of attachment scores, wherein the attachment grade is indicative of a degree of attachment of the first portion of the engineered tissue to an attachment site of the device. The one or more processors of the control unit are further configured to output the attachment grade.

[0008] In accordance with the foregoing and the disclosure herein, the present disclosure includes the application of certain features or aspects with or through the use of a particular machine (e.g., a bioreactor). In various aspects, the bioreactor can include a device configured for growing or manipulating human tissue (e.g., human tissue such as muscle tissue, cardiac tissue, and / or skeletal muscle tissue). Additionally or alternatively, the bioreactor can include a sensor assembly configured to assess the quality of the engineered tissue (e.g., during maturation) and encode the conditional-dependent phenotype of the engineered tissue.

[0009] Further, the present disclosure includes the implementation of a transformation or reduction of certain artifacts to different states or things, e.g., a transformation or reduction of an image of human tissue (e.g., within a bioreactor, as sensed by a sensor assembly) to different states or things (e.g., the generation, creation, or otherwise development of a QC report based on initially received imaging data sensed from the engineered tissue (e.g., by one or more sensors)) that is then transformed or reduced into segments and features as part of an imaging analysis.

[0010] Still further, the present disclosure includes improvements in computer functionality or other technology, at least because the disclosure herein discloses systems and methods for reducing errors in underlying computing devices, e.g., by determining the quality of engineered tissue within a bioreactor and flagging issues therewith. The automated and image-based systems and methods herein reduce errors, particularly during the maturation period of engineered tissue, e.g., by reducing user assessment bias. This improves the underlying system because it relies on image-based processing that utilizes feature extraction to specifically identify engineered tissue or its phenotype, so it will have fewer false positives (as compared to conventional quality control observations). This also allows for the generation of more accurate and less error-prone outputs, and thus, the production of high-quality pharmaceutical products, such as therapeutic drugs.

[0011] Further, the present disclosure relates to improvements in other arts or fields of technology, at least because the systems and methods of the present disclosure provide or otherwise use a two-step segmentation process for performing segmentation of regions from images of engineered tissue. The two-step segmentation process provides an efficient and accurate method for identifying tissue regions. This in turn leads to improvements in the efficiency, accuracy, and overall performance of the underlying system. Specifically, when tissue regions are more accurately identified, machine learning models used in the extraction unit can be more efficiently trained in less time and with higher prediction accuracy. Further, as a result, the accuracy of the machine learning models is also improved, which improves the efficiency and performance of downstream drug discovery and development tasks.

[0012] Still further, these systems and methods, when deployed on an underlying computing device, allow the systems and methods of the present disclosure to perform with fewer iterations and using fewer computing resources as compared to systems and methods related to the prior art. That is, the present disclosure describes improvements in the functioning of a computer itself or "any other art or technology field" in that the underlying computing device can generate tissue descriptors based on geometric features, where, for example, a given tissue descriptor can encode the phenotype of an artificial tissue thereunder. The encoding of tissue phenotypes provided by the tissue descriptors provides a powerful and discriminative feature that can be effectively used for a number of downstream drug discovery and development tasks, including reducing the number of testing cycles and resources expended thereby. That is, the improvements provided by generating tissue descriptors based on geometric features allows the underlying computer system to utilize fewer processing and memory resources as compared to prior art systems and methods. This is at least because such features can be used to generate or determine the phenotype of an artificial tissue under a set of conditions without the need for various testing and / or empirical computer simulations across a wide range of testing using a number of computing cycles and data. In other words, the systems and methods of the present disclosure improve upon the prior art in that the systems and methods of the prior art require an empirical or trial-and-error approach that can involve real-world trials on human tissue that can incur and require substantial database and memory utilization and processor usage to obtain similar real-world or simulated results with the same or similar results. On the other hand, the disclosed systems and methods describe the generation and / or use of bioreactors for growing and testing tissues for defining a limited set of data specific to the tissue (e.g., human engineered tissue) that requires less memory usage and / or processing utilization as compared to conventional approaches that use or require a large set of unknown, potentially irrelevant data. Moreover, the disclosure herein allows for the identification and use of high-fidelity data sets that further reduce the need for additional computing cycles. Furthermore, the tissue descriptors can complement existing tissue assays to improve the interpretability of such models in the presence of experimental variability (e.g., by providing a quantitative explanation of the physiological or morphological state of an artificial tissue).

[0013] Still further, the present disclosure includes particular features beyond those that are well-known, routine, conventional in the art, and / or otherwise adds steps that would otherwise limit the present disclosure to a particular useful application (e.g., systems and methods for automated assessment of artificial tissues) that can be used, for example, to automate the assessment of artificial or engineered tissue quality (quality control) based on one or more tissue parameters and / or to perform automated assessment of artificial or engineered tissue phenotypes based on one or more geometric tissue features that can improve downstream tasks such as drug discovery and development.

[0014] Advantages will become apparent to those skilled in the art upon reading the following description of the preferred embodiments in conjunction with the accompanying drawings. As will be realized, the embodiments are capable of having other different embodiments and its details are capable of being modified in various respects. Accordingly, the drawings and description are to be regarded as illustrative in nature and not as restrictive. BRIEF DESCRIPTION OF DRAWINGS

[0015] Embodiments of the present disclosure will now be described, by way of example only, with reference to the attached figures, wherein:

[0016] Figure 1 A system for automated assessment of artificial tissue is shown in accordance with an aspect of the present disclosure;

[0017] Figure 2 An image of artificial tissue is shown in accordance with an embodiment of the present disclosure;

[0018] Figure 3 A first segmentation step of a morphological based segmentation pipeline by a segmentation unit of a system for automated assessment of artificial tissue is illustrated in accordance with an embodiment of the present disclosure;

[0019] Figure 4 A second segmentation step of a morphological based segmentation pipeline by a segmentation unit of a system for automated assessment of artificial tissue is illustrated in accordance with an embodiment of the present disclosure;

[0020] Figure 5 Features extracted from a tissue region of a segmented image are shown in accordance with an aspect of the present disclosure;

[0021] Figure 6 A decomposition of a well-cropped image of artificial tissue for tissue attachment grading is shown in accordance with an embodiment of the present disclosure;

[0022] Figure 7 A hierarchical tissue attachment grading is shown in accordance with an embodiment of the present disclosure;

[0023] Figure 8 A cropped region of three artificial tissues with different predicted tissue attachment grading is shown in accordance with an embodiment of the present disclosure;

[0024] Figures 9A-9B A multi-task learning (MTL) network for tissue attachment grading is shown in accordance with an aspect of the present disclosure;

[0025] Figure 10 A sequence of images of artificial tissue during maturation period is shown in accordance with an embodiment of the present disclosure;

[0026] Figure 11A portion of a QC report obtained during the maturation period of the artificial tissue shown; Figure 10 A portion of a QC report obtained during the maturation period of the artificial tissue shown;

[0027] Figure 12 A portion of a QC report obtained during the maturation period of the artificial tissue shown; Figure 10 A tissue attachment grading portion of a QC report obtained during the maturation period of the artificial tissue shown;

[0028] Figure 13 Variations in tissue phenotype as encoded by geometric tissue descriptors are exemplified according to one embodiment of the present disclosure;

[0029] Figure 14 A method for quality control (QC) of artificial tissue according to one aspect of the present disclosure is shown;

[0030] Figure 15 A method for engineered tissue attachment grading according to one aspect of the present disclosure is shown;

[0031] Figure 16 A method for geometric encoding of tissue phenotype according to one aspect of the present disclosure is shown;

[0032] Figure 17 A method for determining effects associated with variations in phenotype according to one embodiment of the present disclosure is shown;

[0033] Figure 18 A bioreactor system according to embodiments of the present disclosure is shown; and

[0034] Figure 19 An example computing system according to embodiments of the present disclosure is shown. TECHNICAL FIELD

[0035] The present disclosure relates to artificial or engineered tissue. In particular, but not exclusively, the present disclosure relates to automated assessment of artificial or engineered tissue quality based on one or more tissue parameters. Additionally, but not exclusively, the present disclosure relates to automated assessment of artificial or engineered tissue phenotype based on one or more geometric tissue features. DETAILED DESCRIPTION

[0036] Figure 1 A system 100 for automated assessment of artificial tissue according to one aspect of the present disclosure is shown.

[0037] The system 100 includes a bioreactor 102, a segmentation unit 104, an extraction unit 106, a quality control (QC) unit 108, and an assay unit 110 (alternatively referred to as an analysis unit, an evaluation unit, or an assessment unit). In various aspects, the bioreactor is used to grow tissue (e.g., engineered tissue), such as human tissue, including but not limited to muscle tissue, cardiac tissue, skeletal muscle tissue, or other human tissue. The bioreactor 102 includes an artificial tissue 112 under a first set of conditions and a sensor assembly 114 configured to obtain an image 116 of the artificial tissue 112 at a first time point. The segmentation unit 104 extracts one or more regions (or segments) from the image 116 of the artificial tissue 112. Features or parameters associated with the artificial tissue 112 are extracted from the one or more regions by the extraction unit 106. The QC unit 108 uses the features or parameters extracted by the extraction unit 106 to generate a QC report 118 including information indicative of a state (e.g., a physiological or morphological state) of the artificial tissue 112 under the first set of conditions at the first time point. The assay unit 110 uses the features or parameters extracted by the extraction unit 106 to generate a tissue descriptor 120 encoding a phenotype of the artificial tissue 112 under the first set of conditions at the first time point. Figure 1 Further shown are a first subsystem 122 and a second subsystem 124. The first subsystem 122 includes the segmentation unit 104 and the extraction unit 106. The second subsystem 124 includes the segmentation unit 104, the extraction unit 106, the QC unit 108, and the assay unit 110.

[0038] Figure 1 The illustrated system 100 includes an automated system for the qualitative and quantitative assessment of artificial tissue. Beneficially, the system 100 can quantify the quality of artificial tissue and flag artificial tissue that does not pass quality control conditions. Automation of quality control, particularly during the maturation period of artificial tissue, helps to reduce the requirement for human involvement, thereby reducing user assessment bias and improving the scalability of the system 100. Furthermore, the geometric tissue descriptor utilized by the system 100 provides an efficient and discriminative method for encoding tissue phenotype. As such, the geometric tissue descriptor can be used to support a number of downstream tasks, such as drug discovery and development. Moreover, the tissue descriptor can complement existing tissue assays to improve the interpretability of such models in the presence of experimental variability (e.g., by providing a quantitative explanation of the physiological or morphological state of the artificial tissue).

[0039] Figure 1 Each of the illustrated units (e.g., the segmentation unit 104, the extraction unit 106, etc.) can be implemented using one or more computing systems, such as Figure 19 The illustrated computing systems (other computing systems can also be used) are implemented. The one or more functional units discussed herein (e.g., as with respect to theFigure 1 The computing tasks performed by the one or more functional units described can be implemented as software, hardware, firmware or a combination thereof, which is described in more detail below. The computing tasks can also be performed by at least one of the systems described, or alternatively, by a combination of some of the systems described, or by a combination of some of the systems described in combination with other devices. These configurations can be implemented in hardware, software, firmware or a combination thereof, without departing from the scope of the present disclosure. The use of computer-based systems allows for a great variety of possible configurations, combinations, and partitions of tasks and functionality between and among components. Computer-implemented operations can be performed on a single component or across multiple components. Computer-implemented tasks and / or operations can be performed sequentially or in parallel. Data and instructions can be stored in a single storage medium or across multiple storage media.

[0040] In one embodiment, the segmentation unit 104 and the extraction unit 106 form part of a first subsystem 122 that is separate from the bioreactor 102, the QC unit 108, and the assay unit 110. The first subsystem 122 includes hardware and / or processing logic configured to receive or otherwise obtain data (e.g., images 116) from the bioreactor 102 and output the data to the QC unit 108 and / or the assay unit 110. The first subsystem 122 can be physically co-located with the bioreactor 102, the QC unit 108, and / or the assay unit 110, or the first subsystem 122 can be remotely located (e.g., the first subsystem 122 is deployed as a cloud service and is accessible via one or more application programming interfaces (APIs)).

[0041] In a further embodiment, the segmentation unit 104, the extraction unit 106, the QC unit 108, and the assay unit 110 form part of a second subsystem 124 that is separate from the bioreactor 102. The second subsystem 124 is communicatively coupled to the bioreactor 102 and includes hardware and / or processing logic configured to receive or otherwise obtain data (e.g., images 116) from the bioreactor 102. The second subsystem 124 can be physically co-located with the bioreactor 102, or the second subsystem 124 can be remotely located (e.g., the second subsystem 124 is deployed as a cloud service and is accessible via one or more application programming interfaces (APIs)).

[0042] As will be described below with respect to Figure 18In more detail, bioreactor 102 includes one or more means for growing and / or maintaining artificial or engineered tissue under one or more conditions (e.g., artificial tissue 112 under a first set of conditions). Artificial tissue 112 includes engineered muscle tissue. In one embodiment, the engineered tissue is engineered heart tissue; more specifically, the engineered heart tissue is a 3D construct of human induced pluripotent stem cell-derived cardiomyocytes (CMs) (iCells, Fujifilm / CDI). In an alternative embodiment, the engineered muscle tissue is engineered skeletal muscle tissue. Artificial tissue 112 within bioreactor 102 is associated with one or more conditions, such as reference or control conditions and perturbation conditions. Generally, a reference condition refers to a condition that provides a baseline comparison to the perturbation conditions. In embodiments, reference conditions are conditions associated with a control setting or environment. Reference conditions may correspond to artificial or engineered tissue in its default, natural, or unaltered state (i.e., without drug or agent dosage). Alternatively, reference conditions may correspond to carrier-treated artificial tissue. A perturbation condition refers to a condition under which an artificial tissue is disturbed in some way. Examples of perturbation conditions include the administration of a drug or compound (i.e., the perturbed animal), disease state, different cell lines, physical perturbation applied to the artificial tissue, or changes in the environment of the artificial tissue. In the case of perturbations involving drugs or compounds, these conditions may also be referred to as therapeutic conditions and are further associated with the drug or compound's effects, such as mechanisms of action or toxicity. Given the range of different perturbation conditions, an artificial tissue may be associated with more than one perturbation condition (e.g., a disease-engineered tissue that has been treated with a specific compound).

[0043] Sensor assembly 114 is configured to acquire an image, such as image 116, of the artificial tissue 112 at a first time point (e.g., 1 day after inoculation, 2 weeks after inoculation, etc.). As will be described in more detail below, image 116 is segmented and analyzed to determine one or more features or descriptors associated with the artificial tissue 112 under one or more conditions at the first time point.

[0044] Figure 2 An image 200 of an artificial tissue according to an embodiment of the present disclosure is shown.

[0045] Image 200 is from a bioreactor (such as...) Figure 1 Bright-field images of artificial tissue within the apparatus (bioreactor 102) are shown. Multiple regions are labeled within image 200, including well region 202, first tissue scaffold region 204, second tissue scaffold region 206, and tissue region 208. These regions are automatically identified and used by the bioreactor 102, as will be discussed in more detail below. Figure 1downstream tasks (e.g., quality control, estimation of the organization phenotype, etc.) performed by the system 100.

[0046] The well region 202 constitutes a portion of the image 200 that demarcates a well within which artificial tissue grows from cells seeded therein. The first and second tissue scaffold regions 204 and 206 constitute portions of the image 200 that include tissue scaffolds of devices containing artificial tissue. Of the devices captured within the image 200, two tissue scaffolds are shown, one within each tissue scaffold region. Each tissue scaffold shown in the image 200 includes flexible polymer filaments disposed across a pore. The tissue scaffolds are configured to: (a) permit attachment of artificial tissue formed therebetween; and (b) deform in response to contractile forces exerted thereon by the artificial tissue.

[0047] The tissue region 208 demarcates artificial tissue within the image 200 (i.e., the black region positioned at the center of the image 200). As can be seen, the tissue region 208 is located within the well region 202 and overlaps the first and second tissue scaffold regions 204 and 206. Due to the large variability in the size and shape of the captured tissue across images and extraneous structures within the images (such as those seen in the image 200), efficient and accurate automated extraction of the tissue region 208 from the image 200 is a challenging task.

[0048] Referring again to Figure 1 , the segmentation unit 104 is configured to extract one or more of the regions described above with respect to Figure 2 The segmentation unit 104 utilizes a two-step segmentation process to perform segmentation of regions from images of artificial tissue. A first segmentation step segments a well region from the image 116 (i.e., the well region 202 shown in Figure 2 A second segmentation step segments a tissue region from the image 116 cropped according to the well region identified in the first segmentation step (i.e., the tissue region 208 shown in Figure 2 The segmentation unit 104 is configured to perform further segmentation steps to extract tissue scaffold regions (e.g., the first and second tissue scaffold regions 204 and 206 shown in Figure 2

[0049] Advantageously, utilizing a two-step segmentation process provides for identifying Figure 2 ​An effective and accurate method of delineating tissue regions is shown. Specifically, by utilizing the well region as a common reference frame, the tissue scaffold region and the tissue region can be more accurately identified. This in turn leads to improvements in efficiency, accuracy, and overall performance of the extraction unit 106, the QC unit 108, and the assay unit 110, which depend on the segmentation generated by the segmentation unit 104. Specifically, when the tissue region is more accurately identified, the machine learning models used in the extraction unit 106 (as described below) can be more effectively trained in less time.

[0050] Any suitable image segmentation method can be used to segment the well region, and subsequently the tissue region. Example methods include clustering-based segmentation (e.g., fuzzy c-means clustering, Gaussian mixture modeling, etc.), neural network-based segmentation methods, and shape-based segmentation. In one embodiment, a morphological-based segmentation pipeline is used to implement a two-step segmentation process. Beneficially, this pipeline is unsupervised, requiring no prior training, and is yet more robust to noise. In comparison to supervised machine learning models (e.g., trained neural networks), this pipeline is also able to quickly and effectively segment various regions from an input image, and has lower processor and memory requirements.

[0051] Figure 3 A first segmentation step of a morphological-based segmentation pipeline by a segmentation unit (such as the segmentation unit 104 of the system 100) according to one embodiment of the present disclosure is illustrated.

