Method implemented by computer system, and quality analysis system and computer-readable medium

Abstract

The present disclosure relates to a method implemented by a computer system, and an oocyte quality analysis system and a non-transitory computer-readable medium that stores instructions, and particularly relates to a system and method for assessing the viability of an oocyte. In some embodiments, an image sequence related to an oocyte drawn into a pressure tool is acquired. By using a segmentation model, objects related to the oocyte can be identified. On the basis of the identified objects, features such as morphological features and the drawing depth related to the oocyte can be determined. Then, at least some of the features can be input into a machine learning model to generate an oocyte grade that indicates the probability of the oocyte developing into a usable blastocyst. Optionally, the oocyte grade can be presented to a user by means of an interactive user interface.

Description

Computer system implemented method, quality analysis system, and computer readable medium

[0001] Patent application related documents

[0002] This application claims priority to U.S. Provisional Patent Application No. 63 / 663,627, filed June 24, 2024, entitled “Oocyte Quality Analysis System,” the disclosure of which is incorporated herein by reference in its entirety and for all purposes. TECHNICAL FIELD

[0003] The present disclosure relates to medical analysis using machine learning models, and more specifically, to analyzing oocytes using sensor information and machine learning models, a computer system implemented method, quality analysis system, and computer readable medium. BACKGROUND

[0004] In recent years, progress has been made in the treatment of infertility. Treatments include, for example, in vitro fertilization (IVF). IVF begins with an ovarian stimulation phase, which stimulates the production of eggs. Eggs (oocytes) can be removed from a patient and fertilized in vitro to form embryos. Embryos can be subjected to a number of tests and analyses in an effort to select the most viable and / or most advantageous embryo for implantation. However, these tests and analyses present technical problems, and thus, an accurate protocol to achieve such selection presents technical challenges.

[0005] For example, different scoring systems have been developed to assist in determining the viability of each embryo. These scoring systems typically involve manual annotation of embryo images or time-lapse dynamic images. As can be appreciated, the selection process proves to be error prone, for example, due to subjective judgment by embryologists. Moreover, there is currently no universally adopted standard scoring system to select high quality oocytes. Current automated techniques for embryo analysis are inaccurate and fail to utilize different sensor information. SUMMARY

[0006] The present disclosure provides a computer system implemented method, system, and non-transitory computer readable medium storing instructions, particularly systems and methods for assessing oocyte viability.

[0007] A method implemented by a computer system is provided, comprising the following steps. A plurality of images forming an image sequence related to a time period are acquired, the image sequence depicting an oocyte and a portion of a tool exerting pressure on the oocyte, wherein individual ones of the images are associated with individual pressure values exerted on the oocyte at respective image capture times. Thereafter, an object associated with the oocyte is identified by a segmentation model. Next, a feature associated with the oocyte is determined based on geometric measurements of at least part of the object associated with the oocyte, the feature comprising a morphological feature indicative of the oocyte measurement over the time period, and a suction depth of the oocyte into the portion of the tool exerting pressure on the oocyte. Then, an oocyte grade is generated by a machine learning model at input values comprising the suction depth, wherein the oocyte grade is indicative of at least a likelihood of the oocyte developing into a usable blastocyst.

[0008] In an embodiment of the present invention, determining the feature associated with the oocyte comprises: determining geometric information associated with the object; and calculating the feature associated with the oocyte based on the geometric information associated with the object.

[0009] In an embodiment of the present invention, the object associated with the oocyte comprises at least one of a zona pellucida of the oocyte, a perivitelline space of the oocyte, a first polar body of the oocyte, a cytoplasm of the oocyte, and a bounding box associated with the portion of the tool exerting pressure on the oocyte.

[0010] In an embodiment of the present invention, the morphological feature associated with the oocyte comprises at least one of an ellipticity of the first polar body, a thickness of the zona pellucida, a diameter of the oocyte, an area of the cytoplasm, a compactness of the cytoplasm, a circularity of the cytoplasm, and a ratio between the area of the cytoplasm and a total area of cytoplasm and perivitelline space.

[0011] In an embodiment of the present invention, the segmentation model is a U-Net.

[0012] In an embodiment of the present invention, the machine learning model is a regression model.

[0013] In an embodiment of the present invention, the regression model comprises a plurality of weights, and wherein each of the plurality of weights is associated with one of the features used to generate the oocyte grade.

[0014] In one embodiment of the invention, the input value further includes at least a subset of the morphological features, wherein the subset of morphological features includes at least one of the following: the ellipticity of the first polar body of the oocyte, the thickness of the zona pellucida of the oocyte, the diameter of the oocyte, the area of ​​the cytoplasm of the oocyte, the compactness of the cytoplasm of the oocyte, the roundness of the cytoplasm of the oocyte, and the ratio between the area of ​​the cytoplasm and the total area of ​​the cytoplasm and the perivitelline space.

[0015] In one embodiment of the invention, the portion of the tool is a micropipette, and wherein during the time period, the micropipette contacts the oocyte and is configured to apply the pressure.

[0016] In one embodiment of the invention, the feature includes the inner diameter of the micro-pipette, wherein the inner diameter is used to normalize the morphological feature or the inhalation depth.

[0017] In one embodiment of the invention, determining the feature includes: for each individual image in the images, identifying bounding boxes based at least on the position of the portion of the tool in the individual image to obtain a plurality of bounding boxes; and calculating the inhalation depth based on a first bounding box having the maximum distance in a first direction among the plurality of bounding boxes.

[0018] In one embodiment of the invention, determining the feature includes: tagging each individual image in the images; identifying a bounding box based at least on the position of the portion of the tool in the individual image; and using the pixel intensity of at least a portion of the pixels in the individual image within the bounding box to determine the aspiration depth of the oocyte entering the portion of the tool.

[0019] The present invention provides a system comprising one or more processors and a non-transitory computer storage medium storing instructions, wherein, when the instructions are executed by one or more processors, the one or more processors cause the processors to: acquire a plurality of images forming a time-related image sequence, the image sequence depicting an oocyte and a portion of a tool applying pressure to the oocyte, wherein individual images are associated with individual pressure values ​​applied to the oocyte at their respective image capture times; identify objects associated with the oocyte using a segmentation model; determine features associated with the oocyte based on geometric measurements of at least partially related objects, the features including morphological features indicative of measurements of the oocyte within the time period, and the aspiration depth of the oocyte into the portion of the tool applying pressure to the oocyte; and generate an oocyte grade using a machine learning model based on an input value including the aspiration depth, wherein the oocyte grade at least indicates the likelihood that the oocyte will develop into a usable blastocyst.

[0020] In one embodiment of the present invention, determining the features associated with the oocyte includes: determining geometric information associated with the object; and calculating the features associated with the oocyte based on the geometric information associated with the object.

[0021] In one embodiment of the invention, the object associated with the oocyte includes at least one of the following: the zona pellucida of the oocyte, the perivitelline space of the oocyte, the first polar body of the oocyte, the cytoplasm of the oocyte, and a bounding box associated with the portion of the tool that applies pressure to the oocyte.

[0022] In one embodiment of the invention, the morphological features associated with the oocyte include at least one of the following: the ellipticity of the first polar body, the thickness of the zona pellucida, the diameter of the oocyte, the area of ​​the cytoplasm, the compactness of the cytoplasm, the roundness of the cytoplasm, and the ratio between the area of ​​the cytoplasm and the total area of ​​the cytoplasm and the perivitelline space.

[0023] In one embodiment of the invention, the input value further includes at least a subset of the morphological features, wherein the subset of morphological features includes at least one of the ellipticity of the first polar body of the oocyte, the thickness of the zona pellucida of the oocyte, the diameter of the oocyte, the area of ​​the cytoplasm of the oocyte, the compactness of the cytoplasm of the oocyte, the roundness of the cytoplasm of the oocyte, and the ratio between the area of ​​the cytoplasm and the total area of ​​the cytoplasm and the perivitelline space.

[0024] In one embodiment of the invention, determining the feature includes: for each individual image in the images, identifying bounding boxes based at least on the position of the portion of the tool in the individual image to obtain a plurality of bounding boxes; and calculating the inhalation depth based on a first bounding box having the maximum distance in a first direction among the plurality of bounding boxes.

[0025] The present invention provides a non-transitory computer-readable medium storing one or more instructions, which, when executed by one or more processors, cause the one or more processors to: acquire a plurality of images forming a time-related image sequence, the image sequence depicting a portion of an oocyte and a tool applying pressure to the oocyte, wherein individual images are associated with individual pressure values ​​applied to the oocyte at their respective image capture times; identify objects associated with the oocyte using a segmentation model; and determine features associated with the oocyte based on geometric measurements of at least partially related objects, the features including morphological features indicating measurements of the oocyte within the time period, and the oocyte... The aspiration depth of the cell into the portion of the tool that applies pressure to the oocyte; and the generation of an oocyte grade by a machine learning model based on input values ​​including the aspiration depth and at least a subset of the morphological features, wherein the oocyte grade at least indicates the likelihood that the oocyte will develop into a usable blastocyst, and wherein the subset of the morphological features includes at least one of the ellipticity of the oocyte's first polar body, the thickness of the zona pellucida of the oocyte, the diameter of the oocyte, the area of ​​the oocyte's cytoplasm, the compactness of the oocyte's cytoplasm, the roundness of the oocyte's cytoplasm, and the ratio between the area of ​​the cytoplasm and the total area of ​​the cytoplasm and the perivitelline space.

[0026] In one embodiment of the invention, determining the feature includes: for each individual image in the images, identifying bounding boxes based on the position of the portion of the tool in the individual image to obtain a plurality of bounding boxes; and calculating the inhalation depth based on a first bounding box having the maximum distance in a first direction among the plurality of bounding boxes.

[0027] The oocyte analysis system of this invention provides accurate, objective, automated, and time-efficient oocyte quality assessment, helping embryologists and clinicians make informed decisions in the in vitro fertilization process. Attached Figure Description

[0028] Figure 1 is a block diagram of an example oocyte analysis system including a feature preprocessing engine and a machine learning model according to some embodiments of the present disclosure;

[0029] Figure 2A illustrates the morphological features of an example oocyte generated by the feature preprocessing engine in Figure 1;

[0030] Figure 2B illustrates the mechanical features of oocytes generated by the feature preprocessing engine in Figure 1;

[0031] Figure 3A illustrates a sample user interface including example processing results;

[0032] Figure 3B illustrates the example user interface of the example oocyte analysis system in Figure 1, which receives user input.

[0033] Figure 3C illustrates another example user interface that presents the processing results;

[0034] Figure 4 is a flowchart of an example process for determining indicators of oocyte quality;

[0035] Figure 5 is a flowchart of an example process for selecting a subset of majority oocytes based on indicators of quality.