[0052] Figure 3 An input image 302, an equalized image 304, a first binarized image 306, an eroded image 308, a first filled image 310, a first single-object image 312, and a rough crop image 314 are shown. Figure 3 A second binarized image 316, an edge image 318, a second filled image 320, a second single-object image 322, an image object with center image 324, an adjusted image 326, and an output image 328 are further shown.

[0053] Figure 3 The illustrated morphological segmentation pipeline includes a plurality of sequential morphological operations and other image processing operations. Generally, the pipeline includes two stages. At the first stage (from the input image 302 to the rough crop image 314), a rough crop of the well region is extracted from the input image 302, and extraneous structures are removed. By identifying the well region within the input image 302, the image can be rotated to align with the axis of the well region, improving the fine-grained segmentation of the well region during the second stage. The second stage (from the second binarized image 316 to the output image 328) performs a fine crop of the well region by utilizing edge-based features within the rough crop.

[0054] At the first stage, an input image 302 is provided as input to the segmentation pipeline. As described above, this image comprises a brightfield image of engineered tissue, and includes several irrelevant structures as well as a region of interest (e.g., a well region, a tissue region, etc.).

[0055] A histogram equalization operation is then applied to the input image 302 and results in an equalized image 304. The histogram equalization operation adjusts the histogram of pixel intensities within the input image 302 such that the intensity values are more evenly distributed across the full range of possible intensities. As a result, the global contrast of the equalized image 304 is improved compared to that of the input image 302. Beneficially, this helps to highlight details within the input image 302, and also allows later morphological operations within the pipeline to be applied consistently (i.e., the parameter values of the operations do not need to be adapted on a per-image basis).

[0056] A binarization operation is applied to the equalized image 304, resulting in a first binarized image 306. As is known, binarization is the process of converting a grayscale image to its binary representation. That is, all pixels within the grayscale image are replaced with either a "0" or a "1" depending on the pixel intensity value and the thresholding method used. Typically, if a pixel has an intensity value above a predetermined threshold value, it is assigned the binary value of "1" (i.e., the pixel is considered foreground), otherwise it is assigned the binary value of "0" (i.e., the pixel is considered background). The predetermined threshold value of the equalized image 304 is determined using the Otsu method. Alternatively, the predetermined threshold value is set at a predetermined value (e.g., the median pixel intensity value), or using an optimization method such as variational min-max optimization. In yet further alternatives, a clustering method such as k-means clustering, Gaussian mixture model, or fuzzy k-means clustering is used to identify foreground and background regions of the image.

[0057] A first cycle of morphological operations is then applied to the first binarized image 306 to identify an approximate well region and remove noise. The morphological operations include an erosion of the first binarized image 306 (as shown by the eroded image 308), followed by a fill operation applied to the eroded image 308 (as shown by the first single-object image 312).

[0058] Erosion is a morphological operation that applies a structuring element, a matrix defining the shape of a neighborhood around a pixel that will be used to process that pixel, to all pixels within an image to generate an output. The erosion operation sets the value of a pixel in the output image to the minimum value of all pixels within the neighborhood of that pixel in the input image (where the neighborhood is defined by the structuring element). Thus, the erosion operation applied to the first binarized image 306 removes small objects and lines within the first binarized image 306 (e.g., due to noise), such that only larger structures remain within the eroded image 308. In one implementation, the erosion operation is performed using a cross-shaped structuring element with square connectivity of one.

[0059] The fill operation applied to the eroded image 308 performs a hole-filling (or region-filling) algorithm on the eroded image 308 to produce a first filled image 310. The hole-filling algorithm fills holes within "hollow" shapes in the eroded image 308. A hole is a contiguous background region of the eroded image 308 (i.e., all pixels within the region have a value of "0") surrounded by a connected boundary of foreground pixels (i.e., the background region is bounded by pixels having a value of "1"). The hole-filling algorithm fills these background regions with foreground pixels.

[0060] After the fill operation is applied, the first filled image 310 includes multiple connected components - collections of foreground pixels (with a pixel value of "1") that are connected in some way (e.g., pixels are connected if they share a common edge or corner). The largest of these connected components corresponds to a rough segmentation of the well, as seen in the first filled image 310. Thus, the largest connected component is identified from the first single-object image 312 and used as a rough mask of the well region within the input image 302. Since the well region can not be aligned with the axis of the image, the long axis of the largest connected component is determined (as exemplified by the line within the first single-object image 312) and the angle between this long axis and the vertical axis of the input image 302 is used to calculate the degree of rotation to be applied to the input image to align the well region with the image axis. The first filled image 310 and the input image 302 are rotated by the calculated angle, and the input image 302 is cropped to the bounding box of the largest component within the (rotated) filled image. This rough cropping of the well region is shown in the rough cropped image 314.

[0061] At the second stage, the roughly cropped image 314 is binarized to produce a second binarized image 316. An edge detection operation is then applied to produce an edge image 318. The edge image 318 corresponds to a transformation of the second binarized image 316 with connected edges. The edge image 318 is generated using a pair of dilation operations. A first dilation operation is applied to the second binarized image 316 using a rectangular structuring element that extends along a first direction (e.g., horizontal), and a second dilation operation is applied to the output of the first dilation operation using a rectangular structuring element that extends along a second direction (e.g., horizontal). In one implementation, the first structuring element is a rectangular structuring element of shape (5 x 1), and the second structuring element is a rectangular structuring element of shape (1 x 5).

[0062] The edge image 318 is filled to produce a second filled image 320 using similar operations as described above with respect to the first filled image 310 and the first single object image 312, and the largest connected component is extracted as shown in a second single object image 322. An object image with center 324 shows the center of the single connected component of the second single object image 322 superimposed on the input image 302 (masked using the second single object image 322). An adjusted image 326 shows the input image 302 rotated according to the transformation described above with respect to the first single object image 312 and the roughly cropped image 314. The single connected component shown in the object image with center 324 is superimposed on the adjusted image 326.

[0063] The output image 328 includes the portion of the input image 302 that contains the well region, such that the vertical axis of the well region is aligned with the vertical axis of the output image 328. The output image 328 is then used to extract the tissue scaffold region (e.g., the first tissue scaffold region 204 and the second tissue scaffold region 206 shown below), as described in more detail below. Figure 2 The output image 328 is also used as input for the second step of the segmentation pipeline, as shown below. Figure 4

[0064] Figure 4 A second segmentation step of a morphology-based segmentation pipeline performed by a segmentation unit, such as the segmentation unit 104 of the system 100, according to one embodiment of the present disclosure is illustrated.

[0065] Figure 4 An input image 402, an adjusted image 404, a denoised image 406, a blurred image 408, a binarized image 410, a closed image 412, an eroded image 414, a dilated image 416, a filled image 418, a single object image 420, a tissue region image 422, and a segmentation boundary image 424 are shown.

[0066] ​Figure 4 The illustrated morphological segmentation pipeline includes a plurality of sequential morphological operations and other image processing operations. The morphological operations accurately identify tissue boundaries within the input image 402 that correspond to tissue regions. The use of morphological operations can allow for accurate extraction of geometric descriptors from the tissue regions, thereby improving the performance of downstream tasks that utilize these descriptors, such as quality control and estimation of tissue phenotypes.

[0067] The input image 402 corresponds to the output image of the first segmentation step of the morphological-based segmentation pipeline (i.e., Figure 3 The illustrated output image 328).

[0068] The adjusted image 404 is generated from the input image 402 to reduce the intensity of the well boundary (as seen by the black border surrounding the input image 402). The generation of the adjusted image 404 helps to improve the performance of the tissue segmentation process because the pixel intensity values of the well boundary are in a similar range as the pixel intensity values of the tissue regions within the input image 402. The adjusted image 404 is generated by identifying a 50-pixel boundary region within the input image 402 and then replacing any pixel within the boundary region that has an intensity value below 0.1 (i.e., below 10% of the input image 402 pixel intensity range) with a constant value. In one implementation, the constant value is the mean pixel intensity value of the input image 402, plus the standard deviation of the pixel intensity values of the input image 402.

[0069] A denoising algorithm is applied to the adjusted image 404 to produce a denoised image 406. Any suitable denoising algorithm can be used, such as Gaussian blur, median filtering, etc. In one implementation, a non-local means method is used with a filter strength of 100, a template window size of 50, and a search window size of 21.

[0070] The denoised image 406 is then blurred using an asymmetric filter to generate a blurred image 408. The use of the asymmetric filter filters out the tissue scaffolding from the denoised image 406 by reducing the intensity of the pixels within the denoised image 406 that correspond to the tissue scaffolding. Since the tissue scaffolding appears as substantially horizontal lines in the denoised image 406, the use of the asymmetric filter filters out the tissue scaffolding without removing important tissue boundary details. More specifically, the asymmetric filter suppresses any horizontal edges and subsequently enhances any vertical edges. The asymmetric filter initially convolves the denoised image 406 with a kernel of the form:

[0071]

[0072] where ω1is normalized prior to the convolution such that the denoised image 406 is convolved with the normalized kernel ω1', where The convolution result is then convolved with a kernel of the form:

[0073]

[0074] where ω2is normalized prior to convolution such that the denoised image 406 is convolved with ω1' to generate a blurred image 408, where .

[0075] The blurred image 408 is then binarized to generate a binarized image 410. A morphological closing operation is applied to the binarized image 410 to generate a closed image 412. The morphological closing operation includes a dilation operation followed by an erosion operation, where both operations use the same structuring element. The dilation operation is the opposite of the erosion operation: the value of a pixel in the output image is set to the maximum value of all pixels within the neighborhood of that pixel in the input image (where the neighborhood is defined by the structuring element). The purpose of the morphological closing operation is to fill in (or close) small holes within the binarized image 410 while maintaining the geometric shape (shape and size) of larger objects within the binarized image 410, such as connected components that roughly correspond to tissue regions. In one implementation, the morphological closing operation is performed with a square structuring element with connectivity of one (i.e., only nearest neighbors are considered, elements connected diagonally are not considered as nearest neighbors).

[0076] The closing operation is followed by separate erosion and dilation operations. That is, the closed image 412 is eroded to generate an eroded image 414, and the eroded image 414 is subsequently dilated to generate a dilated image 416. Unlike a morphological opening operation in which the structuring element is the same for both erosion and dilation, the structuring element used to produce the eroded image 414 is different from the structuring element used to produce the dilated image 416. The erosion and dilation operations remove small objects and lines from the closed image 412 while maintaining the geometric shape (shape and size) of larger objects within the closed image 412, such as connected components that roughly correspond to tissue regions. In one implementation, the erosion and dilation operations are performed using a cross-shaped structuring element with square connectivity of one.

[0077] The fill operation is then applied to the dilated image 416 to produce a filled image 418. A largest connected component is then extracted from the filled image 418 to produce a single object image 420. Thus, the single object image 420 corresponds to a segmentation map of the tissue region within the input image 402. This is illustrated in a tissue region image 422 and a segmentation boundary image 424, which show the tissue region extracted from the input image 402 and the boundary of the tissue region superimposed on the input image 402 (i.e., the boundary of the segmentation of the image region within the single object image 420), respectively. The boundary of the tissue region is shown as a white line that bounds the tissue within the segmentation boundary image 424. As can be seen, the two-step segmentation process accurately extracts the tissue region despite the presence of extraneous structures and noise within the image.

[0078] Referring again to Figure 1 After the region is segmented by the segmentation unit 104, the extraction unit 106 extracts one or more features (or segmentation maps) from the region produced by the segmentation unit 104.

[0079] Figure 5 Features extracted from a tissue region of a segmented image according to one aspect of the present disclosure are illustrated.

[0080] Figure 5 A tissue region 502 bounded by a boundary 504 is shown. Figure 5 Further shown are a centroid 506 of the tissue region 502, a first boundary point 508 along the boundary 504 of the tissue region 502, and a second boundary point 510 along the boundary 504 of the tissue region 502. A plurality of intermediate points 512-1, 512-2, 512-3, 512-4 are spaced between the first boundary point 508 and the second boundary point 510. Figure 5 Further shown are a first tissue scaffold region 514 and a second tissue scaffold region 516.

[0081] A first expanded view 518 of the first tissue scaffold region 514 shows a portion of the tissue region 502 surrounding a first tissue scaffold 520 of a device within which an image of artificial tissue was taken. It will be appreciated by those skilled in the art that, for ease of illustration, the first tissue scaffold 520 is superimposed on the tissue region 502 and does not form a portion of the tissue region 502 extracted by a segmentation unit (e.g., the segmentation unit 104 in FIG. 4). Figure 1 The first expanded view 518 shows a first intersection point 522 associated with a first edge of the first tissue scaffold 520 and a second intersection point 524 associated with a second edge of the first tissue scaffold 520.

[0082] A second extended view 526 of the second tissue scaffold region 516 shows a portion of the tissue region 502 surrounding the second tissue scaffold 528 of the device, within which an image of artificial tissue has been captured. Those skilled in the art will understand that, for illustrative purposes, the second tissue scaffold 528 superimposes the tissue region 502 and does not form a segmented structure consisting of units (e.g., such as...). Figure 1 A portion of the tissue region 502 extracted by the segmentation unit 104 in the second view 528. The second extended view 526 shows a third intersection 530 associated with a first edge of the second tissue scaffold 528 and a fourth intersection 532 associated with a second edge of the second tissue scaffold 528.

[0083] Organizational area 502 is used in accordance with the above-mentioned Figure 3 and Figure 4 The described segmentation method (e.g., by) Figure 1 The segmentation unit 104 in the image is extracted from the artificial tissue image. Therefore, the tissue region 502 corresponds to a single connected component or object within the segmented or labeled image. For example, the tissue region 502 corresponds to a connected region of a foreground pixel or a pixel assigned a value indicating that it belongs to a part of the tissue region. This will be described in more detail below. Figure 5 Points within or near the illustrated tissue region 502 (e.g., centroid 506, first boundary point 508, etc.) form landmarks for estimating one or more geometric features or parameters of the artificial tissue, such as tissue length, tissue width, tissue area, etc. These features are used to form a tissue descriptor, which is used in downstream tasks of quality control and tissue phenotypic coding.

[0084] These points (e.g., centroid 506, first boundary point 508, etc.) are determined from tissue region 502 by initially calculating the central moments of tissue region 502. The centroid 506 of tissue region 502 is then determined using the central moments. The remaining points are determined by identifying the major axis of tissue region 502. Figure 5 As seen in the example, tissue region 502 is not necessarily oriented towards the image axis; that is, the orientation of tissue region 502 is not necessarily consistent with the orientation of the image. The orientation of tissue region 502 is determined by calculating the inertia tensor of tissue region 502 based on its central moments. The eigenvector of the inertia tensor, associated with the largest eigenvalue of the inertia tensor, corresponds to the major / principle axis of tissue region 502. Therefore, the minor axis of tissue region 502 is determined to be perpendicular to the major axis of tissue region 502. A line is then fitted from the centroid 506 along the major axis to the boundary 504 of tissue region 502 to identify the first boundary point 508 and the second boundary point 510.

[0085] A plurality of intermediate points 512-1, 512-2, 512-3, 512-4 are then determined. The plurality of intermediate points includes a first intermediate point 512-1 associated with the first tissue scaffold region 514 and a fourth intermediate point 512-4 associated with the second tissue scaffold region 516, as well as a second intermediate point 512-2 and a third intermediate point 512-3. The second intermediate point 512-2 and the third intermediate point 512-3 are determined by identifying three equidistant points between the first boundary point 508 and the second boundary point 510. Those skilled in the art will appreciate that these equidistant points will include the centroid 506, which is identified separately in Figure 5 for ease of illustration. Thus, references to the plurality of intermediate points are taken to mean the plurality of points (i.e., the plurality of intermediate points 512-1, 512-2, 512-3, 512-4 and the centroid 506 shown in Figure 5 ). In Figure 5 five intermediate points are determined. Alternatively, between 5 and 50 intermediate points are determined between the first boundary point 508 and the second boundary point 510.

[0086] The first tissue scaffold region 514 and the second tissue scaffold region 516 correspond to predetermined bounding boxes. That is, since the segmented images from which Figure 5 are extracted are of constant size and resolution (i.e., the segmentation process described with respect to Figure 3 and Figure 4 produces images of constant or uniform size / resolution), the locations of the first tissue scaffold region 514 and the second tissue scaffold region 516 are substantially uniform across different images. Thus, the first tissue scaffold region 514 and the second tissue scaffold region 516 are extracted from images (e.g., the image 200 shown in Figure 2 or the input image 302 shown in Figure 3 ) by extracting portions of the images that are located within the predetermined bounding boxes. The predetermined bounding boxes can be identified from one or more training (or example) images using a manual or automatic identification process (e.g., image registration and feature detection). In one implementation, where the images have dimensions (height x width) of 1500px x 500px such that the tissue region extends vertically within the image, the bounding boxes have dimensions (height x width) of 400px x 500px, the bounding box for the first tissue scaffold region 514 has a top-left anchor point at image location (0, 185), and the bounding box for the second tissue scaffold region 516 has a top-left anchor point at image location (0, 965).

[0087] The first intersection point 522 shown within the first tissue scaffold region 514 is determined from the inner edge of the first tissue scaffold 520 (i.e., the edge of the first tissue scaffold 520 that is closest to the centroid 506). The second intersection point 524 shown within the second tissue scaffold region 516 is determined from the inner edge of the second tissue scaffold 522 (i.e., the edge of the second tissue scaffold 522 that is closest to the centroid 506). Figure 5The first intersection 522 is determined as the intercept (e.g., the y-intercept if the tissue extends vertically within the image) of the first reference line with the edge of the image or image region. Once the first reference line is fitted, the gradient of the first reference line is determined, and the second intersection 524 is determined from a second reference line having the same gradient as the first reference line (i.e., the two reference lines are parallel). The second reference line is fitted such that it passes through or contacts the point of the inner edge of the first tissue scaffold 520 that is farthest from the first reference line. The intercept value (e.g., the y-intercept) of the second reference line corresponds to the second intersection 524. In effect, the second intersection 524 corresponds to the fixed point of the first tissue scaffold 520 to the well, such that the absolute distance between the first intersection 522 and the second intersection 524 approximates the curvature or tension of the first tissue scaffold 520. This measure is used to determine the passive tension of the tissue scaffold 520, i.e., the estimated tension exerted by the tissue on the first tissue scaffold 520 when the tissue is in a resting state.