[0036] Figure 6 illustrates the general architecture of an example oocyte analysis system according to some embodiments of this disclosure;

[0037] Figure 7A illustrates an example implementation of the example oocyte analysis system of Figure 1 for generating oocyte grades according to some embodiments of this disclosure;

[0038] Figures 7B, 7C, 7D, 7E, and 7F illustrate example features generated by an example oocyte analysis system using a segmentation model according to some embodiments of this disclosure;

[0039] Figure 8 is a flowchart of an example process for generating oocyte grades according to some embodiments of the present disclosure. Detailed Implementation

[0040] This specification describes a technique for selecting oocytes (e.g., oocytes) of good quality, such as those with a higher probability of blastocyst formation. As described, the system can utilize machine learning techniques to analyze oocytes using different sensor information. Sensor information may include, for example, images of oocytes (unfertilized eggs) undergoing deformation due to applied pressure, and pressure measurements indicating the pressure applied to the oocytes. This sensor information may allow for understanding the morphological and mechanical properties of the oocytes during deformation, which allows for a more accurate assessment of the quality of each oocyte. In contrast, existing techniques rely on manually tuned models or scoring systems that use a single type of information (e.g., images) and are prone to errors. As described, the disclosed technique utilizes different sensor information, along with specially trained models, to more accurately and effectively assess oocyte quality or viability at an early stage.

[0041] To assess oocyte viability, some existing techniques rely on embryologists visually evaluating embryos (e.g., fertilized eggs). Some clinics record images of embryos, and embryologists may score them based on various scoring systems and their visual assessments. A major challenge in embryo selection is the high degree of labor, subjectivity, and variability among embryologists of varying skill levels and between scoring systems of varying performance. Specifically, after spending considerable time visually evaluating embryos, embryologists often disagree with each other, and even with their own opinions on which embryos have the best chance of successful implantation. Furthermore, it remains unclear which embryo characteristics associated with a particular scoring system ultimately predict the success rate of each embryo.

[0042] Other existing technologies include automated techniques for selecting high-quality oocytes. These technologies may rely on specific characteristics of the fertilized oocyte. Typically, these specific characteristics are obtained by analyzing (e.g., using microscopy and computer vision techniques) videos that capture changes in the oocyte during its growth and development. However, selecting oocytes based on these characteristics (e.g., characteristics derived from observing the growth and development of the oocyte) may not yield satisfactory results.

[0043] In contrast, the revealed technology allows for the analysis of oocyte viability at an earlier stage (e.g., when the oocyte is still unfertilized rather than when it is already fertilized). Therefore, the revealed technology eliminates the additional complexity associated with the possibility that the fertilized oocyte may not subsequently be implanted. By utilizing machine learning models based on oocyte-derived features for oocyte selection, the revealed technology provides a more objective and quantitative analysis of oocyte viability.

[0044] Furthermore, the disclosed techniques utilize machine learning to analyze the morphological and mechanical characteristics of oocytes. As will be described, morphological characteristics may include geometric information associated with the oocyte, such as the size or length of the zona pellucida, cytoplasm, polar body, perivitelline space, and the degree to which the oocyte is aspirated by the pressure-applying tool (e.g., aspiration depth). As will be described, mechanical characteristics may include parameters determined or derived based on the oocyte's deformability properties. For example, mechanical characteristics may include at least morphodynamic parameters or similar parameters described in a specific model (e.g., the Zener model).

[0045] These machine learning techniques can utilize features to indicate one or more metrics representing oocyte quality. Example metrics may include values ​​indicating the likelihood or probability of blastocyst formation. Additional example metrics may include indicators of an oocyte being "good," the likelihood or probability of aneuploidy, and implantation rate, etc.

[0046] More specifically, the system can acquire image sequences depicting the deformation of an oocyte due to mechanical stimulation. Example stimulation may include a portion of the oocyte being aspirated into a pressure tool (e.g., a micropipette) that applies pressure to the oocyte. The system can process the image sequences using example computer vision techniques to derive the aforementioned morphological and mechanical features. In some examples, regularization techniques may be utilized to handle the different resolutions of the hardware used to capture the images (e.g., a microscope camera). Regularization techniques may be integrated as part of image processing techniques performed on the captured image sequences associated with the oocyte. Regularization techniques will be described in more detail below.

[0047] Morphological and mechanical features extracted from oocytes can be used to train or infer using machine learning models. The training process may include training the machine learning model using a subset of morphological and mechanical features. The trained machine learning model can then be used to generate one or more indicators of oocyte quality, where specific indicators may indicate blastocyst formation.

[0048] Furthermore, optionally, the indicators can be presented to users or professionals (e.g., embryologists) through an interactive user interface. This allows for more efficient further analysis or evaluation of oocyte viability. Therefore, based on the embodiments disclosed herein, a more objective, automated, and time-efficient assessment of oocyte viability can be achieved.

[0049] As an example of accuracy related to the revealing technique, a total of 185 oocytes were evaluated to assess their blastocyst formation capability. The 185 samples were divided into 80% for training the machine learning model and 20% for testing it. The machine learning used for viability assessment was a support vector machine (SVM). The machine learning model was trained to predict whether a specific oocyte in the sample would form a blastocyst or not. Statistical analysis of the prediction results showed an accuracy of 73%, a sensitivity of 85%, a specificity of 59%, a positive predictive value (PPV) of 77%, and a negative predictive value (NPV) of 71%. Compared to statistics compiled based on embryologist predictions, which showed an accuracy of 45%, a sensitivity of 59%, a specificity of 33%, a PPV of 43%, and an NPV of 49%, the system and method disclosed herein achieved statistically better performance.

[0050] The foregoing aspects and numerous incidental advantages of this disclosure will be more readily understood, as they become more readily apparent, when taken in conjunction with the accompanying drawings and the following description.

[0051] Example block diagram

[0052] Figure 1 is a block diagram of an example oocyte analysis system 100 according to some embodiments of the present disclosure, which includes a feature preprocessing engine 120 and a machine learning model 130. As shown, the feature preprocessing engine 120 may receive an image sequence 102 of an oocyte (e.g., a mammalian oocyte, such as a human oocyte), stress values ​​106, and clinical information 108 as input. The feature preprocessing engine 120 may then produce oocyte-related features 122 as output. Features 122 may be input to the machine learning model 130, which outputs oocyte quality information 132. The oocyte quality information 132 may be presented to a user for further analysis.

[0053] Image sequence 102 may include multiple images (e.g., images 104A to 104N), each depicting an oocyte 110 and a pressure tool 112. More specifically, images 104A to 104N may form a video depicting the process of aspirating the oocyte 110 into the pressure tool 112. For example, image 104A may represent the first frame of the video, while image 104N may represent the last frame. Image 104A depicts the oocyte 110 before it has been aspirated into the pressure tool 112, while image 104N depicts the oocyte 110 at least partially aspirated into the pressure tool 112. In some examples, image sequence 102 may have a frame rate of 10Hz, 20Hz, 70Hz, 3000Hz, and a total video length of 1 second, 2 seconds, 10 seconds, etc.

[0054] Furthermore, in some embodiments, the pressure tool 112 is a micropipette with a diameter between specific thresholds (e.g., between 10 micrometers (μm), 20 μm, 40 μm and 60 μm, 70 μm, 100 μm, etc.). The micropipette can apply negative pressure to the oocyte 110 (e.g., the pressure inside the pipette is lower than the pressure outside the pipette) to aspirate the oocyte 110 into the micropipette without damaging the oocyte 110. Example pressures may include between -0.01 psi and -0.5 psi. For example, the micropipette may be in close contact with or otherwise contact the oocyte. While image sequence 102 illustrates the aspiration of the oocyte 110 into the pressure tool 112, in some embodiments, the pressure tool 112 may apply other forms of mechanical stimulation (e.g., positive pressure). In this way, different morphological responses of the oocyte 110 can be obtained. The morphological responses can be used by the feature preprocessing engine 120 to analyze and / or extract different morphological features.

[0055] In Figure 1, the feature preprocessing engine 120 is illustrated as using image sequence 102, stress value 106, and clinical information 108 to output features 122 associated with oocyte 110. In some embodiments, all of this information may be used to determine feature 122. In some embodiments, a subset of the information may be used.

[0056] Regarding the pressure value 106, the pressure value 106 may include multiple values ​​that indicate how much force was applied to the oocyte 110 within the time range or time period from images 104A to 104N. For example, the pressure value 106 may indicate that the first pressure (e.g., -0.3 psi) was applied to the oocyte 110 at the moment image 104A was captured (e.g., time or timestamp), and the Nth pressure was applied to the oocyte 110 at the moment image 104N was captured. The pressure value applied to the oocyte 110 may be the same or different in different images of image sequence 102.

[0057] In some cases, the pressure applied to the oocyte 110 may increase over time, while in other cases, the applied pressure may decrease over time. Furthermore, the pressure value 106 may include the forces applied to the oocyte 110, which can be calculated based on the pressure generated and applied to the oocyte 110 by the pressure tool 112. The forces applied to the oocyte 110 can be used to derive the mechanical characteristics of the oocyte 110, which will be described in more detail later in the section concerning Figure 2B.

[0058] Clinical information 108 may include the age and body mass index (BMI) of the patient from whom the oocytes 110 were derived. Additionally, clinical information 108 may include information indicating whether the oocytes 110 have been cryopreserved (CP). Clinical information 108 may also indicate the number of available mature oocytes (MII) relevant to the patient.

[0059] Based on at least some image sequences 102, stress values ​​106, and clinical information 108, the feature preprocessing engine 120 can extract features 122 associated with the oocyte 110. Features 122 may include morphological features of the oocyte 110 (size or length of the zona pellucida, cytoplasm, polar body, or perivitelline) and mechanical features (e.g., elasticity and / or viscosity). Some features may be generated for each image in image sequence 102, while others may be determined based on all images or a subset of images. These features 122 will be described in more detail in the sections relating to Figures 2A-2B below.

[0060] Based on feature 122, machine learning model 130 can generate oocyte quality information 132 for oocyte 110. Examples of information 132 may include the probability of blastocyst formation. In some embodiments, machine learning model 130 may be a support vector machine (SVM) trained to output information 132. In other embodiments, machine learning model 130 may be a deep learning model. For example, a deep learning model may include a recurrent neural network (RNN) trained to output information 132. In this example, the RNN may take feature 122 as a sequence input and output information 132 for that sequence. The model may also be a convolutional neural network or a fully connected network. Machine learning model 130 may use all or a subset of feature 122 to determine oocyte quality information 132. For example, an SVM may be trained to utilize a subset of feature 122.

[0061] In some embodiments, oocyte quality information 132 may further indicate the likelihood that oocyte 110 will form a “usable” blastocyst, where “usable” means that the blastocyst formed from oocyte 110 is suitable for transfer or implantation. Alternatively, oocyte quality information 132 may indicate the likelihood that the blastocyst formed from oocyte 110 will become an “unusable” blastocyst, where “unusable” means that the blastocyst formed from oocyte 110 will be in such a poor condition that it is unsuitable for further in vitro fertilization (IVF) treatment.