[0088] Using the reference points (e.g., the plurality of intermediate points 512-1, 512-2, 512-3, 512-4, the centroid 506, the first intersection 522, etc.), one or more tissue parameters are extracted that can be subsequently used to determine descriptors of the tissue. The tissue parameters include geometric parameters (or geometric features) such as tissue length, tissue width, tissue area, and tissue volume. The tissue parameters further include non-geometric parameters (or non-geometric features) such as passive tension and one or more irregularity ratings.

[0089] Tissue length is a geometric feature or parameter of the tissue corresponding to the longitudinal length of the artificial tissue (i.e., the length of the long axis of the artificial tissue). The length is determined by calculating the distance between the first boundary point 508 and the second boundary point 510. In one embodiment, based on the resolution of the tissue image from which the tissue region 502 is obtained, a pixel-based length is converted to pm.

[0090] Tissue width is a geometric feature or parameter of the tissue corresponding to a measurement of the tissue along a short axis of the artificial tissue (i.e., a cross-sectional measurement). The width of the artificial tissue is determined at each of a plurality of intermediate points 512-1, 512-2, 512-3, 512-4, and the centroid 506, generating a plurality of width measurements. At a first point along the longitudinal center axis of the artificial tissue (e.g., the centroid 506), the tissue width is determined by extending a line at the first point along the short axis of the artificial tissue (which is perpendicular to the long axis) and identifying a pair of points at which the line intersects the boundary 504. Thus, the distance between the pair of points corresponds to the width at the first point. This operation is repeated for each of the plurality of intermediate points 512-1, 512-2, 512-3, 512-4, and the centroid 506, resulting in a plurality of widths of the artificial tissue. In one embodiment, pixel-based lengths are converted to pm based on the resolution of the tissue image from which the tissue region 502 is obtained. In one embodiment, five tissue widths are determined from five intermediate points. Alternatively, five to fifty tissue widths are determined from five to fifty intermediate points spaced between the first boundary point 508 and the second boundary point 510.

[0091] Tissue area is a geometric feature or parameter corresponding to an estimated surface area of the artificial tissue captured within the image. The tissue area is determined from the number of pixels within the tissue region 502. In one embodiment, pixel-based areas are converted to pm based on the resolution of the tissue image from which the tissue region 502 is obtained. 2 or mm 2 .

[0092] Tissue volume is a geometric feature or parameter corresponding to an estimated volume of the artificial tissue. A slice volume model is used to estimate a volume measurement from a 2-dimensional segmentation of the tissue (i.e., the tissue region 502). In general, the slice volume model estimates a volume of a slice of the artificial tissue at a respective location or point along the longitudinal axis based on an estimated cross-sectional area of the artificial tissue at the respective location or point. The estimated cross-sectional area is determined from the width of the artificial tissue at the respective location or point. As will be described in greater detail below, the estimated cross-sectional area is determined according to a model parameterized by the width of the artificial tissue and a predetermined tissue elongation parameter.

[0093] The tissue volume is determined as a piecewise combination of local slice volumes determined at n locations or points evenly distributed along the longitudinal line extending between the first boundary point 508 and the second boundary point 510. Those skilled in the art will appreciate that these locations are omitted for brevity from Figure 5 In one embodiment, n > 10, more particularly 10 < n < 100, and again more particularly n = 40. For the i-th slice (i.e., at the i-th point along the longitudinal line), the slice width w iis computed as the distance along the short axis extension line of the artificial tissue at point i and between the pair of points where this line intersects the boundary 504. The tissue volume V is then computed as:

[0094]

[0095] The expression within the brackets in equation (1) above corresponds to a local model of the cross-sectional area of the artificial tissue at the i-th slice, where a is a predetermined tissue elongation parameter. In general, the tissue elongation parameter a is related to the eccentricity of the cross-sectional shape of the artificial tissue. In one embodiment, a > 1, more specifically 1 < a < 5, and again more specifically a = 2.5. The term t i is the thickness of the i-th slice. In one embodiment, where l is the length of the tissue region 502 (i.e., the distance between the first boundary point 508 and the second boundary point 510). In one embodiment, the pixel-based volume is converted to pm 3 or mm 3 .

[0096] As described in more detail below with respect to Figure 13 estimating the volume of the artificial tissue from the 2-dimensional image of the artificial tissue provides an effective and discriminative encoding of the phenotype of the artificial tissue. Moreover, the model described with respect to equation (1) above allows for the volume of the artificial tissue to be estimated efficiently without the need to obtain an expensive and computationally intensive volume measurement (e.g., a computed tomography (CT) scan, a magnetic resonance imaging (MRI) scan, etc.). Indeed, any existing 2-dimensional tissue imaging system can be quickly and efficiently extended to provide volume estimates using the model described above.

[0097] The above geometric features are used to generate one or more tissue descriptors that encode the phenotype of the artificial tissue. The one or more tissue descriptors include an area descriptor, a volume descriptor, and a smoothness descriptor. The area descriptor and the volume descriptor correspond to the area of the tissue region 502 and the volume of the artificial tissue, respectively, as described in detail above. The smoothness descriptor encodes the smoothness of the boundary 504 based on the plurality of widths. That is, for the plurality of widths determined using the method described above , the smoothness descriptor S includes the variance of the plurality of widths determined as:

[0098]

[0099] where is the mean tissue width. Alternatively, the smoothness descriptor can be determined using any other suitable measure of variability or dispersion of the plurality of widths.

[0100] As will be described in greater detail below, the one or more tissue descriptors provide a comparable geometric encoding of the tissue phenotype that can be used to determine phenotypic changes caused by drug treatments, disease models, different cell lines, or other perturbations. The one or more tissue descriptors and / or the tissue parameters from which the one or more tissue descriptors are determined also provide a mechanism for tracking the maturation of artificial tissue for quality control (QC) purposes.

[0101] Non-geometric features that can be extracted from the tissue region identified by the segmentation unit include a passive tension measure and a tissue irregularity grading.

[0102] The passive tension measure or passive tension parameter is an estimated tension exerted by the artificial tissue on the tissue scaffold when the artificial tissue is in a resting state (i.e., when the artificial tissue is not stimulated by an external source of stimulation). The passive tension measure is estimated for each of the tissue scaffolds on which the artificial tissue is attached. Alternatively, a single passive tension measure is determined for one of the tissue scaffolds. As Figure 5 As illustrated, the passive tension is measured from the first tissue scaffold region 514 and / or the second tissue scaffold region 516 identified by the segmentation unit 104 as described above.

[0103] For the first tissue scaffold region 514, the passive tension is calculated by determining the absolute difference between the first intersection point 522 and the second intersection point 524. This absolute difference corresponds to the (vertical) distance between the anchoring point of the first tissue scaffold 520 and the most extreme point of curvature of the first tissue scaffold 520. The passive tension is reported in pixels and can be converted to microns based on the resolution of the image from which the tissue region 502 was obtained. Similarly, for the second tissue scaffold region 516, the passive tension is calculated by determining the absolute difference between the third intersection point 530 and the fourth intersection point 532.

[0104] As mentioned above, the tissue irregularity grading (or simply irregularity grading / score) is another non-geometric feature that can be extracted from the tissue region. The tissue irregularity grading is primarily used for quality control (QC) purposes and quantifies irregularities or issues associated with the artificial tissue. Irregularities include: holes or discontinuities within the artificial tissue; irregularities in the shape of the tissue; shreds within the tissue; broken tissue; broken tissue scaffolds; and irregularities associated with the attachment of the artificial tissue to the device in which the artificial tissue is contained (e.g., attachment to the sides of the well, or non-attachment to the tissue scaffold). Beneficially, the tissue irregularity grading can be used to complement experimental procedures and predictions (e.g., by providing an explanation of the variability in experimental results) by providing a measure of the quality of the artificial tissue. Figure 1(As shown in the measurement unit 110). For example, if the phenotypic or functional response of the artificial tissue does not correspond to the expected value, the tissue irregularity grading of the artificial tissue can indicate that there is a problem with the artificial tissue (e.g., abnormal attachment, pores in the artificial tissue, etc.), which can at least partially explain the unexpected or abnormal response.

[0105] One or more morphological operations are used to identify pores or discontinuities within an artificial tissue. In one embodiment, the number of pores within the artificial tissue is determined by calculating the Euler number or Euler characteristic of the tissue region 502. The Euler number corresponds to the number of objects (i.e., connected components) minus the number of pores. For a tissue region with no pores, the Euler number will be 1, because there is only a single object (tissue region) and no pores. For a tissue region with a single pore, the Euler number will be 0, because there is a single object (tissue region) and a single pore. Similarly, for a tissue region with two pores, the Euler number will be -1 (minus one), because there is a single object (tissue region) and two pores. Thus, the irregularity score of pores or discontinuities within the artificial tissue can be determined based on the Euler number of the tissue region 502. Pores within the artificial tissue can be reported as binary values ​​(e.g., "true" if the Euler number is less than 1; "false" if the Euler number is equal to 1) or numerical values ​​(e.g., 1 minus the Euler number provides a numerical value for the number of pores within the tissue region).

[0106] One or more machine learning models are used to identify irregularities in tissue shape, fragmentation within tissue, fractures within tissue, and fractures in tissue scaffolds. More specifically, machine learning models are trained to analyze images of tissue (e.g., images of well regions such as...) Figure 2Each of these irregularities is identified in an image of the tissue within the well region 202) shown. Tissue shape irregularities are classified as regular (normal) or irregular. Debris within the tissue is classified as present (i.e., debris is present within the tissue) or not present (i.e., debris is not present within the tissue). Disruptions in the tissue or scaffold are identified as present (i.e., at least one disruption is present in the tissue or scaffold) or not present (i.e., no disruptions are present in the tissue or scaffold). Thus, each of the irregularities can be identified by a binary classifier trained to predict the presence or absence of the irregularity. In one embodiment, the binary classifier trained for each irregularity is a VGG-16 convolutional neural network (CNN). Each binary classifier or irregularity classifier is trained on a training dataset of approximately 2,000 images within each of the two binary classes for approximately half of the images. For each classifier, training is performed using small batch gradient descent with momentum, with a batch size of 128 and a momentum of 0.9. Each of the irregularities mentioned above is reported in the form of a binary value -“true” if the irregularity is present and“false” if the irregularity is not present.

[0107] In addition to the irregularities described above, the irregularity grading can also include one or more tissue attachment grades that indicate the degree of attachment of the artificial tissue to the device and / or its elements, such as the tissue scaffold containing the artificial tissue within. The tissue attachment grade provides an important quantification of potential irregular formation of the artificial tissue, as the tissue attachment grade identifies irregular attachment points of the artificial tissue to the device within which it is contained. In an exemplary setup, after maturation, the artificial tissue is attached to the tissue scaffold of the device, but not to any other elements of the device (e.g., the sidewall of the well). Attachment of the artificial tissue to other elements of the device can inhibit the functional response of the artificial tissue, leading to irregular or inaccurate phenotypic readouts or functional response values. Thus, the tissue attachment grade helps identify and quantify issues related to the state of the artificial tissue (e.g., during maturation), which helps improve the quality of the artificial tissue produced by identifying sources of systematic manufacturing variability. In an example QC process, artificial tissues identified as having tissue attachment that meets certain criteria (e.g., significant attachment to the side of the well distal from the tissue scaffold) can be removed, allowing new artificial tissues to grow in place.

[0108] Generally, tissue attachment grading of an artificial tissue image includes three elements: (i) an indication of whether the artificial tissue is attached to the attachment site of the device; (ii) a measure or indication of the degree of attachment of the artificial tissue to the attachment site (or an attachment severity score); and (iii) a measure of the translucency of the artificial tissue at or near the attachment site. Potential attachment sites for calculating tissue attachment grading include the tissue scaffold, the side (or sidewall) of the well adjacent to the tissue scaffold, and the side (or sidewall) of the well away from the tissue scaffold. The side of the well adjacent to the tissue scaffold corresponds to a portion of the side of the well that is close to or near the tissue scaffold (e.g., a portion of the well's sidewall over which the tissue scaffold passes, or which the tissue scaffold overlaps). The side of the well away from the tissue scaffold corresponds to a portion of the side of the well that is far from or away from the tissue scaffold. For each "corner" of the artificial tissue, one or more potential attachment sites are selected, and a tissue attachment grade is calculated, resulting in four attachment grade labels for each attachment site for each artificial tissue (e.g., upper left, upper right, lower left, lower right scores for attachment to a tissue scaffold, upper left, upper right, lower left, lower right scores for attachment to the side of the well closest to the tissue scaffold, etc.). This is in Figure 6 Example in.

[0109] Figure 6 An exploded image of a well-cropped artificial tissue for tissue attachment grading according to an embodiment of the present disclosure is shown.

[0110] Figure 6 Image 602 of an artificial tissue is shown, corresponding to a well region (as described above, consisting of, for example,...) Figure 1 The image 602 is a cropped image of artificial tissue (identified by the segmentation unit 104). The image 602 is split into four quadrant images: upper left quadrant 604, upper right quadrant 606, lower left quadrant 608, and lower right quadrant 610. Figure 6 Further illustration shows an indication of attachment sites including a first attachment site 612 at the tissue scaffold, a second attachment site 614 on the side of the well closer to the tissue scaffold, and a third attachment site 616 on the side of the well farther from the tissue scaffold.

[0111] All four quadrant images have the same width and height, corresponding to half the width and height of image 602. Each quadrant is transformed such that the artificial tissue within each quadrant has the same orientation. That is, the upper right quadrant is extracted from image 602 and then horizontally mirrored to generate... Figure 6 The upper right quadrant 606 is shown; the lower left quadrant is extracted from image 602 and vertically mirrored to generate... Figure 6 The lower left quadrant 608 is shown; and the lower right quadrant is extracted from image 602 and mirrored horizontally and vertically to generate... Figure 6the lower right quadrant 610.

[0112] Rather than determining a single attachment grade for the artificial tissue within the image 602 (for the respective attachment site), four attachment grades are determined from the four quadrant images for the respective attachment site (e.g., the first attachment site 612, the second attachment site 614, etc.). Beneficially, this helps provide more detailed details of the attachment of the tissue to the device, while also enabling the model for predicting the tissue attachment grade to be trained more effectively, as discussed in more detail below.

[0113] The tissue attachment grades obtained from the tissue attachment area images (e.g., the quadrant images, such as any of the images illustrated Figure 6 may be hierarchically represented, as illustrated Figure 7 in the following description with respect to Figure 7 and FIG. 9, it will be understood by those skilled in the art that a general reference to a tissue attachment grade encompasses any of the tissue attachment grades described above (i.e., any of the tissue attachment sites).

[0114] Figure 7 A hierarchical tissue attachment grade 700 is shown in accordance with an embodiment of the present disclosure.

[0115] The hierarchical tissue attachment grade 700 is associated with an input image 702 of a region of artificial tissue (e.g., a quadrant image, such as any of the images illustrated Figure 6 The hierarchical tissue attachment grade 700 includes three levels: an attachment output level 704, a severity output level 706, and a translucency output level 708. The attachment output level 704 includes an attachment not present label 710 or an attachment present label 712. The severity output level 706 includes a slight attachment label 714, a moderate attachment label 716, or a severe attachment label 718. In some embodiments, the severity output level 706 includes an attachment not present label 720. The translucency output level 708 includes a non-translucent label 722 or a translucent label 724. In some embodiments, the translucency output level 708 includes an attachment not present label 726.

[0116] The labels assigned at the attachment output level 704 can be represented by the labels {no attachment, attached} (e.g., no attachment present label 710 or attachment present label 712) or via a 2-valued indicator vector, where a first value is used to identify no attachment (e.g., vector [1, 0]) and a second value is used to identify attachment (e.g., vector [0, 1]). The labels assigned at the severity output level 706 can be represented by the labels {none, mild, moderate, severe} (e.g., no attachment present label 720, mild attachment label 714, moderate attachment label 716, or severe attachment label 718) or via a 4-valued indicator vector, where a first value is used to identify no attachment (e.g., vector [1, 0, 0, 0]), a second value is used to identify mild attachment (e.g., vector [0, 1, 0, 0]), a third value is used to identify moderate attachment (e.g., vector [0, 0, 1, 0]), and a fourth value is used to identify severe attachment (e.g., vector [0, 0, 0, 1]). The labels assigned at the translucency output level 708 can be represented by the labels {no attachment, non-translucent, translucent} (e.g., no attachment present label 726, non-translucent label 722, or translucent label 724) or via a 3-valued indicator vector, where a first value is used to identify no attachment (e.g., vector [0, 1, 0]), a second value is used to identify non-translucent (e.g., vector [0, 1, 0]), and a third value is used to identify translucent (e.g., vector [0, 0, 1]).

[0117] The hierarchical tissue attachment classification 700 can be represented as a combination of labels (e.g., {attached, mild, non-translucent} for non-translucent tissue with mild attachment at the respective tissue attachment site), as a numerical value representing the seven different possible outcomes, or as an indicator vector representing each of the different possible outcomes (e.g., vector [1, 0, 0, 0, 0, 0, 0] represents the no attachment outcome shown by the "=0" path in Figure 7 [0, 0, 0, 1, 0, 0, 0] represents the moderate attachment non-translucent outcome shown by the "=3" path in Figure 7 [0, 0, 0, 0, 0, 0, 1] represents the severe attachment translucent outcome shown by the "=6" path in Figure 7 [0, 0, 0, 0, 0, 0, 1] represents the severe attachment translucent outcome shown by the "=6" path in

[0118] The attachment classification is determined using a classification model. Generally, the classification model is configured or trained to receive an input image (e.g., input image 702) or a set of features extracted from the input image, and estimate an attachment classification based on the input image. In one embodiment, the classification model includes a hierarchy of classifiers such that a classifier is associated with each output level in the hierarchical organization of attachment classification 700. That is, an attachment classifier determines a label for attachment output level 704 based on input image 702, a severity classifier determines a label for severity output level 706 based on input image 702, and a translucency classifier determines a label for translucency output level 708 based on input image 702. The attachment classification is then determined by combining the outputs of each of the classifiers within the hierarchy of classifiers.

[0119] As stated above, one or more feature extraction processes can be applied to input image 702 such that features obtained from the one or more feature extraction processes are used as input to the classification model (e.g., each of the classifiers within the hierarchy of classifiers) during both the training phase and the inference phase. Example feature extraction processes include principal component analysis (PCA), bag-of-words, histogram of oriented gradients (HOG) features, speeded up robust features (SURF), and local binary pattern (LBP) features.

[0120] The attachment classifier assigns either the attachment not present label 710 or the attachment present label 712 to input image 702 (or features extracted therefrom). Thus, the attachment classifier is a binary classifier trained to predict the presence or absence of an attachment within an input image. In one embodiment, the attachment classifier determines a probability or score associated with the presence of an attachment within an input image. The probability or score is then converted to a label based on a threshold (e.g., if the probability is greater than 0.5, then the attachment present label 712 is assigned, otherwise the attachment not present label 710 is assigned). The attachment classifier is any suitable binary machine learning classifier, such as a Naive Bayes classifier, a support vector machine (SVM), a multilayer perceptron, a decision tree, a k-nearest neighbor (kNN) classifier, etc.

[0121] The severity classifier assigns a slight attachment label 714, a moderate attachment label 716, or a severe attachment label 718 to the input image 702 (or features extracted therefrom). In one embodiment, the severity classifier can also assign an attachment not present label 720 to the input image 702 (i.e., instead of the slight attachment label 714, the moderate attachment label 716, or the severe attachment label 718). Thus, the severity classifier is a multi-class classifier trained to predict the severity or degree of attachment of artificial tissue to a tissue scaffold (as represented within the input image). In one embodiment, the severity classifier determines a probability or score for each of the possible outputs (slight, moderate, severe (and optionally, no attachment)), and assigns the label based on the most likely output (i.e., the output with the highest probability or score). The severity classifier is any suitable multi-class classifier, such as a Naive Bayes classifier, a multilayer perceptron, a decision tree, a kNN classifier, etc.

[0122] The translucency classifier assigns a non-translucent label 722 or a translucent label 724 to the input image 702 (or features extracted therefrom). In one embodiment, the translucency classifier can also assign an attachment not present label 726 to the input image 702 (i.e., instead of the non-translucent label 722 or the translucent label 724). Thus, the translucency classifier is a binary (or multi-class) classifier trained to predict the translucency of artificial tissue at or near a tissue scaffold (as represented within the input image). In one embodiment, the translucency classifier determines a probability or score for each of the possible outputs (translucent, non-translucent (and optionally, no attachment)), and assigns the label based on the most likely output. The translucency classifier is any suitable binary (or multi-class) classifier, such as a Naive Bayes classifier, a support vector machine (SVM), a multilayer perceptron, a decision tree, a k-nearest neighbor (kNN) classifier, etc.

[0123] Each classifier in the hierarchy of classifiers is trained on a training dataset of labeled images having labels related to a particular attachment site. For example, a first attachment classifier is trained on a training dataset of images having labels related to attachment of artificial tissue shown within each image to a tissue scaffold shown within each image (e.g., each label includes an attachment indicator related to whether artificial tissue is attached to a tissue scaffold). A first severity classifier is trained on a training dataset of images having labels related to a degree of attachment of artificial tissue shown within each image to a tissue scaffold. A first translucency classifier is trained on a training dataset of images having labels related to translucency of tissue shown within each image. Thus, the training dataset includes a plurality of images and a plurality of labels, each label associated with a corresponding image of the plurality of images.

[0124] In one implementation, the hierarchy of classifiers is trained for each attachment site. Each classifier within the hierarchy of classifiers includes a support vector machine (SVM) having a kernel based on Kullback-Leibler (K-L) divergence. During training and inference, a feature extraction pipeline is used to convert a quadrant image to a feature vector. Uniform local binary patterns are extracted from the quadrant image having a cell size of 16x16, a radius of 1, and 8 nearest neighbor set points. The set of local binary patterns for each image is then converted to a feature vector by generating a histogram from the set of local binary patterns (where the number of bins is set to the number of unique local binary patterns + 1). Thus, each image is represented by a feature vector corresponding to a histogram of the number of occurrences of each local binary pattern within the image. Because these feature vectors can be viewed as a distribution of local binary patterns in the image, a kernel based on K-L divergence is used for the SVM. Hyperparameters for the SVM are determined using a method based on a standard grid search. Each classifier within the hierarchy of classifiers is trained on a training dataset including 2,000 images obtained from images of 500 artificial tissues, such that each image within the training dataset corresponds to a quadrant image. Each image within the training dataset is associated with a label vector within a corresponding plurality of labels. The label vector includes an attachment score, a severity score, and a translucency score for each of three attachment sites (at the tissue scaffold, at a portion of the well proximate to the tissue scaffold, and at a portion of the well distal to the tissue scaffold), such that each label within the plurality of labels includes a 9-dimensional vector. The classifiers are then trained to predict a given label within the label vector. For example, an attachment classifier for the tissue scaffold attachment site is trained to predict a label within the label vector associated with an attachment score for the tissue scaffold attachment site.

[0125] The output of each classifier in the hierarchy of combined classifiers is combined to produce an attachment ranking. As stated above, the outputs can be combined to generate a label vector (e.g., {attached, slight, non-transparent}), a numerical representation (e.g., “1”), or an indicator vector (e.g., [0, 1, 0, 0, 0, 0, 0]).

[0126] Figure 8 A cropped region of three artificial tissues with different predicted tissue attachment rankings is shown in accordance with embodiments of the present disclosure.

[0127] Figure 8 A first attachment image 802, a second attachment image 804, and a third attachment image 806 are shown. Each of the attachment images is obtained from a corresponding image of an artificial tissue cropped from a region of the image including a side of a well proximate to a tissue scaffold. Those skilled in the art will appreciate that the images are cropped in this manner to aid in understanding and provide a clearer view of the tissue attachment sites. In Figure 8 In the example shown, the attachment site corresponds to a side of the well proximate to a tissue scaffold (e.g., Figure 6 The second attachment site 614 shown).

[0128] The first attachment image 802 shows a predicted slight attachment of the artificial tissue to a side of the well, where the artificial tissue is ranked as non-transparent at or near the attachment site. Thus, the tissue attachment ranking for the first attachment image 802 can be represented as {attached, slight, non-transparent}, a numerical “2”, or an indicator vector [0, 0, 1, 0, 0, 0, 0].

[0129] The second attachment image 804 shows a predicted moderate attachment of the artificial tissue to a side of the well, where the artificial tissue is ranked as non-transparent at or near the attachment site. Thus, the tissue attachment ranking for the second attachment image 804 can be represented as {attached, moderate, non-transparent}, a numerical “3”, or an indicator vector [0, 0, 0, 1, 0, 0, 0].

[0130] The third attachment image 806 shows a predicted severe or strong attachment of the artificial tissue to a side of the well, where the artificial tissue is ranked as non-transparent at or near the attachment site. Thus, the tissue attachment ranking for the third attachment image 806 can be represented as {attached, severe, non-transparent}, a numerical “6”, or an indicator vector [0, 0, 0, 0, 0, 0, 1].

[0131] The tissue attachment rankings for the attachment sites and sub-regions of the artificial tissue are automatically computed (e.g., quadrant images, such as Figure 6The images shown provide accurate and informative location-specific measurements of the potential irregularities in the physiological state of artificial tissues. This automation helps improve the reliability and robustness of reported measurements while reducing the amount of manual input required, thereby reducing user-induced variability. Therefore, tissue attachment grading can be used to guide quality control procedures (e.g., the inclusion or exclusion of artificial tissues) and / or to provide supplementary analyses to help interpret experimental variability.

[0132] As an alternative to the hierarchical structure of classifiers described above, using, for example Figure 9A The single trained multi-task learning (MTL) network shown determines the attachment grading of the corresponding attachment site.

[0133] Figure 9A A multi-task learning (MTL) network 900 for organization attachment hierarchy is shown according to one aspect of this disclosure.

[0134] MTL network 900 includes an input network 902, a shared dense network 904, and multiple output networks (including an attachment network 906-1, a severity network 906-2, a translucency network 906-3, and an auxiliary network 906-4). MTL network 900 receives an input image 908 and generates multiple outputs (including an attachment index 910-1, a severity index 910-2, a translucency index 910-3, and a rating 910-4). In one embodiment, the multiple output networks further include a minor severity network 906-5 that generates a minor severity index 910-5. The following section discusses... Figure 9B and Figure 9C The architecture of the input network 902 is described in more detail. Figure 9A The architecture of the shared dense network 904 shown includes a 2D global average pooling layer 912, a first dense block 914-1, and a second dense block 914-2. The first dense block 914-1 and the second dense block 914-2 share the same dense block architecture, which is related to... Figure 9A The first dense block 914-1 is illustrated here. The dense block architecture (shown with respect to the first dense block 914-1) includes a dense layer 916, an exponential linear unit (ELU) activation layer 918, and a dropout layer 920. Those skilled in the art will understand that... Figure 9A The dense blocks shown all share the same architecture, but may have different hyperparameters (e.g., different drop rates or different numbers of nodes within the dense layer). Multiple output networks all share a common architecture, and this architecture of the output networks is... Figure 9AThe output network architecture is shown with respect to the attachment network 906-1. The output network architecture includes a third dense block 914-3, a fourth dense block 914-4, a dense layer 922, and a Softmax layer 924. As previously stated, the architecture of the third dense block 914-3 and the fourth dense block 914-4 is the same as the dense block architecture described above with respect to the first dense block 914-1. It will be appreciated by those skilled in the art that while the architecture of each output network is the same, the hyperparameters of each output network can differ.

[0135] The MTL network 900 is trained to predict multiple outputs from a single input, enabling the MTL network 900 to take advantage of the supply of relational information to improve staging accuracy by sharing updates of the dense network 904. The input image 908 provided to the MTL network 900 includes a portion of artificial tissue (e.g., a quadrant as described above with respect to Figure 6 In one example implementation, the input image 908 is a grayscale image having dimensions 775px x 175px. As previously stated, the multiple outputs include the attachment indicator 910-1, the severity indicator 910-2, the translucency indicator 910-3, the staging 910-4, and in some embodiments, the mild severity indicator 910-5.

[0136] The attachment indicator 910-1 (alternatively referred to as an attachment indicator vector, an attachment output vector, an attachment score, or an attachment score vector) includes a prediction of whether a portion of artificial tissue within the input image 908 is present with an attachment site. The attachment indicator 910-1 can be a 2-dimensional vector, where the first value of the vector includes a probability of the presence of an attachment, and the second value of the vector includes a probability of the absence of an attachment. Because the attachment indicator 910-1 is generated by the softmax layer of the attachment network 906-1, the 2-dimensional vector is normalized such that the sum of the probabilities is one. The attachment indicator 910-1 can be converted to an attachment label by assigning a label corresponding to the maximum probability value within the attachment indicator 910-1 (e.g., if the first value of the vector is the maximum value, then the label {no attachment} is assigned).

[0137] The severity indicator 910-2 (alternatively referred to as a severity indicator vector, a severity output vector, a severity score, or a severity score vector) includes a prediction of the severity or degree of attachment (as captured within the input image 908) of a portion of the artificial tissue to the attachment site. The severity indicator 910-2 can be a 4-dimensional vector, where the first value of the vector includes a probability of no attachment, the second value of the vector includes a probability of light attachment, the third value of the vector includes a probability of moderate attachment, and the fourth value of the vector includes a probability of significant attachment. Because the severity indicator 910-2 is generated by a softmax layer of the severity network 906-2, the 4-dimensional vector is normalized such that the sum of the probabilities is one. The severity indicator 910-2 can be converted to a severity label by assigning a label corresponding to the maximum probability value within the severity indicator 910-2 (e.g., if the fourth value of the vector is the maximum, then the label {significant} is assigned).

[0138] The translucency indicator 910-3 (alternatively referred to as a translucency indicator vector, a translucency output vector, a translucency score, or a translucency score vector) includes a prediction of the translucency (as captured within the input image 908) of a portion of the artificial tissue at or near the attachment site. The translucency indicator 910-3 can be a 3-dimensional vector, where the first value of the vector includes a probability of no attachment, the second value of the vector includes a probability of the portion of the artificial tissue being translucent, and the third value of the vector includes a probability of the portion of the artificial tissue not being translucent. Because the translucency indicator 910-3 is generated by a softmax layer of the translucency network 906-3, the 3-dimensional vector is normalized such that the sum of the probabilities is one. The translucency indicator 910-3 can be converted to a translucency label by assigning a label corresponding to the maximum probability value within the translucency indicator 910-4 (e.g., if the second value of the vector is the maximum, then the label {translucent} is assigned).

[0139] Classification 910-4 (alternatively referred to as a classification vector, a classification output vector, or an auxiliary output vector) includes a prediction of the overall attachment classification (as captured within input image 908) of a portion of artificial tissue with an attachment site. Classification 910-4 can be a 7-dimensional vector, where the first value of the vector includes a probability of no attachment, the second value of the vector includes a probability of the portion of artificial tissue being translucent and lightly attached, the third value of the vector includes a probability of the portion of artificial tissue being non-translucent and lightly attached, the fourth value of the vector includes a probability of the portion of artificial tissue being non-translucent and lightly attached, the fifth value of the vector includes a probability of the portion of artificial tissue being non-translucent and moderately attached, the sixth value of the vector includes a probability of the portion of artificial tissue being non-translucent and heavily attached, and the seventh value of the vector includes a probability of the portion of artificial tissue being non-translucent and heavily attached. Because classification 910-4 is generated by a softmax layer of auxiliary network 906-4, the 7-dimensional vector is normalized such that the sum of the probabilities is one. Classification 910-4 can be converted into a classification label by assigning a label corresponding to the maximum probability value within classification 910-4 (e.g., if the fifth value of the vector is the maximum value, then assign the label {attached, moderate, non-translucent}). Advantageously, including auxiliary network 906-4 (and classification 910-4) in MTL network 900 helps improve the training of MTL network 900, particularly shared dense network 904. Once MTL network 900 is trained, classification 910-4 can not be used, and instead, hierarchical classifications of the input image determined from each of the other outputs described above are used.

[0140] Light severity indicator 910-5 (alternatively referred to as a light severity indicator vector, a light severity output vector, a light severity score, or a light severity score vector) includes a prediction of whether a portion of artificial tissue has light attachment or normal (i.e., greater than light) attachment (as captured within input image 908). Light severity indicator 910-5 can be a 3-dimensional vector, where the first value of the vector includes a probability of no attachment, the second value of the vector includes a probability of the portion of artificial tissue having light attachment, and the third value of the vector includes a probability of the portion of artificial tissue having normal attachment. Because light severity indicator 910-5 is generated by a softmax layer of light severity network 906-5, the 3-dimensional vector is normalized such that the sum of the probabilities is one. Light severity indicator 910-5 can be converted into a light severity label by assigning a label corresponding to the maximum probability value within light severity indicator 910-5 (e.g., if the first value of the vector is the maximum value, then assign the label {no attachment}).

[0141] Input network 902 includes a convolutional neural network configured to receive input image 908 and output a feature vector to shared dense network 904.

[0142] Figure 9B An embodiment according to this disclosure is shown. Figure 9A The architecture of the input network 902 of the MTL network 900 is shown.

[0143] The input network 902 includes an ingress stream network 926, multiple intermediate stream networks (including a first intermediate stream network 928-1, a second intermediate stream network 928-2, a third intermediate stream network 928-3, and a fourth intermediate stream network 928-4), and an egress stream network 930. The ingress stream network 926 receives the input image (e.g., ...). Figure 9A The input image 908 shown), and from the output stream network 930 (e.g., to... Figure 9A The shared dense network 904 shown outputs a feature vector. The ingress flow network 926 includes a first convolutional block 932-1, a first 2D max-pooling layer 934, a second convolutional block 932-2, a third convolutional block 932-3, a second 2D max-pooling layer 936, a fourth convolutional block 932-4, a fifth convolutional block 932-5, and a third 2D max-pooling layer 938. The architecture of all convolutional blocks (e.g., the first convolutional block 932-1, the second convolutional block 932-2, etc.) is the same, but those skilled in the art will understand that the hyperparameters may differ between the blocks. Figure 9B The architecture of the first convolutional block 932-1 is shown below. This first convolutional block includes a 2D separable convolutional layer 940, a batch normalization layer 942, and an exponential linear unit (ELU) activation layer 944. The following section discusses... Figure 9C The architecture of the multiple intermediate flow networks is described in more detail. The exit flow network 930 includes a sixth convolutional block 932-6, a seventh convolutional block 932-7, a 2D separable convolutional layer 946, and an ELU activation layer 948.

[0144] The input image is provided to a first convolutional block 932-1 of the entry stream network 926. The 2-dimensional separable convolutional layer 940 of the first convolutional block 932-1 has 128 filters (with filter size of (7 x 7)), the batch normalization layer 942 is a normalization layer that transforms the output of the 2-dimensional separable convolutional layer 940 so that the mean is approximately 0 and the standard deviation is approximately 1, and the ELU activation layer 944 applies an exponential linear activation function to the output of the batch normalization layer 942 (with scale set to a = 0.01). The output of the first convolutional block 932-1 (i.e., the output of the ELU activation layer 944) is provided to a first 2-dimensional max pooling layer 934 that applies a max pooling operation (i.e., down-sampling) with a pool size of (2, 2). The output of the first 2-dimensional max pooling layer 934 is provided to a second convolutional block 932-2, the output of which is provided to a third convolutional block 932-3. As described above, the second and third convolutional blocks 932-2, 932-3 share the same architecture as the first convolutional block 932-1. The hyperparameters of these blocks are also the same, but the filter size of the 2-dimensional separable convolutional layer of the second convolutional block 932-2 is (1 x 1), and the filter size of the 2-dimensional separable convolutional layer of the third convolutional block 932-3 is (3 x 3). The output of the third convolutional block 932-3 is provided to a second 2-dimensional max pooling layer 936 that applies a max pooling with a pool size of (2, 2). The output of the second 2-dimensional max pooling layer 936 is provided to a fourth convolutional block 932-4, the output of which is provided to a fifth convolutional block 932-5. The fourth and fifth convolutional blocks 932-4, 932-5 share the same architecture as the first convolutional block 932-1. The hyperparameters of these blocks are also the same, but the filter size of the 2-dimensional separable convolutional layer of the fourth convolutional block 932-4 is (1 x 1), and the filter size of the 2-dimensional separable convolutional layer of the fifth convolutional block 932-5 is (3 x 3). The output of the third convolutional block 932-3 is provided to a third 2-dimensional max pooling layer 938 that applies a max pooling with a pool size of (2, 2).