[0062] Block diagram - morphological feature generation

[0063] Figure 2A illustrates example morphological features of an oocyte generated by the feature preprocessing engine 120. Example morphological features 210 may form part of feature 122 in Figure 1. Example morphological features 210 may include the aspiration depth of the oocyte 110, the diameter of the pressure tool 112 (e.g., a micropipette), the size and / or shape of the cytoplasm and zona pellucida of the oocyte 110, etc. Optionally, these features may be determined for each input image.

[0064] The determination of some example morphological features 210 of oocyte 110 can be illustrated by oocyte measurements 202N, which depict example measurements (e.g., bounding box A, distance B, distance C, distance D, distance E, bounding box F, and distance G). Oocyte measurements 202N correspond to the measurements in image 104N, and oocyte measurements 202A correspond to the measurements in image 104A.

[0065] Regarding the oocyte measurement 202N described, bounding box A may represent a region of interest (ROI), which is cropped for further image processing of image sequence 102. Distance B may indicate the length and / or size of the upper side of the zona pellucida of oocyte 110. Distance C may indicate the length and / or size of the right side of the zona pellucida of oocyte 110. Distance D may indicate the length and / or size of the lower side of the zona pellucida of oocyte 110. Distance E may indicate the inner zona pellucida of oocyte 110. Distance G may indicate the inner diameter of pressure tool 112. Bounding box F may represent a region of interest (ROI), which may be cropped to calculate the aspiration depth of oocyte 110 entering pressure tool 112. The calculation of aspiration depth is described in more detail below.

[0066] Different image processing and object recognition techniques can be used to obtain example morphological features 210. In some examples, computer vision techniques can be used to derive example morphological features 210. For example, edge detection techniques can be used to identify boundaries associated with different parts of the oocyte. In other examples, image segmentation models based on deep learning algorithms can be used to derive example morphological features 210. For example, an image segmentation mask can be generated for feature 210. It can be understood that the image segmentation mask can specify a color or pixel value to indicate whether a pixel forms part of a classification feature (e.g., a luminosity band, etc.). The derivation of each example morphological feature will be discussed in more detail below.

[0067] Bounding box A may define a Region of Interest (ROI) that will be cropped for any image of image sequence 102 (e.g., image 104A or image 104N), where some morphological features 210, such as the inner boundary of the zona pellucida, the size of the pressure tool 112, and the aspiration depth of the oocyte 110, can be calculated within this ROI. In some examples, bounding box A may be identified according to the following example techniques. For example, the tip of the pressure tool 112 (e.g., a micropipette) contacting the oocyte 110 may be identified and / or located based on a potentially pre-acquired image of the micropipette tip. Based on the location of the micropipette tip, bounding box A may be drawn to delineate the ROI around the micropipette tip. Although bounding box A is shown as a rectangle, other shapes may be used to define the ROI.

[0068] Distances B, C, and D define the length or thickness of the zona pellucida on the upper, right, and lower sides, respectively. By averaging the lengths of these distances, morphological characteristics indicating the average size of the zona pellucida of oocyte 110 can be obtained. Furthermore, distance E can be defined based on the position of the micropipette tip and the inner boundary of the zona pellucida of oocyte 110 obtained when defining bounding box A.

[0069] As an example, distance E can be used to derive the horizontal (e.g., along the directions of arrows C and E) length of the cytoplasm of oocyte 110. More specifically, the horizontal length of the cytoplasm of oocyte 110 can be derived by subtracting the combined length of arrows E and C from the horizontal length of oocyte 110 (e.g., defined by the right end of arrow C and the left end of arrow E). The horizontal length of the cytoplasm of oocyte 110 can also be used as part of example morphological feature 210. Similarly, the vertical (e.g., along the directions of arrows B and D) length of the cytoplasm of oocyte 110 can be derived by subtracting the combined length of arrows B and D from the vertical length of oocyte 110 (e.g., defined by the apex of arrow B and the base of arrow D). The vertical length of the cytoplasm of oocyte 110 can also be used as part of feature 122.

[0070] The bounding box F can be used as the ROI for calculating the aspiration depth of the oocyte 110 (e.g., how much the oocyte 110 is aspirated by the pressure tool 112 compared to its position immediately after aspiration). The aspiration depth can be useful for deriving mechanical features associated with the oocyte 110, which is discussed in more detail below regarding Figure 2B. In some embodiments, the aspiration depth can be derived by determining the deepest horizontal position of the oocyte 110 aspirated by the pressure tool 112 (e.g., the end of the aspiration depth). The distance between the end of the aspiration depth and the inner boundary of the zona pellucida of the oocyte 110 can then be calculated. The following are example techniques for determining the aspiration depth.

[0071] The ROI (e.g., bounding box F) can be defined (e.g., identified) based on the position of the tip of the pressure tool 112 in contact with the oocyte 110. As shown in the oocyte measurement 202N, the bounding box F has a rectangular shape; however, in other examples, the bounding box F may be of a different shape (e.g., a square or an irregular shape). In some embodiments, the bounding box F extends to the leftmost portion of the image by a specific threshold distance. That is, this distance can be set to ensure that it includes the distal or extreme portion of the oocyte being drawn into the tool 112.

[0072] Vertically aligned pixels within the bounding box F can be summed to derive a pixel curve. More specifically, assuming the bounding box F spans N pixels horizontally and M pixels vertically, the pixel intensities of pixels at the same horizontal position within the bounding box F can be summed. For example, M pixels can be summed at a specific horizontal position. Therefore, N can be a positive integer depending on the resolution of the image sequence 102 and the horizontal length of the bounding box F. For example, if there are 50 horizontal positions, then for each position, the pixels extending vertically within the box F can be summed to obtain a value.

[0073] These values ​​can then be used to generate a pixel curve that has an additive value on a first axis (e.g., the Y-axis) and a number of horizontal positions on a second axis (e.g., the X-axis). The pixel curve can be smoothed using signal processing techniques such as a sliding window moving average method.

[0074] The first derivative can then be determined on the smoothed pixel curve. This derivative represents the intensity of the vertically added forces. The minimum value of the pixel curve derivative can then be found, which indicates the end of the aspiration depth of the oocyte 110 in a particular image (e.g., image 104A).

[0075] The above-described example techniques can be performed on other images in image sequence 102 to obtain the aspiration depth endpoints for each image 104A to 104N. In some examples, the aspiration depth can be calculated based on the aspiration depth endpoints and the inner boundary of the zona pellucida of the oocyte 110. By calculating the aspiration depth for each image 104A to 104N in image sequence 102, the aspiration depth of the oocyte 110 over time can be plotted as a graph, where a first axis displays time and a second axis displays the aspiration depth of the oocyte 110. In some examples, the aspiration depth can be further normalized based on the zona pellucida size of the oocyte 110. For example, the aspiration depth can be normalized by subtracting the average of distances B, C, and D from the calculated aspiration depth.

[0076] While the foregoing describes an example of determining aspiration depth, it is understood that other techniques may also be used and fall within the scope of this disclosure. For example, deep learning techniques may be used to determine the boundaries or extreme portions of the oocyte entering the tool. As another example, edge detection techniques (e.g., edge detection kernels) may be used to identify the edges of the oocyte within the tool.

[0077] Alternatively, the feature preprocessing engine 120 may not need to analyze all images in the image sequence 102. In some examples, only a portion of the image sequence 102 is analyzed by selecting a starting frame and performing analysis on the starting frame and frames following it. The starting frame may be the frame in which an image of the oocyte 110 is captured just before it is aspirated by the pressure tool 112. For example, the system may identify frames associated with a time or timestamp preceding the time associated with the application of negative pressure. As another example, the system may analyze image frames and identify frames preceding a frame in which the aspiration depth is greater than a threshold (e.g., zero, a small value, etc.). Advantageously, the analysis can be more time-efficient and consume less computational power.

[0078] Alternatively, the preprocessing engine 120 can reduce (e.g., shorten) the number of frames to be analyzed by removing consecutive frames in which the oocyte 110 remains stationary or unmoved. For example, the preprocessing engine 120 may determine that only 0.5 seconds of the images depicted in image sequence 102 show movement of the oocyte 110, while the remaining 1.5 seconds of the images in image sequence 102 show the oocyte 110 remaining stationary or relatively still. The preprocessing engine 120 may then extract 0.5 seconds of the image for further analysis. Advantageously, the analysis time can be reduced by shortening the length of the images used for analysis.

[0079] The feature preprocessing engine 120 can also additionally determine the inner diameter of the pressure tool 112, which is represented by the distance G in the oocyte measurement 202N. In some embodiments, the length of the inner diameter of the pressure tool 112 can be obtained by calculating the number of pixels vertically aligned along the inner diameter of the pressure tool 112. The inner diameter of the pressure tool 112 can be used as a normalization factor to increase the interoperability of the oocyte analysis system 100 between different image capture platforms that may have different hardware specifications (e.g., the resolution of the captured images).

[0080] Regularization may also be related to the Z-axis height of the oocyte within the object containing the oocyte during image capture. That is, the oocyte may have been placed in water or other liquid, with variations in height (e.g., closer to the bottom or surface). Therefore, this adjustment in height may alter the oocyte's size compared to another oocyte or the same oocyte in different images of sequence 102. For example, the oocyte may have experienced slight height variations during image sequence 102.

[0081] More specifically, the oocyte aspiration depth over time can be normalized by the ratio between the inner diameter of the pressure tool 112 and the pixel length of a specific image sequence. Using computer vision techniques, the inner diameter of the pressure tool 112 can be identified, and then the length of the inner diameter can be calculated by counting the number of pixels along the axis covered by the inner diameter of the pressure tool 112. For example, the inner diameter of the pressure tool 112 may span 1,000 pixels, and the aspiration depth can then be normalized by dividing by 1,000. Normalizing the oocyte aspiration depth by the inner diameter of the pressure tool 112 allows different frames of oocytes acquired under different image capture settings to be placed on the same basis to calculate morphological features associated with the oocyte. Therefore, the interoperability and reproducibility of the system and method disclosed herein can be enhanced.

[0082] The feature preprocessing engine 120 can also acquire additional morphological features associated with the oocyte 110. Example features may include the size and / or length of the oocyte 110's polar body, cytoplasm, and / or perivitelline space (e.g., the space between the zona pellucida and cytoplasm, the PVS). Image segmentation models based on deep learning algorithms and computer vision techniques can both be used to determine these additional morphological features. Advantageously, the acquired additional morphological features may be potentially useful for training and / or testing the machine learning model 130 to improve the accuracy, sensitivity, specificity, NPV, and PPV of the results produced by the machine learning model 130.

[0083] Block diagram - mechanical feature generation

[0084] Figure 2B illustrates the oocyte mechanical features generated by the feature preprocessing engine 120 of Figure 1. As shown in Figure 2B, the feature preprocessing engine 120 receives image sequence 102 and pressure value 106 as input and generates example mechanical features 254 based on example elasticity model 252. Example mechanical features 254 can indicate the deformation or movement of oocyte 110 during image sequence 102.

[0085] The pressure value 106 may include the force applied to the oocyte 110 (e.g., by the pressure tool 112 described above). Specifically, the force applied to the oocyte 110 may be calculated based on the following equation (A). Therefore, these forces may be based on the pressure value and geometric information associated with the micropipette.

[0086] The feature preprocessing engine 120 can use the aforementioned force, optionally combined with the inhalation depth described in Figure 2A, to determine feature 254.

[0087] In some embodiments, a linear elastic model, a modified linear elastic model, or a standard linear solid model can be used to obtain features indicative of the viscoelastic behavior of the oocyte 110. Example models may include the Zener model, a modified Zener model, etc. For example, the mechanical characteristics of the oocyte 110 can be obtained by fitting the aspiration depth of the oocyte 110 over time using the following equation (B). For example, as shown in Figure 2B, the mechanical characteristics can be simulated using the model described above (e.g., a spring-damper model) with two springs and two dampers (e.g., shock absorbers). This model can be fitted using example techniques (e.g., as an example, the Broyden-Fletcher-Goldfarb-Shanno algorithm) by minimizing the sum of squared errors between the measured aspiration depth and the simulated aspiration depth. Equation (B) includes parameters k0, k1, τ, η0, and η1 of the example elastic model 252. These parameters can be determined by fitting the aspiration depth. In equation (B), F0 can indicate the force applied to the oocyte 110. For example, F0 can indicate the force associated with a specific aspiration depth (hereinafter referred to as "depth"). In this example, different aspiration depths (e.g., images) associated with different times, along with their respective forces, can be used to determine the parameters. Parameters k0 and k1 simulate the solid-like behavior of oocyte 110, while η0 and η1 simulate the fluid-like behavior of oocyte 110. Therefore, the example elasticity model 252 can take into account both behaviors associated with oocyte 110.

[0088] Parameters k0 and k1 describe the “instantaneous elongation” experienced by oocyte 110 when a force is applied to it. This instantaneous elongation corresponds to or is proportional to 1 / (k0+k1) and can be considered a measure of the “relaxation” in the elastic element of oocyte 110, or the force that can be applied to oocyte 110 before significant resistance is encountered. Parameter k1 can be considered a general measure of stiffness, possibly representing the tightness of protein bonds in the cytoplasm or zona pellucida. Parameter η1 can be considered a measure of the extent to which the zona pellucida continues to deform according to the applied force calculated according to Equation (A). As in a linear elastic solid model, η1 is responsible for the shape change at the molecular level after the spring element is fully extended, causing oocyte 110 to continue elongating. Parameter τ represents how quickly (e.g., velocity) oocyte 110 deforms (e.g., enters pressure tool 112) after the initial instantaneous elongation. η0 can be considered as a measure of the viscosity of the fluid in the space between the zona pellucida and the inner cell of the cytoplasm or oocyte 110 (e.g., PVS).

[0089] In addition to using negative pressure, other non-limiting examples of causing movement or deformation of the oocyte 110 may be used and fall within the scope of this disclosure. For example, positive pressure may be applied to the oocyte 110 to eject it or otherwise deform, fix, or disturb different parts of the oocyte 110. For example, parts may include the zona pellucida, cytoplasm, or portions near the surface of the oocyte 110. Other forms of force (e.g., optical pressure) may also be applied to the oocyte 110.

[0090] In some instances, the pressure or force applied to the oocyte 110 by the pressure tool 112 is appropriately adjusted to avoid causing undue damage to the oocyte 110. For example, excessive pressure applied to the oocyte 110 may damage its structure and reduce its viability. In some embodiments, the pressure applied to the oocyte 110 is between -0.01 psi and -0.5 psi (0.01 psi to 0.5 psi if positive pressure is applied). In some aspects, the pressure applied to the oocyte 110 by the pressure tool 112 is adjusted according to the number of days elapsed since fertilization (e.g., 1, 2, or 3 days). In some embodiments, the inner diameter of the pressure tool 112 is between 40 μm and 70 μm, and the applied pressure can be adjusted according to the inner diameter of the pressure tool 112 to produce an appropriate level of force applied to the oocyte 110.

[0091] Example User Interface

[0092] Referring to Figures 3A to 3C, an illustrative application of the oocyte analysis system 100 in Figure 1 will be described.

[0093] Figure 3A illustrates an example user interface 300A that includes the processing results of the example oocyte analysis system 100 in Figure 1. As shown in Figure 3A, the oocyte quality information 132 generated by the oocyte analysis system 100 can be presented through the user interface 300A. The user interface 300A can be presented as a webpage or application user interface accessible by a browser.

[0094] As shown in the figure, user interface 300A receives oocyte quality information 132 from system 100 described herein. Information 132 may reflect individual quality indicators of multiple oocytes extracted from the patient. In the example shown, user interface 300A presents summary information based on the analysis of five oocytes. User interface 300A indicates that “oocyte 3” has the highest blastocyst formation probability. This indication may be based on the blastocyst formation probability determined by a machine learning model (e.g., model 130).

[0095] User interface 300A can respond to user input related to viewing detailed analysis. In some embodiments, detailed analysis may include all or a subset of the features described above. Detailed analysis may include a graphical representation related to oocyte 3, such as an image from an image sequence. In some embodiments, the image sequence may be presented in user interface 300A as a video or animation.

[0096] In some instances, oocyte quality information 132 can be linked to or compared with assessments obtained from preimplantation genetic testing (PGT) or implantation on the same oocyte sample. For example, for oocytes whose quality information 132 indicates they can form blastocysts (e.g., “good” oocytes), preimplantation chromosomal screening (PGT-A) can be performed to determine the euploidy rate of the oocytes selected by the oocyte analysis system 100. As another example, for oocytes whose quality information 132 indicates they cannot form blastocysts (e.g., “bad” oocytes), preimplantation chromosomal screening (PGT-A) can also be performed to determine the aneuploidy rate of the oocytes selected by the oocyte analysis system 100.

[0097] Another example is that oocyte quality information 132 indicates that oocytes capable of forming blastocysts (e.g., “good” oocytes, such as those with a probability above a threshold) can be further evaluated after implantation. This can, for example, be used to measure the predictive ability of the oocyte analysis system 100 for oocyte viability. The predictability of the oocyte analysis system 100 can be assessed based on the probability of embryo implantation from “good” oocytes indicated by the oocyte analysis system 100. Advantageously, the oocyte analysis system 100 can be improved based on PGT-A and implantation assessment results by connecting different stages of embryo quality assessment. For example, a lower embryo implantation probability might indicate a need to tune the parameters of the machine learning model 130 by using different subsets of features 122 to train the machine learning model 130.

[0098] Figure 3B illustrates an example user interface 300B that allows the example oocyte analysis system 100 of Figure 1 to receive user input, such as patient information. As shown in Figure 3B, user interface 300B allows users to add, edit, and store patient information into the oocyte analysis system 100. As shown, the patient information displayed on user interface 300B is assigned a patient ID (e.g., 1234567890). The patient ID can be used to associate a specific oocyte with a specific patient. Patient information such as height, weight, and the date the patient information was created is displayed below the patient ID. Additional patient information can be added by pressing the "+Add" button in the upper part of user interface 300B.

[0099] Although not illustrated in Figure 3B, the user interface 300B can facilitate other interactions with the example oocyte analysis system 100. For example, by operating the user interface 300B, other patient information such as BMI and / or age can be edited and associated with a specific patient. The oocyte analysis system 100 can use the added patient information to analyze the quality of one or more specific oocytes. Furthermore, the user interface 300B can facilitate the search for specific patient information based on patient ID or other recorded patient information such as MII. Moreover, the user interface 300B can allow users to edit information associated with a specific oocyte, such as the date the oocyte was received and information about the oocyte provider (e.g., from whom the specific oocyte was extracted), such as the provider's date of birth, height, and / or weight.

[0100] User interface 300B can also receive user input that enables the example oocyte analysis system 100 to analyze a specific oocyte to generate oocyte quality information 132 about that specific oocyte. Furthermore, user interface 300B can prompt the user to check or recalibrate the camera position used to capture image sequences 102 for analyzing the quality of one or more oocytes. In some embodiments, the oocyte quality information 132 generated by the example oocyte analysis system 100 may be presented to the user as described below.

[0101] Figure 3C illustrates an example user interface 300C that presents the processing results of the example oocyte analysis system 100 of Figure 1. As shown in Figure 3C, the user interface 300C can present different oocyte quality information 132 based on the analysis performed by the oocyte analysis system 100. As shown in Figure 3C, the oocyte quality information 132 is presented in fractional form to indicate the quality of the oocytes. In addition, a textual description summarizing the quality can be presented. For example, the text can be based on quality indicators and can be selected from pre-stored textual expressions or words (e.g., normal, good, etc.) or determined using a language model (e.g., a large language model).

[0102] The left side shows the analysis result of an oocyte with a good score (e.g., 95), which likely means that the oocyte is very likely to form a "usable" blastocyst. Conversely, the right side shows the analysis result of an oocyte with a poor score (e.g., 14), which likely means that the oocyte is unlikely to form a "usable" blastocyst. In the middle, the user interface 300C shows the analysis result of an oocyte with a "normal" score (e.g., 60), which likely means that the oocyte is more likely to form a usable blastocyst than not.

[0103] After receiving the analysis results of a specific oocyte, the user interface 300C allows the user to view the analysis results of other oocytes or provides suggestions on the quality of oocytes not yet analyzed by the example oocyte analysis system 100. Specifically, the user can view the analysis results of another oocyte by pressing the "Next oocyte" button, or view the analysis results of the currently analyzed oocytes by pressing the "Back" button.

[0104] Example Flowchart

[0105] Figure 4 is a flowchart of an example process 400 for determining indicators of oocyte quality. All or at least part of process 400 can be implemented, for example, by the oocyte analysis system 100 of Figure 1. As described above, process 400 can be used to determine the viability of oocytes 110 (e.g., whether they will form blastocysts) without relying on manual and subjective assessment by an embryologist. Therefore, process 400 can be used to achieve a more objective, time-saving, and automated acquisition of oocyte quality.

[0106] In block 402, the system acquires images that form a sequence of oocyte images. As described above, these images may be captured by a microscope camera and depict a series of events resulting from the deformation or movement of the oocyte caused by the application of force to the oocyte by a tool (e.g., a micropipette).

[0107] At block 404, the system acquires a pressure value associated with the oocyte being aspirated by the aspiration tool. For example, the pressure value may include the pressure or force applied to the oocyte 110 during aspiration by the pressure tool. In some embodiments, the pressure value may remain constant throughout the aspiration process. In some embodiments, the pressure value may vary in a certain way (e.g., applying a lower pressure first, then a higher pressure).

[0108] In block 406, the system determines oocyte-related morphological features based on the acquired image sequence. Illustratively, oocyte-related morphological features include the oocyte aspiration depth, cytoplasmic size and / or length, zona pellucida size and / or length, and the diameter of the tool used to aspirate the oocyte.