[0145] The output of the entry stream network 926 (i.e., the output of the third 2-dimensional max pooling layer 938) is provided to a first intermediate stream network 928-1 of a plurality of intermediate stream networks.

[0146] Figure 9C An architecture of each of the plurality of intermediate stream networks of the input network 902 is shown in accordance with one embodiment of the present disclosure. Figure 9B An architecture of each of the plurality of intermediate stream networks of the input network 902 is shown in accordance with one embodiment of the present disclosure.

[0147] Each intermediate flow network includes a residual block 950, a max pooling block 952, a plurality of starter blocks (including a first starter block 954-1, a second starter block 954-2, a third starter block 954-3, and a fourth starter block 954-4), and a dilation block 956. The input to the intermediate flow network is provided to each of the above-mentioned blocks, and the block output is provided to a concatenation block 958. The output of the intermediate flow network corresponds to the output of the concatenation block 958.

[0148] The residual block 950 includes a 2-dimensional separable convolution layer with 128 filters of size (1 x 1). The max pooling block 952 includes a 2-dimensional max pooling layer 960 and a 2-dimensional separable convolution layer 962. The 2-dimensional max pooling layer 960 applies max pooling with a pool size of (2 x 2), and the output of the 2-dimensional max pooling layer 960 is provided to the 2-dimensional separable convolution layer 962 including 128 filters of size (1 x 1).

[0149] The plurality of starter blocks (e.g., the first starter block 954-1, the second starter block 954-2, etc.) work in parallel to extract multi-level features from the input. Each of the starter blocks shares the same architecture, but some layer hyperparameters are different. In Figure 9C In particular, the common architecture is shown with respect to the first starter block 954-1. The first starter block 954-1 includes a first convolution block 964-1, a second convolution block 964-2, a third convolution block 964-3, and a final 2-dimensional separable convolution layer 966. The first convolution block 964-1 includes a 2-dimensional separable convolution layer 968, a batch normalization layer 970, and an ELU activation layer 972 (with a scale set to a = 0.01). The second convolution block 964-2 and the third convolution block 964-3 share the same architecture as the first convolution block 964-1, but the 2-dimensional separable convolution layer hyperparameters are different.

[0150] For the first starter block 954-1, the 2-dimensional separable convolution layer 968 of the first convolution block 964-1 has 128 filters of size (1 x 1), the 2-dimensional separable convolution layer of the second starter block 964-2 has 128 filters of size (3 x 3), the 2-dimensional separable convolution layer of the third starter block 964-3 has 128 filters of size (1 x 1), and the final 2-dimensional separable convolution layer 966 has 128 filters of size (3 x 3). Beneficially, the use of (1 x 1) convolution layers can enable more efficient processing by reducing the dimensionality of the input data.

[0151] For the second starting block 954-2, the 2-dimensional separable convolution layer of the first convolution block has 128 filters of size (1 x 1), the 2-dimensional separable convolution layer of the second starting block has 128 filters of size (5 x 5), the 2-dimensional separable convolution layer of the third starting block has 128 filters of size (1 x 1), and the final 2-dimensional separable convolution layer has 128 filters of size (5 x 5).

[0152] For the third starting block 954-3, the 2-dimensional separable convolution layer of the first convolution block has 128 filters of size (1 x 1), the 2-dimensional separable convolution layer of the second starting block has 128 filters of size (7 x 7), the 2-dimensional separable convolution layer of the third starting block has 128 filters of size (1 x 1), and the final 2-dimensional separable convolution layer has 128 filters of size (7 x 7).

[0153] For the fourth starting block 954-4, the 2-dimensional separable convolution layer of the first convolution block has 128 filters of size (1 x 1), the 2-dimensional separable convolution layer of the second starting block has 128 filters of size (9 x 9), the 2-dimensional separable convolution layer of the third starting block has 128 filters of size (1 x 1), and the final 2-dimensional separable convolution layer has 128 filters of size (9 x 9).

[0154] The concatenation block 958 includes a plurality of convolution layers 974-1 through 974-7, each convolution layer including a 2-dimensional separable convolution layer of 128 filters of size (2 x 2) and a dilated convolution having a dilation rate of 1. Beneficially, the use of sequentially dilated convolutions within the concatenation block 958 enables the intermediate flow network to have a larger receptive field (i.e., by "skipping" over pixels, enabling the convolution to cover a larger area of the input) without increasing the number of parameters, providing an effective increase in the receptive field without increasing the complexity of the parameter space to be searched during training.

[0155] The concatenation block 958 concatenates the outputs of the residual block 950, the max-pooling block 952, the first starting block 954-1, the second starting block 954-2, the third starting block 954-3, the fourth starting block 954-4, and the dilation block 956. The concatenation block 958 includes a concatenation layer 976 (which concatenates all of the inputs (i.e., the outputs described above)), a batch normalization layer 978, and an ELU activation layer 980 (which is scaled to a = 0.01).

[0156] Referring again to Figure 9B , each of the plurality of intermediate flow networks (e.g., the first intermediate flow network 928-1, the second intermediate flow network 928-2, etc.) has a number of layers that is less than the number of layers of the final flow network 926. Figure 9CThe same architecture is described. The output of the final intermediate flow network—the fourth intermediate flow network 928-4—is provided to the exit flow network 930.

[0157] In the egress network 930, the 2D separable convolutional layers of the sixth convolutional block 932-6 and the seventh convolutional block 932-7, as well as the 2D separable convolutional layer 946, have 256 filters of size (7×7). The ELU activation layers of the sixth convolutional block 932-6 and the seventh convolutional block 932-7, as well as the ELU activation layer 948, have a scale set to α = 0.01.

[0158] Refer again Figure 9A The output of input network 902 (i.e., Figure 9B The output of the ELU activation layer 948 of the outflow network 930 (shown) is provided to the shared dense network 904. A 2D global average pooling layer 912 of the shared dense network 904 applies a global average pooling operation to the input of the shared dense network 904. The dense layer 916 of the first dense block 914-1 of the shared dense network 904 is a densely connected layer with 256 units. Then, the ELU activation layer 918 applies an ELU activation function with a scale set to α = 0.01, and the dropout layer 920 applies a dropout rate of 0.8 to the output of the ELU activation layer 918. The second dense block 914-2 of the shared dense network 904 has the same architecture and hyperparameters as the first dense block 914-1.

[0159] The output of the shared dense network 904 is then provided to each of the plurality of output networks. The architecture and hyperparameters of each of the plurality of output networks are the same, but the number of units in the final dense layer (e.g., dense layer 922 related to attachment network 906-1) is different. For each of the plurality of output networks, the dense layers of the first dense block have 256 units, the ELU activation function of the first dense block has a scale set to a = 0.01, and the dropout rate of the first dense block is set to 0.6. For each of the plurality of output networks, the dense layers of the second dense block have 512 units, the ELU activation function of the second dense block has a scale set to a = 0.01, and the dropout rate of the second dense block is set to 0.8. For attachment network 906-1, dense layer 922 has 2 units because attachment network 906-1 predicts whether there is an attachment or whether there is no attachment. For severity network 906-2, the dense layer has 4 units because severity network 906-3 predicts no attachment, mild attachment, moderate attachment, or major attachment. For translucency network 906-3, the dense layer has 3 units because translucency network 906-3 predicts no attachment, translucent, or non-translucent. For auxiliary network 906-4, dense layer 922 has 7 units because auxiliary network 906-4 predicts whether there is no attachment, mild attachment with translucent tissue, mild attachment with non-translucent tissue, moderate attachment with translucent tissue, moderate attachment with non-translucent tissue, major attachment with translucent tissue, or major attachment with non-translucent tissue. For mild severity network 906-5, the dense layer has 3 units because mild severity network 906-5 predicts no attachment, mild attachment, or normal attachment (i.e., an attachment that is more severe or stronger than mild attachment).

[0160] In one implementation, the MTL network 900 is trained using a training dataset that includes 2,000 quadrant images, with 7 possible labels (e.g., labels 0 through 6 shown) of approximately 286 images (quadrants). The images are assigned labels by expert evaluation. The training dataset is split for training / validation in a ratio of 80 / 20. The MLT network 900 is trained using mini-batch gradient descent (batch size of 128) and ADAM. The ADAM solver has an initial learning rate of le-3 and is stopped early based on validation loss. Figure 7

[0161] The output of the trained MTL network is used to estimate input images (e.g., such as Figure 6 ​The predicted tissue attachment grade indicates the degree of attachment of a portion of the engineered or artificial tissue to an attachment site (as captured within the input image). In one embodiment, the auxiliary network 906-4 is used in a training phase to improve the robustness and accuracy of the predictions produced by the other output networks (e.g., the attachment network 906-1, the severity network 906-2, etc.). The predicted tissue attachment grade is then determined from the outputs of the other output networks. More specifically, the predicted attachment grade can be determined by combining the attachment degree (or severity / size of attachment) prediction within the severity indicator 910-2 with the translucency prediction within the translucency indicator 910-3.

[0162] Referring again to Figure 1 The features and / or tissue descriptors generated by the extraction unit 106 (as described above) are then used by a quality control (QC) unit 108 and / or an assay unit 110. The operation of each of these units is described in turn below.

[0163] In general, the QC unit 108 determines the state (e.g., physiological or morphological state) of the artificial tissue at a given point in time, and can track the development and / or state changes of the artificial tissue over a plurality of predetermined time points (e.g., 1 day post-seeding, 2 days post-seeding, 7 days post-seeding, etc.). At the predetermined time points, the QC unit 108 generates a QC report with one or more tissue parameters obtained from one or more image segments of images (taken at the predetermined time points) of the artificial tissue. The one or more tissue parameters are determined by the extraction unit 106 from the image segments obtained from the segmentation unit 104. The QC report provides an indication of the state of the artificial tissue at the predetermined time points as encoded in the one or more tissue parameters. The QC report also provides an indication of the state changes of the artificial tissue over the plurality of predetermined time points.

[0164] Additionally or alternatively, the QC report includes a tissue quality alert when at least one of the tissue parameters does not satisfy a predetermined quality control threshold. For example, the quality control threshold can require that the tissue attachment grade score of each of the attachment points (i.e., the four corners of the artificial tissue) of the tissue be above a threshold value at a particular time point (e.g., 49 days post-seeding). In one embodiment, the QC unit 108 or another unit of the system 100 transmits a command to the bioreactor 102 based on the tissue quality alert. The bioreactor 102 can be configured to handle the command (e.g., by shutting down a device containing the artificial tissue that failed to satisfy the quality control threshold, or issuing an alert emphasizing the existence of a quality issue associated with the artificial tissue that failed to satisfy the quality control threshold). Additionally or alternatively, the command is transmitted to another device (e.g., a computing device, a display device, etc.) to alert a user of the system that the quality control threshold has been satisfied or has not been satisfied.

[0165] Figures 10-12 The QC reporting process during the maturation period of an artificial tissue is illustrated.

[0166] Figure 10 An image sequence of an artificial tissue during its maturation stage is shown according to an embodiment of the present disclosure.

[0167] Figure 10 The image shows a first image 1002 of the artificial tissue taken at a first predetermined time point during the maturation period, a second image 1004 of the artificial tissue taken at a second predetermined time point during the maturation period, a third image 1006 of the artificial tissue taken at a third predetermined time point during the maturation period, a fourth image 1008 of the artificial tissue taken at a fourth predetermined time point during the maturation period, and a fifth image 1010 of the artificial tissue taken at a fifth predetermined time point during the maturation period.

[0168] exist Figure 10 In the example shown, the predetermined time points correspond to 1 day, 2 days, 7 days, 14 days, and 49 days post-vaccination, respectively. Therefore, the first image 1002 was obtained 1 day post-vaccination, the second image 1004 was obtained 2 days post-vaccination, the third image 1006 was obtained 7 days post-vaccination, the fourth image 1008 was obtained 14 days post-vaccination, and the fifth image 1010 was obtained 49 days post-vaccination.

[0169] As shown in the figure, the morphological state of the artificial tissue evolves within predetermined time points. This state change is encoded in the QC report. Therefore, at each predetermined time point, one or more tissue parameters are extracted from an image of the artificial tissue (or an image fragment extracted from an image) to determine a QC report that provides an indication of the artificial tissue's maturity state. This is in... Figure 11 As shown in the image.

[0170] Figure 11 An embodiment of the present disclosure is shown in Figure 10 This is a portion of the QC report obtained during the maturation period of the artificial tissue shown.

[0171] Figure 11 First chart 1102, second chart 1104, and third chart 1106 are shown. In one embodiment, first chart 1102, second chart 1104, and third chart 1106 are included as part of a QC report that documents the maturation status of the artificial tissue during the maturation period. Figure 11 The QC report shown was generated at a predetermined time point corresponding to 49 days post-vaccination and includes tissue parameters generated at previously predetermined time points (e.g., 1 day post-vaccination, 2 days post-vaccination, etc.). Figure 11The QC report shown provides important feedback and metrics related to the state of the artificial tissue at a predetermined time point (49 days post-seeding) and the development of the artificial tissue over the maturation period.

[0172] The first graph 1102 includes a plot of the length of the tissue obtained during the maturation of the artificial tissue. The first graph 1102 includes three points corresponding to the length of the artificial tissue obtained at 7 days post-seeding, 14 days post-seeding, and 49 days post-seeding. The length plotted in the first graph 1102 is obtained from the third image 1006, the fourth image 1008, and the fifth image 1010 shown in Figure 1 using the segmentation unit 104 (i.e., the segmentation unit 104 in Figure 1 and the feature extraction unit (i.e., the extraction unit 106 in Figure 10 as described in more detail above with respect to Figure 5 .

[0173] The second graph 1104 includes a plot of the width of the tissue obtained during the maturation of the artificial tissue. The second graph 1104 includes five plots of the width of the tissue obtained at 7 days post-seeding, 14 days post-seeding, and 49 days post-seeding (as obtained at the plurality of intermediate points 512-1, 512-2, 512-3, 512-4, and the centroid 506 shown in Figure 5 . The width plotted in the second graph 1104 is obtained from the third image 1006, the fourth image 1008, and the fifth image 1010 shown in Figure 1 using the segmentation unit 104 (i.e., the segmentation unit 104 in Figure 1 and the feature extraction unit (i.e., the extraction unit 106 in Figure 10 as described in more detail above with respect to Figure 5 . The plot labeled “W1” corresponds to the width of the tissue measured at the first intermediate point 512-1 shown in Figure 5 . The plot labeled “W2” corresponds to the width of the tissue measured at the second intermediate point 512-2 shown in Figure 5 . The plot labeled “W3” corresponds to the width of the tissue measured at the centroid 506 shown in Figure 5 . The plot labeled “W4” corresponds to the width of the tissue measured at the third intermediate point 512-3 shown in Figure 5 . The plot labeled “W5” corresponds to the width of the tissue measured at the fourth intermediate point 512-4 shown in Figure 5 .

[0174] Figure 1106 in the third chart includes plots of passive guidewire tension (or passive guidewire curvature PWC) obtained during the maturation of the artificial tissue. Figure 1106 includes two plots of passive guidewire tension obtained at 1 day, 2 days, 7 days, 14 days, and 49 days post-inoculation. The passive guidewire tension plotted in Figure 1106 is obtained using segmented units (i.e., Figure 1 The segmentation unit 104 and the feature extraction unit (i.e., Figure 1 Extraction unit 106 in the middle) from Figure 10 The images 1002, 1004, 1006, 1008, and 1010 shown above were obtained as described in the previous section. Figure 5 To describe in more detail.

[0175] Figure 12 An embodiment of the present disclosure is shown in Figure 10 The tissue adhesion grading section of the QC report obtained during the maturation period of the artificial tissue shown.

[0176] Figure 12 Chart 1202, Chart 1204, Chart 1206, and Chart 1208 are shown. In one embodiment, Chart 1202, Chart 1204, and Chart 1206 are included as part of a QC report that documents the maturation status of the artificial tissue during the maturation period. In one embodiment, Figure 12 The organizational hierarchy section of the QC report shown is Figure 11 This is a portion of the QC report shown. Figure 12 The organizational hierarchy section of the QC report shown is generated at multiple predetermined time points (as mentioned above). Figure 10 (as discussed), and provides important feedback and metrics related to the state of artificial tissues, such as those encoded by organizational gradations.

[0177] Figure 1202 shows a plot of the tissue attachment grading of the upper left portion of the artificial tissue relative to the side of the well closest to the tissue support. Seven days post-inoculation (e.g.) Figure 10 As shown in the third image 1006), there is no attachment of the upper left portion of the artificial tissue to the attachment site (i.e., the side of the pore closest to the tissue scaffold), and the artificial tissue is translucent, as indicated by the solid black circle at the 7-day post-inoculation time point. 14 days post-inoculation (e.g.) Figure 10 As shown in the fourth image 1008), the upper left portion of the artificial tissue is not translucent, as indicated by the white circle at the 14-day post-inoculation time point, and exhibits significant or severe adhesion to the attachment site. 49 days post-inoculation (as shown in the fourth image 1008) Figure 10 (As shown in the fifth image 1010) The tissue attachment grade of the upper left part of the artificial tissue is the same as the attachment grade 14 days after inoculation.

[0178] The second graph 1204 includes a plot of the tissue attachment grading of the right upper portion of the artificial tissue relative to the tissue attachment site of the well proximal to the tissue scaffold at all predetermined time points - 7 days post-seeding, 14 days post-seeding, and 49 days post-seeding. At all predetermined time points, there is no attachment of the right upper portion of the artificial tissue to the attachment site (the well proximal to the tissue scaffold).