[0109] As discussed in Figure 2A, morphological features can be acquired using computer vision and / or machine learning techniques. In some embodiments, one or more neural networks can be trained to output image segmentation masks specific to certain morphological features. In this way, the size or length associated with a portion of an oocyte can be identified. In some embodiments, image segmentation may include two stages: two-dimensional (2D) modeling followed by three-dimensional (3D) modeling. In the 2D modeling stage, features associated with the 2D images are extracted. In the 3D modeling stage, the features extracted from the 2D images can be concatenated to form time-series data, where each time-series data includes features extracted from one 2D image in the sequence of 2D images. A machine learning model (e.g., a deep learning model) can then classify the features generated in the 3D modeling stage, as shown in, for example, in block 410.

[0110] In block 408, the system determines the mechanical features associated with the oocyte. For example, the system determines parameters that indicate the deformation or movement of the oocyte 110 during the image sequence. In this example, these parameters may be related to the elasticity model described in Figure 2B.

[0111] In block 410, the system uses a machine learning model to determine indicators of oocyte quality. The system provides features, such as connectivity features determined for an image or feature sequences determined for individual images, as input to the machine learning model. In some embodiments, the machine learning model 130 may be a support vector machine. In some embodiments, the model may be a deep learning model (e.g., a neural network). In some embodiments, a subset of features may be provided. For example, one feature, two features, three features, ten features, etc., may be provided.

[0112] These indicators may include information that at least indicates oocyte blastocyst formation. In addition, one or more indicators may indicate aneuploidy and / or implantation associated with the oocyte.

[0113] In some embodiments, one or more indicators may indicate whether oocyte 110 will form a “good” blastocyst, where a “good” blastocyst may mean an associated Gardner embryo / blastocyst rating greater than 3CC. Furthermore, in addition to using example morphological and mechanical features to determine indicators of oocyte quality, the machine learning model may further use patient clinical information to determine oocyte quality information. As mentioned in the discussion of Figure 1, clinical information may include age, patient BMI, and / or other clinical information such as oocyte-related CP and MII (e.g., oocyte-related developmental stage).

[0114] After determining the indicators of oocyte quality, process 400 can return to block 402 to determine the quality information of another oocyte.

[0115] Figure 5 is a flowchart of an example process 500 for selecting a subset of multiple oocytes based on indicators of quality. All or at least some parts of process 500 can be implemented by the oocyte analysis system 100 of Figure 1. Process 500 provides the ability to determine viability in multiple oocytes without relying on time-consuming and subjective embryologist assessments. Therefore, process 500 can be used to achieve a more objective, time-saving, and automated quality assessment of multiple oocytes.

[0116] In block 502, the system acquires image sequences and stress values ​​associated with multiple oocytes. As described herein, the image sequences depict oocytes deformed due to the application of a stress tool or pressure.

[0117] In block 504, the system acquires indicators that describe the quality of individual oocytes from a plurality of oocytes. These indicators, described in more detail above, indicate the likelihood of each oocyte developing into a blastocyst.

[0118] In block 506, the system selects a subset from multiple oocytes. For example, the system may identify the oocyte with the highest threshold number based on their respective probabilities of forming blastocysts. Another example is that the system may aggregate or otherwise combine indicators for each oocyte. In this example, the system may select the oocyte with the highest threshold number based on the aggregated or combined indicators. As mentioned above, these indicators may indicate successful blastocyst formation and later stages, or indicators related to chromosomal abnormalities (e.g., PGT-A, euploidy), etc.

[0119] In block 508, the system outputs and / or presents information related to the selected subset of oocytes. This information can be presented through a graphical user interface (GUI), such as user interface 300A shown in Figure 3A. As shown in Figures 3A and 3C, this information can indicate whether a particular oocyte is likely to develop into a blastocyst. Furthermore, other information related to the selected subset of oocytes can also be presented to the user. For example, morphological characteristics, mechanical features, and / or clinical information of the patient who acquired the oocytes can be presented. By implementing process 500 using the oocyte analysis system 100, viability information related to multiple oocytes can be obtained in a time-saving manner.

[0120] Example System

[0121] Figure 6 illustrates the general architecture of the example system. In some embodiments, the system can be used to perform the functions described herein. In some embodiments, the system may be an oocyte analysis system 100, which includes an arrangement of computer hardware and software configured to implement aspects of this disclosure. The oocyte analysis system 100 may include more (or fewer) components than shown in Figure 6. However, not all of these components need to be shown to provide an implementable disclosure.

[0122] As shown, the oocyte analysis system 100 includes a processor 602, a pressure tool 604 (e.g., pressure tool 112 of FIG. 1), a network interface 606, an image sensor 608 (e.g., one or more microscope cameras for capturing image sequences 102 of FIG. 1), and a data storage 610, all of which can communicate with each other via a communication bus 612. In some embodiments, the pressure tool 112 may not be included, and the system 100 may represent a back-end processing system. The network interface 606 provides connectivity to one or more networks or computing systems, thus enabling the oocyte analysis system 100 to receive and send messages and instructions from other computing systems, interfaces (e.g., user interface 300A of FIG. 3A), or services. In some embodiments, the oocyte analysis system 100 may be configured to process requests from other devices or modules, such as requests to analyze oocyte quality. The data storage 610 may be any non-transitory computer-readable data storage and, in various embodiments, may store any or all elements of the loaded memory 614 depicted in FIG. 6.

[0123] Processor 602 may also communicate with memory 614. Memory 614 may contain computer program instructions (grouped into modules or components in some embodiments) that are executable by processor 602 to implement one or more embodiments. Memory 614 typically includes RAM, ROM, and / or other persistent, auxiliary, or non-transitory computer-readable media. Memory 614 may store operating system 616, which provides processor 602 with computer program instructions for the general management and operation of oocyte analysis system 100. Memory 614 may further store specific computer-executable instructions and other messages (which may be referred to herein as “modules” or “engines”) for implementing aspects of this disclosure. For example, memory 614 may include feature preprocessing engine 632 and machine learning model 634, which implement the various aspects of this disclosure described above. Memory 614 may further store, for example, user interface module 618, which enables information to be presented to user interface 300A of FIG. 3A. In addition, memory 614 may store database 620 (e.g., for storing parameters of different types of machine learning models) and features 630 that can be extracted by feature preprocessing engine 632. As various operations are executed, all modules or elements loaded into memory 614 can also be stored in data storage 610.

[0124] It should be recognized that many of the components depicted in Figure 6 are optional, and embodiments of the oocyte analysis system 100 may incorporate or exclude these components. Furthermore, the components need not be unique or separate. Components may also be reorganized. In some embodiments, components that are part of the oocyte analysis system 100 may be additionally or alternatively included in other computing devices, such that some aspects of this disclosure can be performed by the oocyte analysis system 100, while others are performed by another computing device.

[0125] Example of oocyte grade generation

[0126] As discussed above with respect to Figures 1 through 6, the oocyte analysis system 100 can utilize machine learning techniques to analyze the morphological and mechanical characteristics of oocytes to generate one or more indicators indicating oocyte quality. For example, the indicators may include values ​​indicating the likelihood or probability of blastocyst formation.

[0127] As will be discussed below, the oocyte analysis system 100 may additionally and / or selectively utilize segmentation models to identify oocyte-related objects (e.g., different parts of the oocyte and / or geometric information associated with different parts) and determine features associated with the identified objects. Based on these features, the oocyte analysis system 100 may utilize machine learning models (e.g., regression models), or other statistical or artificial intelligence techniques to generate oocyte grades. Oocyte grades indicate the likelihood of an oocyte developing into a usable blastocyst. Based on oocyte grades, the oocyte analysis system 100 can provide accurate, objective, automated, and time-efficient oocyte quality assessments, helping embryologists and clinicians make informed decisions in the in-vitro fertilization process.

[0128] Figure 7A illustrates an example implementation of the oocyte analysis system 100 for generating oocyte grades, based on some embodiments of this disclosure. Unless otherwise stated, the blocks, inputs, and outputs of Figure 7A may be the same or substantially similar to the blocks, inputs, and outputs with the same numbering in Figure 1. As shown in Figure 7A, the feature preprocessing engine 120 may receive an image sequence 102 of an oocyte (e.g., oocyte 110). Based on the image sequence 102, the segmentation model 740 may identify objects 760 associated with the oocyte. Based on the objects 760, the feature extractor 750 may determine or extract features 122 associated with the objects 760 identified by the segmentation model 740. Although not illustrated in Figure 7A, the feature preprocessing engine 120 may additionally and / or selectively utilize stress values ​​106 and clinical information 108 as inputs for generating features 122. Based on at least a subset of feature 122, machine learning model 130 can generate oocyte quality information 132 containing at least an oocyte grade 780, which indicates the likelihood of the oocyte developing into a usable blastocyst. As shown, oocyte grade 780 can be one of grade A, grade B, grade C, or indeterminate (INC), which will be explained in detail below. Oocyte grade 780 can be presented to the user through an interactive user interface for further analysis.

[0129] As shown in Figure 7A, segmentation model 740 can process image sequences 102 of oocyte 110 (e.g., including images 104A to 104N) to identify objects 760 associated with oocyte 110 (e.g., objects representing different parts of the oocyte). More specifically, the objects 760 identified by segmentation model 740 may include bounding boxes (BBOX) 760A, zona pellucida (ZP) 760E, perivitelline space (PVS) 760D, first polar body (FPB) 760B, and cytoplasm (CPM) 760C. In some examples, segmentation model 740 may utilize machine learning algorithms or architectures, such as the U-Net architecture selectively using the MobileNet_v2 encoder backbone, to generate image segmentation masks for identifying BBOX 760A, FPB 760B, CPM 760C, PVS 760D, and ZP 760E. It can be understood that image segmentation masks can help isolate and identify different parts of the oocyte.

[0130] In some examples, the segmentation model 740 can be trained, tuned, and / or validated using machine learning techniques. Training the segmentation model 740 may include data selection, model training, and model validation to ensure that the segmentation model 740 can accurately segment and recognize BBOX 760A, FPB 760B, CPM 760C, PVS 760D, and ZP 760E. Data selection may include collecting and organizing a dataset of oocyte images that can be used for model training and validation. This dataset may include images from a variety of sources (e.g., human, bovine, and porcine oocytes) to ensure diversity and robustness. These images may be annotated with bounding boxes, first polar bodies, cytoplasm, perivitelline space, and zona pellucida. Raw and annotated images may be further separated using Git and Data Version Control (DVC) for better data control and management. The dataset may be divided into training and validation sets.

[0131] Once the dataset is ready, the segmentation model 740 can be trained. The model training process may involve selecting an appropriate model architecture (e.g., a convolutional neural network such as U-Net, a visual converter architecture, or similar), defining hyperparameters (e.g., encoder depth, decoder channels, batch size, initial learning rate, optimizer, scheduler, or similar), and using data augmentation techniques (e.g., mesh warping, optical warping, random cropping, translation, scaling, and rotation) to improve model performance. Furthermore, combinations of loss functions, such as multi-class focus loss and Dice loss, can be used to handle class imbalance and improve segmentation accuracy. Training progress can be monitored using metrics such as mean intersection and union ratio (mIOU) and F1 score.

[0132] After training segmentation model 740, it can be validated using a validation dataset. The validation process ensures that segmentation model 740 generalizes well to unseen data and accurately segments different parts of the oocyte. Based on the data selection and management, model training, and model validation processes described above, segmentation model 740 can accurately identify and segment various parts of the oocyte, achieving precise feature extraction and grading. This, in turn, helps to objectively determine the oocyte grade, increasing the chances of success in in vitro fertilization (IVF) treatment.

[0133] In some cases, ZP 760E may be the outer layer of the oocyte, protecting it and / or facilitating sperm union during fertilization. PVS 760D may be the space between ZP 760E and CPM 760C. PVS 760D may contain FPB 760B. FPB 760B may be a relatively small cell expelled from the oocyte during meiosis. The presence and morphology of FPB 760B can provide insights into the oocyte's developmental potential. CPM 760C may contain various organelles. CPM 760C may be crucial for the oocyte's metabolic activity and developmental capacity.

[0134] Based on some or all of the BBOX 760A, FPB 760B, CPM 760C, PVS 760D, and ZP 760E identified by the segmentation model 740, the feature extractor 750 can determine features 122 associated with the oocyte 110. More specifically, features 122 can be determined or calculated based on various combinations of measurements and / or geometric information associated with the object 760 (e.g., BBOX 760A, FPB 760B, CPM 760C, PVS 760D, and ZP 760E). Features 122 may include morphological features indicating measurements of the oocyte 110 and the depth of aspiration of the oocyte 110, some of which have been mentioned in the discussion with reference to FIG2A. In some examples, morphological features may be associated with the object 760 identified by the segmentation model 740. For example, morphological features may include at least one of the following: the ellipticity of FPB 760B, the thickness of ZP 760E, the diameter of oocyte 110 (e.g., diameter 702E as described with reference to FIG7E), the area of ​​CPM 760C, the compactness of CPM 760C, the roundness of CPM 760C, and the ratio between the area of ​​CPM 760C and the total area of ​​CPM 760C and PVS 760D (e.g., coverage 706C as described with reference to FIG7C).

[0135] Based on feature 122, machine learning model 130 can generate oocyte quality information 132 containing at least oocyte grade 780. In some examples, machine learning model 130 can be a regression model. The regression model can include multiple weights. In some examples, the multiple weights can be iteratively adjusted through a training process associated with the regression model. Each of the multiple weights can be associated with one of the features 122 that determine oocyte grade 780. For example, a first weight can be used to multiply by a first feature (e.g., the ellipticity of FPB 760B) to produce a first product, a second weight can be used to multiply by a second feature (e.g., the area of ​​CPM 760C) to produce a second product, and so on. Oocyte grade 780 can be obtained by summing the first product, the second product, etc.

[0136] In some examples, as described above, oocyte grade 780 can classify oocyte quality into grades: A, B, C, and Indeterminate (INC). Grades A, B, and C can represent the likelihood of the oocyte developing into a usable blastocyst, which is useful for successful in vitro fertilization (IVF) treatment. In some embodiments, grade 780 can represent a value assigned to a specific range, reflecting one of the four grades or different numbers within each grade.

[0137] Grade A represents the highest probability of developing into a usable blastocyst. Grade A oocytes exhibit optimal morphological and mechanical characteristics, such as ideal ellipticity of the first polar body, appropriate thickness of the zona pellucida, and / or favorable compactness and roundness of the cytoplasm. High-quality indicators associated with Grade A oocytes demonstrate their strong potential for successful fertilization and subsequent embryonic development.

[0138] Grade B oocytes may have a good chance of developing into usable blastocysts, but not as high as Grade A oocytes. Grade B oocytes may still exhibit favorable morphological and mechanical characteristics, but may deviate slightly from the optimal values ​​observed in Grade A oocytes. Despite these slight deviations, Grade B oocytes may still be considered viable and have a reasonable chance of successful fertilization and embryonic development.

[0139] Oocytes classified as Grade C may have a lower chance of developing into usable blastocysts. Grade C oocytes may exhibit several deviations from optimal morphological and mechanical characteristics, such as irregular ellipticity of the first polar body, suboptimal zona pellucida thickness, and less desirable cytoplasmic compactness and roundness. Although Grade C oocytes may not be ideal, they may still have some potential for successful fertilization and embryonic development, although the likelihood is lower compared to Grade A and Grade B oocytes.

[0140] Oocytes classified as Indeterminate (INC) may have an uncertain likelihood of developing into a usable blastocyst. This classification may stem from insufficient or ambiguous data (e.g., blurry image sequences), making accurate assessment of oocyte quality challenging. Indeterminate oocytes may require further analysis or additional data to determine their viability. An INC grade indicates that the current assessment cannot provide a definitive conclusion regarding the oocyte's successful fertilization and embryonic development potential. By classifying oocytes into these grades, the Oocyte Analysis System 100 provides a more objective, automated, and time-efficient assessment of oocyte quality, helping embryologists and clinicians make informed decisions in the in-vitro fertilization process.

[0141] As an example demonstrating the accuracy of the correlation between oocyte grading and the revealed grading, a total of 488 oocytes were evaluated to assess their likelihood of developing into usable blastocysts. Of the 488 oocytes, 144 oocytes were grade A, 283 oocytes were grade B, and 61 oocytes were grade C. For grade A oocytes, 78.47% were fertilized, 75.34% reached good day 3 (D3) embryonic development, 75.22% developed into blastocysts by day 5, and 69.03% developed into blastocysts suitable for implantation and / or cryopreservation (e.g., usable blastocysts). For grade B oocytes, 71.73% were fertilized, 59.11% reached good day 3 (D3) embryonic development, 71.92% developed into blastocysts by day 5, and 54.19% developed into blastocysts suitable for implantation and / or cryopreservation (e.g., usable blastocysts). For grade C oocytes, 67.21% were fertilized, 53.66% reached good day 3 (D3) embryonic development, 60.98% developed into blastocysts by day 5, and 39.02% developed into blastocysts suitable for implantation and / or embryo cryopreservation (e.g., available blastocysts).

[0142] Example features of oocyte level

[0143] Figures 7B to 7F illustrate example features 122 generated by feature extractor 750 according to some embodiments of this disclosure. As shown in Figure 7B, segmentation model 740 may identify bounding boxes 760A for each of images 104A to 104N. In some examples, bounding boxes 760A may be identified by segmentation model 740 based on the location of at least tool 770A (e.g., micropipette) and / or the depiction of oocytes 110 being aspirated into pressure tool 112 in images 104A to 104N.

[0144] Based on the segmentation model 740 for each identified bounding box 760A in images 104A to 104N, the feature extractor 750 can calculate or obtain the aspiration depth 702A. For example, the feature extractor 750 can calculate the aspiration depth 702A of the oocyte 110 based on one of the identified bounding boxes 760A that has the maximum distance along the X-axis from images 104A to 104N. In this example, the aspiration depth 702A can be the length of the bounding box 760A along the X-axis (e.g., distance A). In other examples, the aspiration depth 702A can be calculated using other methods. For example, the aspiration depth 702A can be calculated based on the method discussed above with reference to FIG2A (e.g., using the pixel intensity of pixels within the bounding box).

[0145] Additionally and / or optionally, the feature extractor 750 may calculate distance B (e.g., indicating the length of bounding box 760A along the Y-axis), the minimum value of bounding box 760A along the X-axis, the maximum value of bounding box 760A along the X-axis, the minimum value of bounding box 760A along the Y-axis, and / or the maximum value of bounding box 760A along the Y-axis to generate other features that may be useful for generating oocyte grade 780.

[0146] As shown in Figure 7C, the segmentation model 740 can identify the first polar body 760B of the oocyte. Based on the first polar body 760B identified by the segmentation model 740, the feature extractor 750 can calculate the ellipticity 702B, representing the ellipticity of the first polar body 760B. Additionally and / or optionally, the feature extractor 750 can calculate the area of ​​the first polar body 760B, which may be useful for generating oocyte grade 780. In some examples, the ellipticity 702B can be calculated as the ratio of the major axis length (distance K) to the minor axis length (distance L) of the ellipse that best fits the shape of the first polar body 760B. A higher ellipticity value (e.g., a higher ratio) may indicate that the first polar body 760B is more elongated, while a ratio close to 1 indicates that the first polar body 760B is more rounded.

[0147] As shown in Figure 7D, the segmentation model 740 can identify the cytoplasm 760C of the oocyte. Based on the cytoplasm 760C identified by at least the segmentation model 740, the feature extractor 750 can calculate area 702C, compactness 704C, roundness 708C, and coverage 706C. Area 702C can be the area of ​​the cytoplasm 760C. In some examples, area 702C can be derived or estimated based on the radius of the cytoplasm 760C, and / or using computer vision techniques (e.g., calculating the number of pixels within the cytoplasm 760C and calculating area 702C based on the resolution of image sequence 102). Compactness 704C can represent the compactness of the cytoplasm 760C and can indicate a compactness measurement associated with the cytoplasm 760C. Compactness 704C can refer to a measurement of the tightness or density of the cytoplasm 760C. Compactness 704C can indicate the health status of the oocyte and its ability to develop into a usable blastocyst. In some cases, the density 704C can be obtained by normalizing the area 702C using the area of ​​a perfect circle that may have the maximum density. Example formulas for calculating the density 704C are provided in Table 1 below.

[0148] Circularity 708C can refer to the roundness of the cytoplasm 760C and can indicate the roundness of the cytoplasm 760C. Coverage 706C can be the ratio between the area of ​​the cytoplasm 760C and the total area of ​​the cytoplasm 760C and the perivitelline space 760D. More specifically, coverage 706C can be a fraction, where the area of ​​the cytoplasm 760C is the numerator and the total or combined area of ​​the cytoplasm 760C and the perivitelline space 760D is the denominator.

[0149] Additionally and / or optionally, feature extractor 750 may calculate geometric information associated with cytoplasm 760C, such as distance C (e.g., indicating the length of cytoplasm 760C along the X-axis), distance D (e.g., indicating the length of cytoplasm 760C along the Y-axis), minimum value of cytoplasm 760C along the X-axis, maximum value of cytoplasm 760C along the X-axis, minimum value of cytoplasm 760C along the Y-axis, maximum value of cytoplasm 760C along the Y-axis, radius 712C of cytoplasm 760C, and / or perimeter 710C of cytoplasm 760C, to produce other features that may be useful for generating oocyte grade 780. In some examples, roundness 708C may be calculated based on the area of ​​cytoplasm 760C and perimeter 710C of cytoplasm 760C.