[0179] The third graph 1206 includes a plot of the tissue attachment grading of the left lower portion of the artificial tissue relative to the tissue attachment site of the well proximal to the tissue scaffold. At 7 days post-seeding (as shown in the third image 1006), there is no attachment of the left lower portion of the artificial tissue to the attachment site and the artificial tissue is translucent. At 14 days post-seeding (as shown in the fourth image 1008), the left lower portion of the artificial tissue is translucent and has slight attachment to the attachment site. At 49 days post-seeding (as shown in the fifth image 1010), the left lower portion of the artificial tissue is not translucent and has major or severe attachment to the attachment site. Figure 10 Figure 10 The fourth graph 1208 includes a plot of the tissue attachment grading of the right lower portion of the artificial tissue relative to the tissue attachment site of the well proximal to the tissue scaffold. At all predetermined time points - 7 days post-seeding, 14 days post-seeding, and 49 days post-seeding - there is no attachment of the right lower portion of the artificial tissue to the attachment site (the well proximal to the tissue scaffold). Figure 10

[0180] Using the information provided in this portion of the QC report shown in

[0181] Using the information provided in this portion of the QC report shown in Figure 12 it is possible to determine that the artificial tissue has irregular attachment limited to the left side of the artificial tissue. This information can be used to modify or audit the manufacturing process and / or provide an explanation of the variability in the experimental output.

[0182] While the above description is directed to the use of the QC report during the maturation process of the artificial tissue, the tissue parameters, tissue descriptors, and / or the QC report can also be used post-maturation. For example, the QC report can be generated at the point in time when experimental data is generated or extracted from the artificial tissue. In such examples, the QC report (and / or the tissue parameters and tissue descriptors) provide an indication of the state of the artificial tissue at the point in time when the data is generated or extracted. Thus, the QC report can be used to gain interpretable insights into the experimental results and / or to track developmental, physiological, or morphological changes in the artificial tissue during the experimental procedure.

[0183] Having described the functionality of the quality control (QC) unit 108 shown in Figure 1 the description will now turn to the assay unit 110 shown in Figure 1

[0184] ​​​In general, the assay unit 110 determines changes in the artificial tissue phenotype as encoded by the tissue descriptors generated by the extraction unit 106. More specifically, the assay unit 110 is configured to compare a first tissue descriptor encoding a phenotype of the artificial tissue under a first set of conditions to a second tissue descriptor encoding a phenotype of the artificial tissue under a second set of conditions. The first set of conditions and the second set of conditions can be a reference condition (e.g., a control condition) or a perturbation condition associated with a perturbation of the artificial tissue (such as a drug treatment, a disease model, a different cell line, or a physical perturbation of the artificial tissue). For example, the first set of conditions can be a control condition and the second set of conditions can be a perturbation condition associated with a drug treatment. The relative change in the tissue descriptors provides a comparable encoding or characterization of the change in phenotype due to the change from the first set of conditions to the second set of conditions. The relative change can be used as a feature or feature vector in several downstream tasks such as drug development and discovery.

[0185] For example, a first tissue descriptor of the artificial tissue is obtained under a reference condition. The reference condition includes a carrier treatment of the artificial tissue. The first tissue descriptor includes the use of a plurality of features extracted from the reference condition using a feature extraction algorithm. The plurality of features includes a plurality of features extracted from the reference condition using a feature extraction algorithm. Figure 1The estimated smoothness of the artificial tissue obtained by the segmentation unit 104 and extraction unit 106 of system 100 is described in detail above. Because the first tissue descriptor is associated with the artificial tissue under reference conditions, it is considered as an encoding of the baseline phenotype of the artificial tissue. A second tissue descriptor of the artificial tissue is obtained under perturbation conditions. The perturbation conditions include drug treatment of the artificial tissue according to a first dose of the compound. That is, at the time point between obtaining the first tissue descriptor and obtaining the second tissue descriptor, the first dose of the compound is applied to the artificial tissue. The second tissue descriptor includes the estimated smoothness of the artificial tissue. The second tissue descriptor is considered as an encoding of the phenotype of the artificial tissue under the perturbation conditions caused by the first dose of the compound. The determination unit 110 is configured to compare the second tissue descriptor with the first tissue descriptor to determine the phenotypic change of the artificial tissue. That is, the relative change between the tissue descriptor obtained under reference conditions and the tissue descriptor obtained under perturbation conditions corresponds to the quantification of the phenotypic change of the artificial tissue due to the change in conditions. This relative change is then compared to a database of predetermined relative changes obtained from artificial tissues under conditions with known effects (e.g., changes from reference conditions to a set of perturbation conditions associated with a drug having a known mechanism of action). Based on this comparison, one or more closely matching relative changes are identified from the database. Because the identified relative changes are similar to those exhibited due to changes in conditions, it can be determined that compounds applied to artificial tissues may have effects similar to those associated with closely matching relative changes (e.g., similar toxicity, mechanism of action, etc.). Therefore, the geometric tissue descriptor described in this disclosure provides an effective and discriminative characterization of the phenotype of artificial tissues, which can be used to identify one or more effects associated with unknown compounds.

[0186] Figure 13 Variations in tissue phenotypes encoded by a geometric tissue descriptor according to one embodiment of the present disclosure are illustrated.

[0187] Figure 13 The figure shows the hypothetical effect of 2 μm on tissue volume (relative to the size of tissue treated with dimethyl sulfoxide (DMSO)). Figure 13 The plots depict the tissue volumes of drug-treated tissues (referred to as "compounds") and artificial tissues treated with compounds of 10 μm in size. The DMSO plots include tissue volume measurements obtained under control conditions from a cohort of eight engineered cardiac tissue samples. The mean tissue volume for the DMSO cohort is 0.3374 mm². 3 The standard deviation is 0.0214 mm. 3The first treatment plot (2 pm of compound) includes tissue volume measurements obtained from a population of five engineered heart tissue samples under perturbation conditions corresponding to a 2 pm dose of compound. The mean tissue volume for the first treatment population is 0.4125 mm 3 with a standard deviation of 0.0226 mm 3 . The second treatment plot (10 pm of compound) includes tissue volume measurements obtained from a population of seven engineered heart tissue samples under perturbation conditions corresponding to a 10 pm dose of compound. The mean tissue volume for the second treatment population is 0.4433 mm 3 with a standard deviation of 0.0305 mm 3 .

[0188] The relative difference between the DMSO population and the first treatment population is shown in Figure 13 as d1. The relative difference between the DMSO population and the second treatment population is shown in Figure 13 as d2. The relative difference between the first treatment population and the second treatment population is shown in Figure 13 as d3. Figure 13 Each of the relative differences shown can be used as an encoding of a phenotypic change. For example, the relative difference d1 encodes a phenotypic change from a reference condition to a perturbation condition including treatment of the compound at 2 pm.

[0189] An analysis of variance (ANOVA) test applied to the three populations shows a significant difference in volume between the populations (p < 9.6e -7). A Tukey test indicates that the DMSO population volume is significantly different from the first treatment population (p < 1.9e -4) and the second treatment population (p < 8.7e -7), with no statistically significant difference between the first treatment group and the second treatment group (p = 0.12).

[0190] Accordingly, the geometric encoding of tissue phenotypes provided by the tissue volume descriptors provides a powerful and discriminative feature that can be effectively used for a number of downstream drug discovery and development tasks.

[0191] The description will now turn to methods for quality control of artificial tissue, tissue attachment grading, and geometric encoding of tissue phenotypes using the components, processes, and operations of the systems described above.

[0192] Figure 14 A method 1400 for quality control (QC) of artificial tissue according to one aspect of the present disclosure is shown.

[0193] The method 1400 includes the steps of obtaining 1402 an image, extracting 1404 one or more image segments, determining 1406 one or more tissue parameters, and outputting 1408 a QC report. In one embodiment, the method 1400 is performed by a QC system 1000.Figure 1 The system 100 illustrated proceeds.

[0194] In general, the QC report generated by the method 1400 provides an automated indication of the state of the artificial tissue at the predetermined time point as encoded within one or more tissue parameters. Automation of quality control, in particular during the maturation period of the artificial tissue, helps to reduce the requirement for human involvement, thereby reducing user assessment bias and improving the scalability of the tissue engineering system. Furthermore, the QC report generated by the method 1400 can complement existing tissue assays to improve the interpretability of such models in the presence of experimental variability (e.g., by providing a quantitative explanation of the physiological state of the artificial tissue).

[0195] At an obtaining step 1402, a first image of the artificial tissue is obtained at a first predetermined time point. The first image comprises an image of the artificial tissue taken at the first predetermined time point.

[0196] The artificial tissue comprises muscle tissue, such as cardiac tissue or skeletal muscle tissue. In one embodiment, the first image comprises a brightfield microscopy image obtained from a sensor assembly of a bioreactor.

[0197] In one embodiment, the first predetermined time point comprises a predetermined time point during a maturation period of the artificial tissue. For example, the predetermined time point is 1 day post-seeding, 2 days post-seeding, 7 days post-seeding, 14 days post-seeding, or 49 days post-seeding. The first image can be obtained automatically (e.g., by a control unit or system such as the system 100 illustrated) at the predetermined time point. Figure 1 The system 100 illustrated).

[0198] At an extracting step 1404, one or more image segments are extracted from the first image. Each image segment of the one or more image segments comprises a region of interest of the artificial tissue within the first image (e.g., Figure 2 The well region 202, the first tissue scaffold region 204, the second tissue scaffold region 206, and the tissue region 208 illustrated).

[0199] As described in more detail above with respect to Figure 5 The one or more image segments are extracted from the first image using a segmentation unit (e.g., Figure 1 The segmentation unit 104 of the system 100 illustrated). In one embodiment, the one or more image segments are extracted from the first image using a segmentation pipeline. The segmentation pipeline utilizes a two-step segmentation process. A first segmentation step segments a well region (i.e., Figure 2 The well region 202 illustrated) from the first image. A second segmentation step segments a tissue region (i.e., Figure 2the tissue region 208 shown). In one embodiment, the segmentation pipeline includes a plurality of sequential morphological operations including an image equalization operation, an affine transformation, an image cropping operation, and binarization (as described above with respect to Figure 3 and 4 the segmentation pipeline shown).

[0200] At the determining step 1406, one or more tissue parameters are determined from the one or more image segments. The one or more tissue parameters are indicative of a physiological state of the artificial tissue at the first predetermined point in time. The physiological state of the artificial tissue includes one or more of: a maturation state, a presence of a disease phenotype, an effect of a drug treatment, and an irregularity associated with the tissue.

[0201] In one embodiment, the one or more tissue parameters include one or more of: a tissue irregularity grade; a geometric parameter; and a passive tension measure. The tissue irregularity grade is a non-geometric feature that quantifies an irregularity or issue associated with the artificial tissue, including a hole or discontinuity within the artificial tissue, an irregularity of a shape of the tissue, a fragment within the tissue, a broken tissue, and a broken tissue scaffold. In one embodiment, the tissue irregularity grade includes a tissue attachment grade associated with an attachment of the artificial tissue to a device within which the artificial tissue is contained. The tissue attachment grade is indicative of a degree of attachment of a portion of the artificial tissue to an attachment site of the device (e.g., a tissue scaffold of the device, a side wall of the device proximate to the tissue scaffold, or a side wall of the tissue distal to the tissue scaffold). The tissue attachment grade includes a plurality of scores, including an attachment score (i.e., a score or a label of the attachment output level 704 shown), an attachment severity score (i.e., a score or a label of the severity output level 706 shown), and a translucency score (i.e., a score or a label of the translucency output level 708 shown). The tissue attachment grade is determined using a grading model based on an image segment (e.g., a quadrant image such as the quadrant image shown) of the one or more image segments that includes the portion of the tissue. The grading model can be a hierarchy of classifiers, or a trained multi-task learning (MTL) network as shown and described in detail above. Figure 7 Figure 7 Figure 7 The attachment severity score is determined using a severity model based on the image segment (e.g., the quadrant image) of the one or more image segments that includes the portion of the tissue. The severity model can be a hierarchy of classifiers, or a trained MTL network as shown and described in detail above. Figure 6 Figures 9A-9C The translucency score is determined using a translucency model based on the image segment (e.g., the quadrant image) of the one or more image segments that includes the portion of the tissue. The translucency model can be a hierarchy of classifiers, or a trained MTL network as shown and described in detail above.

[0202] As described above with respect to Figure 5 ​​​In more detail, the geometric parameters include one of: a tissue length, a tissue width, a tissue volume, or a tissue area. The geometric parameters are determined from an image segment of the one or more image segments extracted at the extracting step 1404. The image segment includes a region of the first image that bounds the tissue (e.g., the tissue region 502 shown). The tissue length is associated with a length of a long axis of the image segment (e.g., the length between the first boundary point 508 and the second boundary point 510 of the tissue region 502 shown). The tissue width is associated with a length of a short axis of the image segment at a predetermined relative position along the long axis of the tissue (e.g., the short axis length at the first intermediate point 512-1, the second intermediate point 512-2, the centroid 506, etc.). The tissue volume is determined from a plurality of tissue widths according to a slice volume model. As described in more detail above with respect to Figure 5 In more detail, the slice volume model estimates a volume of a slice of the artificial tissue at a position along the longitudinal axis based on an estimated cross-sectional area of the artificial tissue at the position determined from the width of the artificial tissue at the position. The estimated cross-sectional area is determined according to a model parameterized by the width of the artificial tissue and a predetermined tissue elongation parameter, as shown in Equation (1). The predetermined tissue elongation parameter is in the range of 1 to 5. The tissue area is an estimated surface area of the artificial tissue captured within the image. Figure 5 Figure 5 The passive tension measure corresponds to an estimated tension exerted by the artificial tissue on the tissue support when in a resting state (i.e., without any external stimulus). The passive tension measure is determined from an image segment of the one or more image segments that includes the tissue support of the device (e.g., the first tissue support region 204 shown). The passive tension measure is estimated from a displacement model of the tissue support. The displacement model includes one or more linear models, or a linear model and a quadratic model. The model estimates a (vertical) distance between an anchor point of the tissue support and a most extreme curvature point of the first tissue support.

[0203] The passive tension measure corresponds to an estimated tension exerted by the artificial tissue on the tissue support when in a resting state (i.e., without any external stimulus). The passive tension measure is determined from an image segment of the one or more image segments that includes the tissue support of the device (e.g., the first tissue support region 204 shown). The passive tension measure is estimated from a displacement model of the tissue support. The displacement model includes one or more linear models, or a linear model and a quadratic model. The model estimates a (vertical) distance between an anchor point of the tissue support and a most extreme curvature point of the first tissue support. Figure 2 At the outputting step 1408, a QC report is outputted based on the one or more tissue parameters.

[0204] The QC report provides a representation of the physiological state of the artificial tissue at the first predetermined time point, and can track the physiological development and / or physiological state changes of the artificial tissue over the predetermined time points up to the first predetermined time point. The QC report is generated from the one or more tissue parameters such that the QC report provides an indication of the physiological state (and changes thereof) of the artificial tissue at the predetermined time points as encoded in the one or more tissue parameters.

[0205] and Figure 10 An example portion of the QC report is shown in FIGS. 1 1A-1 1C, as described in more detail above. Figure 12

[0206] ​​Additionally or alternatively, the QC report includes a tissue quality alert when at least one of the tissue parameters does not satisfy a predetermined quality control threshold. For example, the quality control threshold can require that the tissue attachment grade score for each of the attachment points of the tissue (i.e., the four corners of the artificial tissue) be above a threshold at a given point in time (e.g., 49 days post-seeding). In one embodiment, the method 1400 further includes a step of sending a command to the bioreactor based on the tissue quality alert. The bioreactor can be configured to process the command (e.g., by shutting down a device containing artificial tissue that failed to satisfy the quality control threshold, or issuing an alert at the bioreactor that emphasizes the existence of a quality issue associated with the artificial tissue that failed to satisfy the quality control threshold). Additionally or alternatively, the command is transmitted to another device (e.g., a computing device, a display device, etc.) to alert a user of the system that the quality control threshold has been satisfied or has not been satisfied.

[0207] In one embodiment, outputting the QC report includes storing or saving the QC report to a persistent storage device (such as a non-volatile memory, a non-transitory medium, etc.). Additionally or alternatively, outputting the QC report includes transmitting the QC report via a network (e.g., a local area network, a wide area network, etc.), or displaying the QC report or a portion thereof for viewing by a user.

[0208] Figure 15 A method 1500 for engineering tissue attachment grading is shown in accordance with one aspect of the present disclosure.

[0209] The method 1500 includes the steps of obtaining 1502 an image, extracting 1504 a first region, determining 1506 a plurality of attachment scores, determining 1508 an attachment grade, and outputting 1510 the attachment grade. In one embodiment, the steps of extracting 1504, determining 1506, determining 1508, and outputting 1510 are repeated for a plurality of regions extracted from the image. For example, these steps are repeated four times such that an attachment grade is determined for each quadrant image extracted from the image obtained at the obtaining step 1502. In one embodiment, the method 1500 is performed by the system 100 shown. Figure 1

[0210] ​Generally, Method 1500 uses a multi-task learning (MTL) network to determine the attachment grading of artificial tissue at the corresponding attachment site. Attachment of artificial tissue to non-tissue scaffold elements of the device, or insufficient attachment of artificial tissue to the tissue scaffold, can inhibit the functional response of the artificial tissue, resulting in irregular or inaccurate phenotypic measurements or functional response values. Therefore, the tissue attachment grading determined by Method 1500 helps identify and quantify problems related to the state of the artificial tissue (e.g., during maturation), which helps improve the quality of the resulting artificial tissue by identifying sources of systematic manufacturing variability. In the example QC process, artificial tissue identified as having attachments that meet specific criteria (e.g., severe attachment to the side of the well furthest from the tissue scaffold) can be removed, allowing new artificial tissue to grow in the appropriate location.

[0211] At step 1502, an image of the engineered tissue grown within the device is obtained. The device includes a tissue scaffold for attaching to the engineered tissue.

[0212] Bioreactors containing artificial tissues, directly or indirectly (e.g., Figure 1 The image shows a bioreactor 102. The bioreactor includes a device containing a tissue scaffold. The artificial tissue includes engineered muscle tissue, such as cardiac or musculoskeletal tissue. A first set of conditions associated with the artificial tissue includes reference or control conditions (such as carrier treatment of the artificial tissue). Alternatively, the first set of conditions includes perturbation conditions associated with perturbation of the artificial tissue (such as drug treatment, disease models, physical perturbation of different cell lines or artificial tissues).

[0213] At extraction step 1504, a first region is extracted from the image. The first region of the image includes a first portion of the engineered tissue and a portion of the tissue scaffold.

[0214] As mentioned above Figure 5 and Figure 6 To elaborate further, the first region corresponds to a portion of the image that includes engineered tissue (i.e., the "corner") and a portion of the tissue scaffold (such as...). Figure 6 The quadrant shown.