[0150] As shown in Figure 7E, the segmentation model 740 can identify the perivitelline space 760D of the oocyte. Based on the perivitelline space 760D identified by the segmentation model 740, the feature extractor 750 can calculate the thickness 702D and the coverage 706C. The thickness 702D can represent the thickness of the perivitelline space 760D. As mentioned above, the coverage 706C can be the ratio between the area of ​​the cytoplasm 760C and the total area of ​​the cytoplasm 760C and the perivitelline space 760D. An example of calculating the thickness 702D will be described below with reference to Figure 7E.

[0151] Additionally and / or optionally, feature extractor 750 may calculate distance E (e.g., indicating the length of the perivitelline space 760D along the X-axis), distance F (e.g., indicating the length of the perivitelline space 760D along the Y-axis), the minimum value of the perivitelline space 760D along the X-axis, Xmax 750D (e.g., the maximum value of the perivitelline space 760D along the X-axis), Ymin 740D (e.g., the minimum value of the perivitelline space 760D along the Y-axis), Ymax 730D (e.g., the maximum value of the perivitelline space 760D along the Y-axis), and / or the area 708D of the perivitelline space 760D to generate other features that may be useful for producing oocyte grade 780. The area 708D may be the area between cytoplasm 760C and 760E.

[0152] As shown in Figure 7F, the segmentation model 740 can identify the zona pellucida 760E of the oocyte. Based on the zona pellucida 760E identified by the segmentation model 740, the feature extractor 750 can calculate the thickness 702D and the diameter 702E. The diameter 702E can be the diameter of the oocyte (e.g., oocyte 110). The feature extractor 750 can further calculate the distance G (e.g., indicating the length of the zona pellucida 760E along the X-axis), the distance H (e.g., indicating the length of the zona pellucida 760E along the Y-axis), the Ymax 730E (e.g., the maximum value of the zona pellucida 760E along the Y-axis), the Ymin 740E (e.g., the minimum value of the zona pellucida 760E along the Y-axis), and the Xmax 750E (e.g., the maximum value of the zona pellucida 760E along the X-axis).

[0153] In some examples, feature extractor 750 may calculate diameter 702E based on distance G and distance H. For example, feature extractor 750 may calculate diameter 702E by adding distance G and distance H. In some examples, thickness 702D may be calculated based on Ymax 730E, Ymin 740E, Xmax 750E, Ymax 730D, Ymin 740D, and Xmax 750D. More specifically, thickness 702D may be the median of the differences between Ymax 730E and Ymax 730D, the differences between Xmax 750E and Xmax 750D, and the differences between Ymin 740D and Ymin 740E.

[0154] Table 1 illustrates a list of 122 examples of features that can be used to generate oocyte grade 780, as well as example formulas and / or measurements for deriving the list of 122 examples of features.

[0155] Table 1. Characteristic examples of oocyte-level production

[0156] In some examples, machine learning model 130 may generate an oocyte grade 780 to indicate the likelihood of an oocyte developing into a usable blastocyst, based on some or all of the aspiration depth 702A, ellipticity 702B, area 702C, density 704C, coverage 706C, roundness 708C, thickness 702D, and diameter 702E. In some embodiments, only aspiration depth 702A may be used. As described above, machine learning model 130 may be a regression model. The regression model may include multiple weights. The regression model may be represented using the following equation (C), where βi represents a corresponding weight used to multiply the value of one of the features 122 (e.g., xi), and β0 represents an offset used to generate oocyte grade 780. More specifically, each of the multiple weights may be associated with one of the aspiration depth 702A, ellipticity 702B, area 702C, density 704C, coverage 706C, roundness 708C, thickness 702D, and diameter 702E for generating oocyte grade 780. For example, a first weight can be multiplied by the aspiration depth 702A to produce a first product, a second weight can be multiplied by the ellipticity 702B to produce a second product, and so on. Oocyte grade 780 can be derived by adding the first product, the second product, and so on. In some examples, the oocyte analysis system 100 can set some of the weights to zero, so that some of the features 122 are not considered when generating oocyte grade 780. In other examples, the regression model can use all of the aspiration depth 702A, ellipticity 702B, area 702C, density 704C, coverage 706C, roundness 708C, thickness 702D, and diameter 702E to generate oocyte grade 780.

[0157] Oocyte grade and =β0+Σ i∈s β i x i

[0158] Where S = {702A, 702B, 702C, 704C, 706C, 708C, 702D, 702E}

[0159] In some examples, machine learning model 130 can generate an oocyte grade 780 to indicate the likelihood of an oocyte developing into a usable blastocyst, based on partial or complete aspiration depth 702A, ellipticity 702B, area 702C, density 704C, coverage 706C, roundness 708C, thickness 702D, and diameter 702E. As described above, machine learning model 130 can be a regression model. The regression model can include multiple weights. The regression model can be represented using the following equation (C), where βi represents the corresponding weight used to multiply the value of one of the features 122 (e.g., xi), and β0 represents the offset used to generate oocyte grade 780. More specifically, each of the multiple weights can be associated with one of aspiration depth 702A, ellipticity 702B, area 702C, density 704C, coverage 706C, roundness 708C, thickness 702D, and diameter 702E for generating oocyte grade 780. For example, a first weight can be multiplied by the aspiration depth 702A to produce a first product, a second weight can be multiplied by the ellipticity 702B to produce a second product, and so on. Oocyte grade 780 can be derived by adding the first product, the second product, and so on. In some examples, the oocyte analysis system 100 can set some of the weights to zero, so that some of the features 122 are not considered when generating oocyte grade 780. In other examples, the regression model can use all of the aspiration depth 702A, ellipticity 702B, area 702C, density 704C, coverage 706C, roundness 708C, thickness 702D, and diameter 702E to generate oocyte grade 780.

[0160] Example Oocyte Grading Flowchart

[0161] Figure 8 is a flowchart of an example process 800 for generating oocyte grades. All or at least part of process 800 can be implemented by the oocyte analysis system 100 of Figure 7A. As described above, process 800 can be used to determine the viability of oocytes 110 (e.g., whether the oocyte will develop into a usable blastocyst) without relying on manual and subjective assessment by an embryologist. Therefore, process 800 can be used to achieve more accurate, objective, time-efficient, and automated acquisition of oocyte 110 quality.

[0162] In block 802, the system acquires images that form a sequence of oocyte images. As described above, these images can be captured by a microscope camera and depict a series of events resulting from the application of force to the oocyte by a tool (e.g., a micropipette), resulting in the oocyte's geometry, deformation, and / or movement. In some examples, the system acquires multiple images that form a time-related image sequence. This image sequence depicts a portion of an oocyte (e.g., oocyte 110) and a tool (e.g., pressure tool 112) applying pressure to the oocyte. Each individual image in the sequence can be associated with a single pressure value applied to the oocyte at its respective image capture time.

[0163] In block 804, the system identifies oocyte-related objects based on a segmentation model using oocyte image sequences. In some examples, the system identifies oocyte-related objects via segmentation model 740. These objects may include various parts of the oocyte, such as the zona pellucida (e.g., ZP 760E), perivitelline space (e.g., PVS 760D), first polar body (e.g., FPB 760B), cytoplasm (e.g., CPM 760C), and bounding boxes associated with the tool portion that applies pressure to the oocyte (e.g., BBOX 760A).

[0164] In block 806, the system determines features associated with the oocyte based on the identified object (e.g., feature 122). In some examples, feature 122 may include morphological features that indicate measurements of the oocyte over a time period and the aspiration depth of the oocyte into the pressure-applied tool section (e.g., aspiration depth 702A). Morphological features may include features such as the ellipticity of the first polar body (e.g., ellipticity 702B), the thickness of the zona pellucida (e.g., thickness 702D), the diameter of the oocyte (e.g., diameter 702E), the area of ​​the cytoplasm (e.g., area 702C), the compactness of the cytoplasm (e.g., compactness 704C), the roundness of the cytoplasm (e.g., roundness 708C), and the ratio between the cytoplasmic area and the total area of ​​the cytoplasm and perivitelline space (e.g., coverage 706C).

[0165] In block 808, the system generates an oocyte grade using a machine learning model based on at least one subset of features determined in block 806. In some examples, the system generates an oocyte grade (e.g., oocyte grade 780) using machine learning model 130 based on input including at least one subset of features 122. In some embodiments, this subset may include only aspiration depth. In some embodiments, this subset may include aspiration depth and at least one other feature. Oocyte grade 780 may indicate at least one probability that the oocyte will develop into a usable blastocyst. Machine learning model 130 may be a regression model that includes multiple weights, each weight associated with a feature, to generate oocyte grade 780. Oocyte grade 780 may then be provided via an interactive user interface for further analysis.

[0166] All methods and tasks described herein can be performed and fully automated by a computer system. In some cases, a computer system may include multiple different computers or computing devices (e.g., physical servers, workstations, storage arrays, cloud computing resources, etc.) that communicate and interoperate via a network to perform the described functions. Each such computing device typically includes a processor (or multiple processors) that executes program instructions or modules stored in memory or other non-transitory computer-readable storage media or devices (e.g., solid-state storage devices, disk drives, etc.). The various functions disclosed herein may be embodied in these program instructions or implemented in specific application circuitry of the computer system (e.g., ASICs or FPGAs). When a computer system includes multiple computing devices, these devices may be located in the same location, but not necessarily. The results of the disclosed methods and tasks can be persistently stored by converting physical storage devices such as solid-state memory chips or disks into different states. In some embodiments, the computer system may be a cloud-based computing system whose processing resources are shared by multiple different business entities or other users.

[0167] The processes described or illustrated in this disclosure may be initiated in response to an event, such as according to a predetermined or dynamically determined schedule, upon request by a user or system administrator, or in response to certain other events. When these processes are initiated, a set of executable program instructions stored on one or more non-transitory computer-readable media (e.g., hard disk, flash memory, removable media, etc.) may be loaded into the memory (e.g., RAM) of a server or other computing device. These executable instructions may then be executed by the computer processor, the hardware underlying the computing device. In some embodiments, these processes, or portions thereof, may be implemented serially or in parallel on multiple computing devices and / or multiple processors.

[0168] According to embodiments, certain actions, events, or functions of any process or algorithm described herein may be performed in a different order, and may be added, combined, or omitted entirely (e.g., not all described operations or events are necessary for the practice of the algorithm). Furthermore, in some embodiments, operations or events may be performed in parallel, for example, through multithreading, interrupt handling, or multiple processors or processor cores or other parallel architectures, rather than sequentially.