[0215] At step 1506, multiple attachment scores are determined based on the first region using a multi-task learning (MTL) network.

[0216] The architecture of the MTL network is described in more detail above with reference to Figure 9. The MTL network comprises an input convolutional neural network (e.g., ...) configured to receive an input image. Figure 9A The input network 902 shown is configured to estimate attachment scores associated with the input image, and multiple independent output networks (e.g., Figure 9AThe attached network 906-1, severity network 906-2, and semi-transparency network 906-3, auxiliary network 906-4, and in some embodiments, mild severity network 906-5 shown, as well as each independent output network coupled to the input convolutional neural network and multiple independent output networks (e.g., Figure 9A The shared network (904) shown is a shared network among multiple convolutional neural networks. The input convolutional neural network includes multiple convolutional neural networks, each including multiple initial layers. The shared network includes at least one pooling layer and multiple dense layers. Each of the multiple independent output networks includes multiple dense layers and at least one output layer. In one embodiment, the multiple independent output networks further include an auxiliary output network configured to predict attachment hierarchy (e.g., ...). Figure 9A The auxiliary output network shown is 906-4.

[0217] Multiple adhesion scores include adhesion metrics (i.e., Figure 7 The attached output level 704 score or label shown), attached severity score (i.e., Figure 7 The severity output level 706 is shown as a score or label) and the translucency score (i.e., Figure 7 The output level of semi-transparency is shown as a score or label of 708.

[0218] In one embodiment, method 1500 further includes the step of training an MTL network. As described above with respect to Figure 9, the MTL network is trained on multiple images of engineered tissue and associated tissue scaffolds. Each of the multiple images is associated with a corresponding plurality of attachment scores.

[0219] At step 1508, an adhesion grading for the first region is determined based on multiple adhesion scores. The adhesion grading indicates the degree of adhesion between the first portion of the engineered tissue and the adhesion sites of the device. In one embodiment, the adhesion grading includes a combination of multiple adhesion scores.

[0220] The attachment site associated with the attachment grading is a tissue scaffold (e.g., the attachment grading indicates the degree of attachment of a first portion of engineered tissue to that portion of the tissue scaffold shown in the first image), a first portion of the well of the device on the side closer to the tissue scaffold (e.g., the attachment grading indicates the degree of attachment of the first portion of engineered tissue to a first portion on one side of the well), or a second portion of the well of the device on the side farther from the tissue scaffold (e.g., the attachment grading indicates the degree of attachment of the first portion of engineered tissue to a second portion on one side of the well).

[0221] At output step 1510, the attachment grading is output.

[0222] In one embodiment, outputting the attachment grade includes storing or saving the attachment grade to a persistent storage device (such as a non-volatile memory, a non-transitory medium, etc.). Additionally or alternatively, outputting the attachment grade includes transmitting the attachment grade via a network (e.g., a local area network, a wide area network, etc.), or displaying the attachment grade or a portion thereof for a user to view. Additionally or alternatively, outputting the attachment grade includes including the attachment grade or a portion thereof in a QC report (e.g., as shown above Figure 12

[0223] Additionally or alternatively, the attachment grade is output as part of an alert when the attachment grade does not satisfy the predetermined attachment criteria. In one embodiment, a command is transmitted to a bioreactor containing the first tissue based on the alert. The bioreactor can be configured to process the command (e.g., by shutting down a device containing the artificial tissue that failed to satisfy the predetermined attachment criteria, or by issuing an alert at the bioreactor that emphasizes the existence of a quality issue associated with the artificial tissue that failed to satisfy the predetermined attachment criteria). Additionally or alternatively, the command is transmitted to another device (e.g., a computing device, a display device, etc.) to alert a user of the system that the predetermined attachment criteria has been satisfied or has not been satisfied.

[0224] Figure 16 A method 1600 for geometric encoding of tissue phenotypes according to one aspect of the present disclosure is shown.

[0225] The method 1600 includes the steps of obtaining 1602 an image, extracting 1604 a first region from the image, extracting 1606 one or more geometric features from the first region, and generating 1608 a first tissue descriptor. The method 1600 further includes the optional step of outputting 1610 the tissue descriptor. In one embodiment, the method 1600 is performed by the system 100 shown and described in detail above. Figure 1

[0226] In general, the method 1600 provides an efficient mechanism for encoding the phenotype of an artificial tissue according to one or more geometric features extracted from an image of the artificial tissue. The geometric features provide a shape-based characterization of the artificial tissue that can be used as a feature to determine changes in the phenotype of the artificial tissue.

[0227] At the obtaining step 1602, an image of an artificial tissue under a first set of conditions is obtained (e.g., the image 116 shown in Figure 1 The image is obtained directly or indirectly from a bioreactor (e.g., the bioreactor 102 shown in Figure 1 The image includes, for example, a brightfield microscopy image obtained from a sensor assembly of the bioreactor.

[0228] ​​Artificial tissues include engineered muscle tissues, such as cardiac or skeletal muscle tissues. The first set of conditions associated with artificial tissues includes reference or control conditions (such as vector treatment of the artificial tissue). Alternatively, the first set of conditions includes perturbation conditions associated with perturbations of the artificial tissue (such as drug treatment, disease models, physical perturbations of different cell lines or artificial tissues).

[0229] At extraction step 1604, a first region is extracted from the image, such that the first region defines the artificial tissue within the image. This is done using segmentation units (such as...). Figure 1 The segmentation unit 104 shown is used to extract the first region.

[0230] The first region corresponds to a specific portion of the image containing artificial tissue and / or a device containing artificial tissue. More specifically, the first region corresponds to a tissue region (such as...) Figure 2 The tissue region shown is 208.

[0231] A segmentation pipeline is used to extract the first region from the image. The segmentation pipeline utilizes a two-step segmentation process. The first segmentation step segments the well region (i.e., Figure 2 The well region 202 shown is illustrated. The second segmentation step segments the tissue region (i.e.,) from the image cropped according to the well region identified in the first segmentation step. Figure 2 The shown organization region 208). In one embodiment, the segmentation pipeline includes multiple sequential morphological operations, including image equalization, affine transformation, image cropping, and binarization (as described above regarding...). Figure 3 and 4 The segmented pipeline shown is described.

[0232] At extraction step 1606, one or more geometric features are extracted from the first region. The one or more geometric features provide a shape-based representation of the artificial tissue.

[0233] One or more geometric features or parameters are included at multiple corresponding locations along the longitudinal axis of the first region (e.g., with...). Figure 5 Multiple tissue widths are defined at the locations of the intermediate points 512-1, 512-2, 512-3, 512-4 and the centroid 506 shown. Each tissue width is perpendicular to the longitudinal axis of the first region. The corresponding positions are evenly spaced along the longitudinal axis of the first region.

[0234] Additionally or alternatively, one or more geometric features or parameters include tissue length corresponding to the longitudinal length of the artificial tissue.

[0235] Additionally or alternatively, one or more geometric features or parameters include tissue area corresponding to the estimated surface area of ​​the artificial tissue captured within the image.

[0236] Additionally or alternatively, the one or more geometric features or parameters include a tissue volume corresponding to an estimated volume of the artificial tissue. The tissue volume is determined from the plurality of tissue widths according to a slice volume model. As described above with respect to equation (3), the slice volume model estimates a volume of a slice of the artificial tissue at a location along the longitudinal axis based on an estimated cross-sectional area of the artificial tissue at the location. The estimated cross-sectional area is determined according to a model parameterized by the width of the artificial tissue and a predetermined tissue elongation parameter. The predetermined tissue elongation parameter is in a range of 1 to 5. Figure 5 In further detail, the slice volume model estimates a volume of a slice of the artificial tissue at a location along the longitudinal axis based on an estimated cross-sectional area of the artificial tissue at the location. The estimated cross-sectional area is determined according to a model parameterized by the width of the artificial tissue and a predetermined tissue elongation parameter, as shown in equation (1). The predetermined tissue elongation parameter is in a range of 1 to 5.

[0237] At a generate step 1608, a first tissue descriptor is generated based on the one or more geometric features. The first tissue descriptor encodes a phenotype of the artificial tissue under the first set of conditions.

[0238] In one embodiment, the first tissue descriptor includes an estimated smoothness of the artificial tissue determined from a variance of the plurality of tissue widths, as described in detail above with respect to equation (2). In another embodiment, the first tissue descriptor includes an estimated volume of the artificial tissue. That is, the first tissue descriptor corresponds to the tissue volume geometric feature. In another embodiment, the first tissue descriptor includes an estimated area of the artificial tissue. That is, the first tissue descriptor corresponds to the estimated area geometric feature.

[0239] At an optional output step 1610, the first tissue descriptor is output. In one embodiment, outputting the first tissue descriptor includes storing or saving the first tissue descriptor to a persistent storage device (such as a non-volatile memory, a non-transitory medium, etc.). Additionally or alternatively, outputting the first tissue descriptor includes transmitting the first tissue descriptor via a network (e.g., a local area network, a wide area network, etc.), or displaying the first tissue descriptor for viewing by a user.

[0240] Figure 17 A method 1700 for determining an effect associated with a phenotypic change is shown in accordance with one embodiment of the present disclosure.

[0241] The method 1700 includes the steps of obtaining 1702 a first tissue descriptor, obtaining 1704 a second tissue descriptor, comparing 1706 the tissue descriptors, determining 1708 a phenotypic change, comparing 1710 the phenotypic change to one or more predetermined changes, and determining 1712 an effect. The method 1700 further includes an optional step of outputting 1714 the effect. In one embodiment, the method 1700 is performed by the system 100 shown and described in detail above. Figure 1

[0242] ​In general, the method 1700 corresponds to a technique for determining an effect associated with a phenotypic change of an artificial tissue from a first set of conditions (e.g., a reference condition) to a second set of conditions (e.g., a perturbed condition). The method 1700 utilizes a relative change in the (geometric) tissue descriptor between the two conditions, thereby providing a comparable encoding or characterization of the phenotypic change due to the change from the first set of conditions to the second set of conditions. By mapping the relative change in the tissue descriptor to a database of predetermined tissue descriptors, the effect associated with the change from the first set of conditions to the second set of conditions can be effectively determined.

[0243] At an obtaining step 1702, a first tissue descriptor is obtained. The first tissue descriptor is associated with an artificial tissue under a first set of conditions. In one embodiment, the first tissue descriptor is obtained using the method 1600 shown. Figure 16 Alternatively, the first tissue descriptor is obtained from a persistent storage device or storage medium or via a network connection or other transmission medium.

[0244] At an obtaining step 1704, a second tissue descriptor is obtained. The second tissue descriptor is associated with the artificial tissue (to which the first tissue descriptor relates) under a second set of conditions. In one embodiment, the second tissue descriptor is obtained using the method 1600 shown. Figure 16 Alternatively, the second tissue descriptor is obtained from a persistent storage device or storage medium or via a network connection or other transmission medium.

[0245] In one embodiment, the first set of conditions (to which the first tissue descriptor relates) is a reference condition or control condition, and the second set of conditions is a perturbed condition. The perturbed condition is associated with a perturbation of the artificial tissue, such as a drug treatment, a disease model, a different cell line, or a physical perturbation of the artificial tissue.

[0246] The first tissue descriptor and the second tissue descriptor encode the phenotype of the artificial tissue under different conditions according to the same set of geometric features. For example, both the first tissue descriptor and the second tissue descriptor can be an estimated smoothness of the artificial tissue, an estimated volume of the artificial tissue, etc.

[0247] In a comparing step 1706, the first tissue descriptor is compared to the second tissue descriptor, and at a subsequent determining step 1708, a phenotypic change of the artificial tissue is determined based on the comparison made at the comparing step 1706.

[0248] Because the first tissue descriptor and the second tissue descriptor are quantitative numerical representations of the phenotype of the artificial tissue under different conditions, the comparison of the two tissue descriptors provides a quantitative numerical representation of the phenotypic change of the artificial tissue from the first set of conditions to the second set of conditions. This can be through Figure 13The relative changes δ1, δ2, δ3 between the conditions shown are illustrated.

[0249] In a comparison step 1710, the phenotypic change is compared to one or more predetermined phenotypic changes. Each predetermined phenotypic change of the one or more predetermined phenotypic changes is associated with a corresponding effect.

[0250] The one or more predetermined phenotypic changes are durably stored in a database (such as a relational or graph-based database, static file, or other structured representation). Each predetermined phenotypic change of the one or more predetermined phenotypic changes includes a relative change and at least one associated effect. For example, a predetermined phenotypic change can correspond to a relative change in tissue volume of an artificial heart tissue under a control condition compared to the artificial heart tissue under a first dose of a compound having a known toxicity; thus, the effect associated with the relative change would correspond to the known toxicity of the compound. As a further example, a predetermined phenotypic change can correspond to a relative change in tissue area of an artificial heart tissue under a condition corresponding to a particular disease state compared to the tissue area after a first drug treatment having a known mechanism of action has been administered; thus, the effect associated with the relative change would correspond to the known mechanism of action of the first drug treatment.

[0251] In a determination step 1712, an effect associated with the second set of conditions is determined based on the comparison made at the comparison step 1710.

[0252] The comparison step 1710 includes searching the one or more predetermined phenotypic changes and identifying a similarity or dissimilarity between the phenotypic change and the one or more predetermined phenotypic changes. The one or more predetermined phenotypic changes are searched using a linear or sequential search or a binary search. As a result of the comparison step 1710, one or more candidate predetermined phenotypic changes (i.e., predetermined phenotypic changes that are most similar to the phenotypic change) are identified. Based on the one or more candidate predetermined phenotypic changes, the effect associated with the second set of conditions is determined. For example, if a dissimilarity between the most similar predetermined phenotypic change and the phenotypic change is less than a predetermined threshold, the effect associated with the most similar predetermined phenotypic change is associated with the second set of conditions.

[0253] In an optional output step 1714, the effect is output. In one embodiment, outputting the effect includes storing or saving the effect to a persistent storage device (such as a non-volatile memory, a non-transitory medium, etc.). Additionally or alternatively, outputting the effect includes transmitting the effect via a network (e.g., a local area network, a wide area network, etc.) or displaying the effect for a user to view.

[0254] Figure 18 A bioreactor system 1800 according to an embodiment of the present disclosure is shown.

[0255] The bioreactor system 1800 includes a bioreactor 1802 and a control unit 1804. The bioreactor 1802 includes a device 1806 for growing tissue, a sensor assembly 1808, and an interface 1810. The interface 1810 communicatively couples the bioreactor 1802 and the control unit 1804 so that data can be exchanged between the bioreactor 1802 and the control unit 1804. In one embodiment, the bioreactor 1802 corresponds to the bioreactor 102 of the system 100 shown in FIG. 1, and the control unit 1804 corresponds to a subsystem of the system 100 (such as the first subsystem 122 or the second subsystem 124). Figure 1 The bioreactor 102 of the system 100 shown in FIG. 1, and the control unit 1804 corresponds to a subsystem of the system 100 (such as the first subsystem 122 or the second subsystem 124).

[0256] As shown in the expanded portion 1806-1 of the device 1806, the device 1806 or substrate includes one or more wells (such as the well 1814), one or more cell culture wells (such as the cell culture well 1816), a pair of electrodes (including the first electrode 1818-1 and the second electrode 1818-2), and a pair of elements (including the first element 1820-1 and the second element 1820-2). The well 1814 is positioned in the cell culture well 1816 and has a bottom on the device 1806, a first end 1822-1, and a second end 1822-2. The well 1814 is configured for growth of an engineered tissue 1824 from cells seeded therein. Media can be added to the cell culture well 1816 for growing and / or maintaining the engineered tissue 1824. The engineered tissue 1824 (alternatively referred to as an artificial tissue) includes an engineered muscle tissue. In one embodiment, the engineered tissue is an engineered cardiac tissue. In an alternative embodiment, the engineered muscle tissue is an engineered skeletal muscle tissue.

[0257] The pair of electrodes are separated by a gap within which the well 1814 is positioned. The pair of electrodes are configured to apply electrical stimulation to a cell culture (e.g., the engineered tissue 1824 within the well 1814 shown in the expanded portion 1806-1) within one or more wells of the device 1806. During maturation of the cell culture within the device 1806, the pair of electrodes apply stimulation to the cell culture according to a multi-week electrical stimulation protocol. After maturation of the cell culture, the pair of electrodes can be configured to stimulate the cell culture (e.g., the engineered tissue 1824) at a set frequency or pacing frequency.

[0258] The first and second elements 1820-1 and 1820-2 are disposed across the well 1814 such that there is a gap between the pair of elements and the bottom of the well 1814. The first and second elements 1820-1 and 1820-2 are configured to: (a) permit attachment of engineered tissue 1824 formed therebetween, thereby suspending the engineered tissue 1824 above the bottom of the well 1814; and (b) deform in response to a contraction force exerted on the pair of elements by the engineered tissue 1824, thereby mimicking the physiological environment of the engineered tissue 1824 itself and / or permitting measurement of the contraction force (e.g., by the sensor assembly 1808). For example, the pair of electrodes can subject the engineered tissue 1824 to electrical stimulation at a frequency of 0.1 Hz. The engineered tissue 1824 will contract in response to the electrical stimulation, causing deformation of at least one of the first and second elements 1820-1 and 1820-2.

[0259] The sensor assembly 1808 is configured to obtain a plurality of images of the tissue (e.g., engineered tissue 1824) at a plurality of time points. For example, the optical sensor can be configured to capture images or frames of the tissue at predetermined time points (e.g., 1 day post-seeding, 2 days post-seeding, etc.). In one embodiment, the control unit 1804 is configured to send a command 1826 to the bioreactor 1802 that causes the sensor assembly 1808 to obtain an image 1828 of the tissue (e.g., engineered tissue 1824), which is then returned to the control unit 1804.

[0260] Figure 19 An example computing system for implementing the methods of the present disclosure is shown. In particular, Figure 19 A block diagram of an embodiment of a computing system according to example embodiments of the present disclosure is shown. Figure 19 The computing system shown can correspond to part or all of any of the functional units described above.

[0261] The computing system 1900 can be configured to perform any of the operations disclosed herein, such as, for example, with reference to the description of the Figure 1Any of the operations described with reference to the functional units discussed. The computing system includes one or more computing devices 1902. The one or more computing devices 1902 of the computing system 1900 include one or more processors 1904 and memory 1906. The one or more processors 1904 can be any general purpose processor configured to execute a set of instructions, such as computing instructions, including instructions implemented in one or more programming languages (such as Python, Go, C, C++, C#, Java, etc.). For example, the one or more processors 1904 can be one or more general purpose processors, one or more field programmable gate arrays (FPGAs), and / or one or more application specific integrated circuits (ASICs). In one embodiment, the one or more processors 1904 include one processor. Alternatively, the one or more processors 1904 include multiple processors operatively connected. The one or more processors 1904 are communicatively coupled to the memory 1906 via an address bus 1908, a control bus 1910, and a data bus 1912. The memory 1906 can be a random access memory (RAM), read only memory (ROM), a persistent storage device (such as a hard disk), an erasable programmable read only memory (EPROM), etc. The one or more computing devices 1902 further include an I / O interface 1914 communicatively coupled to the address bus 1908, the control bus 1910, and the data bus 1912.

[0262] The memory 1906 can store information accessible to the one or more processors 1904. For example, the memory 1906 (e.g., one or more non-transitory computer-readable storage media, i.e., memory devices) can include computer-readable instructions (not shown) (e.g., computing instructions) executable by the one or more processors 1904. The computer-readable instructions can be software written in any suitable programming language or can be implemented in hardware. Additionally or alternatively, the computer-readable instructions can be executed on the one or more processors 1904 in logically and / or virtually separate threads. For example, the memory 1906 can store instructions (not shown) (such as computing instructions) that, when executed by the one or more processors 1904, cause the one or more processors 1904 to perform operations, such as any of the operations and functionalities of the computing system 1900 as described herein. Additionally or alternatively, the memory 1906 can store data (not shown) that is obtainable, received, accessed, written, manipulated, created, and / or stored. The data can include, for example, data and / or information described herein. In some implementations, the one or more computing devices 1902 can obtain data from and / or store data in one or more memory devices that are remote from the computing system 1900. Figures 1-14 The memory 1906 can store information accessible to the one or more processors 1904. For example, the memory 1906 (e.g., one or more non-transitory computer-readable storage media, i.e., memory devices) can include computer-readable instructions (not shown) (e.g., computing instructions) executable by the one or more processors 1904. The computer-readable instructions can be software written in any suitable programming language or can be implemented in hardware. Additionally or alternatively, the computer-readable instructions can be executed on the one or more processors 1904 in logically and / or virtually separate threads. For example, the memory 1906 can store instructions (not shown) (such as computing instructions) that, when executed by the one or more processors 1904, cause the one or more processors 1904 to perform operations, such as any of the operations and functionalities of the computing system 1900 as described herein. Additionally or alternatively, the memory 1906 can store data (not shown) that is obtainable, received, accessed, written, manipulated, created, and / or stored. The data can include, for example, data and / or information described herein. In some implementations, the one or more computing devices 1902 can obtain data from and / or store data in one or more memory devices that are remote from the computing system 1900.

[0263] The computing system 1900 further includes a storage unit 1916, a network interface 1918, an input controller 1920, and an output controller 1922. The storage unit 1916, the network interface 1918, the input controller 1920, and the output controller 1922 are communicatively coupled to the central control unit (i.e., the memory 1906, the address bus 1908, the control bus 1910, and the data bus 1912) via the I / O interface 1914.

[0264] The storage unit 1916 is a computer-readable medium, preferably a non-transitory computer-readable medium, that includes one or more programs including computing instructions that, when executed by the one or more processors 1904, cause the computing system 1900 to perform the method steps of the present disclosure. Alternatively, the storage unit 1916 is a transitory computer-readable medium. The storage unit 1916 can be a persistent storage such as a hard drive, cloud storage, or any other appropriate storage.

[0265] The network interface 1918 can be a Wi-Fi module, a network interface card, a Bluetooth module, and / or any other suitable wired or wireless communication device. In one embodiment, the network interface 1918 is configured to connect to a network such as a local area network (LAN), a wide area network (WAN), the Internet, or an intranet.

Claims

1. A method for geometric coding of organizational phenotypes, the method comprising: Obtain images of the artificial tissue under the first set of conditions; Extract a first region from the image, such that the first region defines the artificial tissue within the image; Extract one or more geometric features from the first region, wherein the one or more geometric features provide a shape-based representation of the artificial tissue; as well as A first tissue descriptor is generated based on the one or more geometric features, wherein the first tissue descriptor encodes the phenotype of the artificial tissue under the first set of conditions.

2. The method according to claim 1, further comprising: Output the first organization descriptor.

3. The method of claim 1, wherein the one or more geometric features include a plurality of tissue widths defined at corresponding plurality of locations along the longitudinal axis of the first region, wherein each of the plurality of tissue widths is perpendicular to the longitudinal axis of the first region.

4. The method of claim 3, wherein the corresponding plurality of positions are uniformly spaced apart along the longitudinal axis of the first region.

5. The method of claim 3, wherein the first tissue descriptor includes an estimated smoothness of the artificial tissue determined from the variance of the plurality of tissue widths.

6. The method of claim 3, wherein the first tissue descriptor comprises the estimated volume of the artificial tissue determined from the plurality of tissue widths according to a slice volume model.

7. The method of claim 6, wherein the slice volume model estimates the volume of a slice of the artificial tissue at a location along the longitudinal axis based on an estimated cross-sectional area of ​​the artificial tissue, wherein the estimated cross-sectional area is determined from the width of the artificial tissue at that location.

8. The method of claim 7, wherein the estimated cross-sectional area is determined based on a model parameterized by the width of the artificial tissue and a predetermined tissue elongation parameter.

9. The method of claim 8, wherein the predetermined tissue elongation parameter is in the range of 1 to 5.

10. The method of claim 1, wherein the first organization descriptor comprises the estimated area of ​​the first region.

11. The method of claim 1, further comprising: Obtain a second tissue descriptor associated with the artificial tissue under the second set of conditions; as well as The phenotypic changes of the artificial tissue are determined based on a comparison between the first tissue descriptor and the second tissue descriptor.

12. The method of claim 11, further comprising: Compare the first organization descriptor and the second organization descriptor.

13. The method of claim 11, wherein the first set of conditions is a reference condition.

14. The method of claim 11, wherein the second set of conditions is a perturbation condition associated with a perturbation of the artificial tissue.

15. The method of claim 14, wherein the perturbation comprises one or more of the following: drug treatment; disease model; or different cell lines.

16. The method of claim 11, further comprising: The phenotypic change is compared with one or more predetermined phenotypic changes, each predetermined phenotypic change being associated with a corresponding effect; as well as The effects associated with the second set of conditions are determined based on the comparison.

17. The method of claim 16, further comprising: Output the effect.

18. The method of claim 16, wherein the effect comprises one or more of a mechanism of action or toxicity.

19. The method of claim 1, wherein the first region is extracted from the image using a segmentation pipeline.

20. The method of claim 19, wherein the segmentation pipeline comprises a plurality of sequential morphological operations.

21. The method of claim 19, wherein the segmentation pipeline comprises one or more of the following: image equalization operation; affine transformation; image cropping operation; and binarization.

22. The method of claim 1, wherein the image comprises a bright-field microscopy image.

23. The method of claim 1, wherein the image is obtained from a bioreactor in which the artificial tissue is grown.

24. The method of claim 1, wherein the artificial tissue comprises muscle tissue.

25. The method of claim 24, wherein the muscle tissue is heart tissue.

26. The method of claim 24, wherein the muscle tissue is skeletal muscle tissue.

27. An apparatus comprising: One or more processors; as well as A memory storing instructions that, when executed by the one or more processors, cause the one or more processors to: Obtain images of the artificial tissue under the first set of conditions; Extract a first region from the image, such that the first region defines the artificial tissue within the image; Extract one or more geometric features from the first region, wherein the one or more geometric features provide a shape-based representation of the artificial tissue; as well as A first tissue descriptor is generated based on the one or more geometric features, wherein the first tissue descriptor encodes the phenotype of the artificial tissue under the first set of conditions.

28. A non-transitory machine-readable medium comprising instructions that, when executed by one or more processors, cause the one or more processors to: Obtain images of the artificial tissue under the first set of conditions; Extract a first region from the image, such that the first region defines the artificial tissue within the image; Extract one or more geometric features from the first region, wherein the one or more geometric features provide a shape-based representation of the artificial tissue; as well as A first tissue descriptor is generated based on the one or more geometric features, wherein the first tissue descriptor encodes the phenotype of the artificial tissue under the first set of conditions.

29. A system for quality control (QC) of artificial tissues, the system comprising: Bioreactor, comprising: Devices configured to promote tissue growth; and Sensor components configured to acquire one or more images of tissue within the device; and A QC unit communicatively coupled to the bioreactor, the QC unit comprising one or more processors and a memory storing instructions, the instructions causing the one or more processors, when executed by the one or more processors, to: A first image of the tissue within the device is obtained from the bioreactor at a first predetermined time point; Extract one or more image segments from the first image, wherein each of the one or more image segments includes a region of interest of the tissue within the first image; Determine one or more tissue parameters from the one or more image segments, wherein the one or more tissue parameters indicate the physiological state of the tissue at the first predetermined time point; and A QC report is output based on one or more of the organizational parameters.

30. The system of claim 29, wherein the physiological state includes one or more of the following: a mature state, the presence of a disease phenotype, a therapeutic effect of a drug, and irregularities associated with the tissue.

31. The system of claim 30, wherein the physiological state includes a mature state, and the first predetermined time point is a time point occurring during the maturation period of the tissue.

32. The system of claim 29, wherein the one or more tissue parameters include one or more of the following: tissue irregularity gradation; geometric parameters; and passive tension strength.

33. The system of claim 32, wherein the tissue irregularity grading includes tissue attachment grading, the tissue attachment grading indicating the degree of attachment of a portion of the artificial tissue to the attachment site of the device.

34. The system of claim 33, wherein the attachment site is one of the tissue scaffold of the device, a sidewall of the device near the tissue scaffold, and a sidewall of the device away from the tissue scaffold.

35. The system of claim 33, wherein the tissue attachment grading comprises a plurality of scores, the plurality of scores including an attachment score, an attachment severity score, and a translucency score.

36. The system of claim 33, wherein the organization irregularity grading is determined based on the image segments in the one or more image segments using a grading model.

37. The system of claim 36, wherein the image segment includes the quadrant of the first image that includes the portion of the tissue.

38. The system of claim 36, wherein the hierarchical model comprises a hierarchical structure of classifiers.

39. The system of claim 38, wherein the hierarchical model comprises a trained multi-task learning (MTL) network.

40. The system of claim 39, wherein the trained MTL network comprises an input convolutional neural network configured to receive an input image, a plurality of independent output networks each configured to estimate a score associated with the input image, and a shared network coupled between the input convolutional neural network and each of the plurality of independent output networks.

41. The system of claim 32, wherein the geometric parameters include one of the following: tissue length, tissue width, tissue volume, or tissue area.

42. The system of claim 41, wherein the geometric parameters are determined from an image segment of the one or more image segments, the image segment including a region of the first image that defines the tissue.

43. The system of claim 42, wherein the tissue length is associated with the length of the major axis of the image segment.

44. The system of claim 43, wherein the tissue width is associated with a predetermined relative position along the long axis of the tissue.

45. The system of claim 44, wherein the tissue width is associated with the length of the minor axis of the image fragment at the predetermined relative position.

46. ​​The system of claim 32, wherein the passive tension strength is determined from an image segment of the one or more image segments, the image segment including a region of the first image comprising the tissue scaffold of the device.

47. The system of claim 46, wherein the passive tension force is estimated from a displacement model of the tissue scaffold.

48. The system of claim 47, wherein the model comprises one or more linear models.

49. The system of claim 48, wherein the model comprises a linear model and a quadratic model.

50. The system of claim 29, wherein the one or more processors of the QC unit are configured to extract the one or more image segments from the first image using a segmentation pipeline.

51. The system of claim 50, wherein the segmentation pipeline comprises a plurality of sequential morphological operations.

52. The system of claim 50, wherein the segmentation pipeline includes one or more of the following: image equalization operation; affine transformation; image cropping operation; and binarization.

53. The system of claim 29, wherein the first image comprises a bright-field microscope image.

54. The system of claim 29, wherein the QC report includes an organizational quality alert.

55. The system of claim 54, wherein the QC report includes the organizational quality warning when at least one of the one or more organizational parameters fails to meet a predetermined quality control threshold.

56. The system of claim 54, wherein the instructions, when executed by the one or more processors of the QC unit, further cause the one or more processors to: Commands are sent to the bioreactor based on the tissue quality warning.

57. The system of claim 29, wherein the tissue comprises muscle tissue.

58. The system of claim 57, wherein the muscle tissue is heart tissue.

59. The system of claim 57, wherein the muscle tissue is skeletal muscle tissue.

60. A method for quality control (QC) of artificial tissues, the method comprising: A first image of the artificial tissue is obtained by one or more processors at a first predetermined time point; One or more image segments are extracted from the first image by the one or more processors, wherein each of the one or more image segments includes a region of interest of the artificial tissue within the first image; One or more tissue parameters are determined from the one or more image segments by the one or more processors, wherein the one or more tissue parameters indicate the maturity state of the artificial tissue at the first predetermined time point; as well as The one or more processors output a QC report based on the one or more organizational parameters.

61. A non-transitory machine-readable medium comprising instructions that, when executed by one or more processors, cause the one or more processors to: Obtain the first image of the artificial tissue at the first predetermined time point; Extract one or more image segments from the first image, wherein each of the one or more image segments includes a region of interest of the artificial tissue within the first image; One or more tissue parameters are determined from the one or more image segments, wherein the one or more tissue parameters indicate the maturity state of the artificial tissue at the first predetermined time point; as well as A QC report is output based on one or more of the organizational parameters.

62. A system for grading engineered tissue attachment, the system comprising: Multi-task learning (MTL) networks, which include: The input is a convolutional neural network, which is configured to receive an input image; Multiple independent output networks, each configured to estimate an attachment score associated with the input image; and A shared network, which is coupled between the input convolutional neural network and each of the plurality of independent output networks; A control unit, communicatively coupled to the MTL network, and including one or more processors and computing instructions, which, when executed by the one or more processors, cause the one or more processors to: Obtaining images of engineered tissue grown within a device, the device including a tissue scaffold for attaching to the engineered tissue; Extract a first region from the image, wherein the first region of the image includes a first portion of the engineered tissue and a portion of the tissue scaffold; The MTL network is used to determine multiple attachment scores based on the first region; An adhesion grade for the first region is determined based on the plurality of adhesion scores, wherein the adhesion grade indicates the degree of adhesion between the first portion of the engineered tissue and the adhesion site of the device; and Output the adhesion classification.

63. The system of claim 62, wherein the attachment site is one of the tissue scaffold, a first portion of the well of the device on the side closer to the tissue scaffold, and a second portion of the well of the device on the side farther from the tissue scaffold.

64. The system of claim 62, wherein the plurality of adhesion scores includes an adhesion index, an adhesion severity score, and a semi-transparency score.

65. The system of claim 62, wherein the adhesion grading comprises a combination of the plurality of adhesion fractions.

66. The system of claim 62, wherein the plurality of independent output networks further includes an auxiliary output network configured to predict the attachment gradation.

67. The system of claim 62, wherein the input convolutional neural network comprises a plurality of convolutional neural networks.

68. The system of claim 67, wherein the plurality of convolutional neural networks comprises a plurality of initial blocks.

69. The system of claim 62, wherein each of the plurality of independent output networks comprises a plurality of dense layers and at least one output layer.

70. The system of claim 62, wherein the shared network comprises at least one pooling layer and a plurality of dense layers.

71. The system of claim 62, wherein the first region includes a first quadrant of the image.

72. The system of claim 62, wherein the computation instructions are further configured to cause the one or more processors, when executed by the one or more processors: The MTL is trained on a training dataset comprising multiple images of engineered tissue and associated tissue scaffolds, each of which is associated with a corresponding plurality of attachment scores.

73. The system of claim 62, wherein the image is obtained from a bioreactor in which the engineered tissue is grown.

74. The system of claim 73, further comprising the bioreactor.

75. The system of claim 62, wherein the adhesion rating is output as part of an alert.

76. The system of claim 75, wherein the warning is issued when the adhesion rating does not meet the predetermined adhesion criteria.

77. The system of claim 62, wherein the engineered tissue comprises muscle tissue.

78. The system of claim 77, wherein the muscle tissue is heart tissue.

79. The system of claim 77, wherein the muscle tissue is skeletal muscle tissue.

80. A method for classifying the attachment of engineered tissues, the method comprising: Obtaining images of engineered tissue grown within a device, the device including a tissue scaffold for attaching to the engineered tissue; Extract a first region from the image, wherein the first region of the image includes a first portion of the engineered tissue and a portion of the tissue scaffold; Multiple attachment scores are determined based on the first region using a multi-task learning (MTL) network; An adhesion grade for the first region is determined based on the plurality of adhesion scores, wherein the adhesion grade indicates the degree of adhesion between the first portion of the engineered tissue and the adhesion site of the device; and Output the adhesion classification.

81. The method of claim 80, wherein the MTL network comprises an input convolutional neural network configured to receive an input image, a plurality of independent output networks each configured to estimate an attachment score associated with the input image, and a shared network coupled between the input convolutional neural network and each of the plurality of independent output networks.

82. A non-transitory machine-readable medium comprising instructions that, when executed by one or more processors, cause the one or more processors to: Obtaining images of engineered tissue grown within a device, the device including a tissue scaffold for attaching to the engineered tissue; Extract a first region from the image, wherein the first region of the image includes a first portion of the engineered tissue and a portion of the tissue scaffold; Multiple attachment scores are determined based on the first region using a multi-task learning (MTL) network; An adhesion grade for the first region is determined based on the plurality of adhesion scores, wherein the adhesion grade indicates the degree of adhesion between the first portion of the engineered tissue and the adhesion site of the device; and Output the adhesion classification.

83. The non-transitory machine-readable medium of claim 82, wherein the MTL network comprises an input convolutional neural network configured to receive an input image, a plurality of independent output networks each configured to estimate an attachment score associated with the input image, and a shared network coupled between the input convolutional neural network and each of the plurality of independent output networks.

Images