[0169] The various example logic blocks, modules, routines, and algorithmic steps described in the embodiments disclosed herein can be implemented as electronic hardware (e.g., ASIC or FPGA devices), computer software running on computer hardware, or a combination of both. Furthermore, the various example logic blocks and modules described in the embodiments disclosed herein can be implemented or executed by a machine, such as a processor device, digital signal processor (DSP), application-specific integrated circuit (ASIC), field-programmable gate array (FPGA) or other programmable logic device, discrete gate or transistor logic, discrete hardware components, or any combination thereof, designed to perform the functions described herein. The processor device may be a microprocessor, but alternatively, it may be a controller, microcontroller, or state machine, a combination thereof, or the like. The processor device may include circuitry configured to process computer-executable instructions. In another embodiment, the processor device includes an FPGA or other programmable device that performs logical operations without processing computer-executable instructions. The processor device may also be implemented as a combination of computing devices, such as a combination of a DSP and a microprocessor, multiple microprocessors, one or more microprocessors combined with a DSP core, or any other such configuration. While this document primarily describes digital technologies, the processor device may also include major analog components. For example, some or all of the rendering techniques described herein can be implemented in analog circuits or mixed analog and digital circuits. The computing environment can include any type of computer system, including but not limited to microprocessor-based computer systems, mainframe computers, digital signal processors, portable computing devices, device controllers, or computing engines within devices, to name just a few.

[0170] Elements of the methods, processes, routines, or algorithms described in the embodiments disclosed herein may be directly embodied in hardware, in software modules executed by a processor device, or a combination of both. Software modules may reside in RAM, flash memory, ROM, EPROM, EEPROM, registers, hard disks, removable disks, CD-ROMs, or any other form of non-transitory computer-readable storage media. Example storage media may be coupled to a processor device, enabling the processor device to read information from and write information to the storage medium. Alternatively, the storage medium may be integrated with the processor device. The processor device and storage medium may reside in an ASIC. The ASIC may reside in a user terminal. Alternatively, the processor device and storage medium may reside as discrete components in a user terminal.

[0171] The conditional language used herein, such as *may*, *possibly*, *may*, *can*, *e.g.*, unless explicitly stated otherwise or understood from the context, is generally intended to express that certain embodiments include certain features, elements, or steps, while other embodiments do not. Therefore, such conditional language is not intended to imply that a feature, element, or step is necessary in any way for one or more embodiments, or that one or more embodiments necessarily include logic for determining whether such features, elements, or steps are included in any particular embodiment, regardless of any additional input or prompting. The terms *comprising*, *including*, *having*, etc., are synonyms and are used inclusively in an open-ended manner, not excluding additional elements, features, behaviors, operations, etc. Furthermore, the term *or* is used in its inclusive sense (rather than its exclusive sense), and thus, when used, for example, to connect lists of elements, the term *or* indicates one, some, or all of the elements in the list.

[0172] Unless otherwise explicitly stated, delimited language such as “at least one of X, Y, or Z” is generally understood in context to mean that an item, term, etc., can be X, Y, or Z, or any combination thereof (e.g., X, Y, or Z). Therefore, such delimited language generally does not, and should not, imply that certain embodiments require at least one of X, at least one of Y, and at least one of Z to be present.

[0173] While the foregoing detailed description has shown, described, and pointed out novel features applicable to various embodiments, it will be understood that various omissions, substitutions, and changes may be made to the form and details of the illustrated apparatus or algorithm without departing from the spirit of this disclosure. As acknowledged, some embodiments described herein may be embodied in forms that do not provide all the features and benefits described herein, as some features may be used or implemented separately from other features. All modifications within the meaning and scope of the claims should be included within their scope.

Claims

1. A method implemented by a system of one or more computers, characterized in that, include: Acquire multiple images to form a time-related image sequence, the image sequence depicting portions of an oocyte and a tool applying pressure to the oocyte, wherein individual images are associated with individual pressure values ​​applied to the oocyte at their respective image capture times; Objects associated with the oocyte were identified using a segmentation model; Based on geometric measurements of the object at least partially associated with the oocyte, characteristics associated with the oocyte are determined, including morphological features indicating measurements of the oocyte within the time period, and the aspiration depth of the oocyte into the portion of the tool applying pressure to the oocyte; and Based on input values ​​including the aspiration depth, an oocyte grade is generated by a machine learning model, wherein the oocyte grade at least indicates the likelihood that the oocyte will develop into a usable blastocyst.

2. The method implemented by a system of one or more computers according to claim 1, characterized in that, The characteristics associated with the oocyte include: Determine the geometric information associated with the object; and Based on the geometric information associated with the object, the features associated with the oocyte are calculated.

3. The method implemented by a system of one or more computers according to claim 1, characterized in that, The object associated with the oocyte includes at least one of the following: the zona pellucida of the oocyte, the perivitelline space of the oocyte, the first polar body of the oocyte, the cytoplasm of the oocyte, and a bounding box associated with the portion of the tool that applies pressure to the oocyte.

4. The method implemented by a system of one or more computers according to claim 3, characterized in that, The morphological characteristics associated with the oocyte include at least one of the following: the ellipticity of the first polar body, the thickness of the zona pellucida, the diameter of the oocyte, the area of ​​the cytoplasm, the compactness of the cytoplasm, the roundness of the cytoplasm, and the ratio between the area of ​​the cytoplasm and the total area of ​​the cytoplasm and the perivitelline space.

5. The method implemented by a system of one or more computers according to claim 1, characterized in that, The segmentation model is U-Net.

6. The method implemented by a system of one or more computers according to claim 1, characterized in that, The machine learning model mentioned is a regression model.

7. The method implemented by a system of one or more computers according to claim 6, characterized in that, The regression model includes multiple weights, each of which is associated with one of the features used to generate the oocyte grade.

8. The method implemented by a system of one or more computers according to claim 1, characterized in that, The input values ​​further include at least a subset of the morphological features, wherein the subset of morphological features includes at least one of the following: the ellipticity of the first polar body of the oocyte, the thickness of the zona pellucida of the oocyte, the diameter of the oocyte, the area of ​​the cytoplasm of the oocyte, the compactness of the cytoplasm of the oocyte, the roundness of the cytoplasm of the oocyte, and the ratio between the area of ​​the cytoplasm and the total area of ​​the cytoplasm and the perivitelline space.

9. The method implemented by a system of one or more computers according to claim 1, characterized in that, The portion of the tool is a micropipette, and during the time period, the micropipette contacts the oocyte and is configured to apply the pressure.

10. The method implemented by a system of one or more computers according to claim 9, characterized in that, The feature includes the inner diameter of the micropipette, and the inner diameter is used to normalize the morphological feature or the inhalation depth.

11. The method implemented by a system of one or more computers according to claim 1, characterized in that, Determining the features includes: For each individual image in the images, bounding boxes are identified at least based on the position of the portion of the tool within the individual image to obtain a plurality of said bounding boxes; and The inhalation depth is calculated based on a first bounding box that has the largest distance in a first direction among the multiple bounding boxes.

12. The method implemented by a system of one or more computers according to claim 1, characterized in that, Determining the features includes: For each individual image in the image: The bounding box is identified at least based on the position of the portion of the tool within the individual images; and The aspiration depth of the oocyte into the portion of the tool is determined using the pixel intensity of at least a portion of the pixels of the individual images within the bounding box.

13. An oocyte quality analysis system, characterized in that, A non-transitory computer storage medium comprising one or more processors and stored instructions, wherein when the instructions are executed by one or more of the processors, the one or more processors: Acquire multiple images to form a time-related image sequence, the image sequence depicting portions of an oocyte and a tool applying pressure to the oocyte, wherein individual images are associated with individual pressure values ​​applied to the oocyte at their respective image capture times; Objects associated with the oocyte were identified using a segmentation model; Based on geometric measurements of the object at least partially associated with the oocyte, characteristics associated with the oocyte are determined, including morphological features indicating measurements of the oocyte within the time period, and the aspiration depth of the oocyte into the portion of the tool applying pressure to the oocyte; and Based on input values ​​including the aspiration depth, an oocyte grade is generated by a machine learning model, wherein the oocyte grade at least indicates the likelihood that the oocyte will develop into a usable blastocyst.

14. The oocyte quality analysis system according to claim 13, characterized in that, The characteristics associated with the oocyte include: Determine the geometric information associated with the object; and Based on the geometric information associated with the object, the features associated with the oocyte are calculated.

15. The oocyte quality analysis system according to claim 13, characterized in that, The object associated with the oocyte includes at least one of the following: the zona pellucida of the oocyte, the perivitelline space of the oocyte, the first polar body of the oocyte, the cytoplasm of the oocyte, and a bounding box associated with the portion of the tool that applies pressure to the oocyte.

16. The oocyte quality analysis system according to claim 15, characterized in that, The morphological characteristics associated with the oocyte include at least one of the following: the ellipticity of the first polar body, the thickness of the zona pellucida, the diameter of the oocyte, the area of ​​the cytoplasm, the compactness of the cytoplasm, the roundness of the cytoplasm, and the ratio between the area of ​​the cytoplasm and the total area of ​​the cytoplasm and the perivitelline space.

17. The oocyte quality analysis system according to claim 13, characterized in that, The input values ​​further include at least a subset of the morphological features, wherein the subset of morphological features includes at least one of the following: the ellipticity of the first polar body of the oocyte, the thickness of the zona pellucida of the oocyte, the diameter of the oocyte, the area of ​​the cytoplasm of the oocyte, the compactness of the cytoplasm of the oocyte, the roundness of the cytoplasm of the oocyte, and the ratio between the area of ​​the cytoplasm and the total area of ​​the cytoplasm and the perivitelline space.

18. The oocyte quality analysis system according to claim 13, characterized in that, Determining the features includes: For each individual image in the images, bounding boxes are identified at least based on the position of the portion of the tool within the individual image to obtain a plurality of said bounding boxes; and The inhalation depth is calculated based on a first bounding box that has the largest distance in a first direction among the multiple bounding boxes.

19. A non-transitory computer-readable medium storing instructions, characterized in that, When the instructions are executed by one or more processors, the one or more processors: Acquire multiple images to form a time-related image sequence, the image sequence depicting portions of an oocyte and a tool applying pressure to the oocyte, wherein individual images are associated with individual pressure values ​​applied to the oocyte at their respective image capture times; Objects associated with the oocyte were identified using a segmentation model; Based on geometric measurements of the object at least partially associated with the oocyte, characteristics associated with the oocyte are determined, including morphological features indicating measurements of the oocyte within the time period, and the aspiration depth of the oocyte into the portion of the tool applying pressure to the oocyte; and Based on input values ​​including the aspiration depth and at least a subset of the morphological features, an oocyte grade is generated by a machine learning model, wherein the oocyte grade at least indicates the likelihood that the oocyte will develop into a usable blastocyst, and wherein the subset of morphological features includes at least one of the ellipticity of the oocyte's first polar body, the thickness of the zona pellucida of the oocyte, the diameter of the oocyte, the area of ​​the oocyte's cytoplasm, the compactness of the oocyte's cytoplasm, the roundness of the oocyte's cytoplasm, and the ratio between the area of ​​the cytoplasm and the total area of ​​the cytoplasm and the perivitelline space.

20. The non-transitory computer-readable medium for storing instructions according to claim 19, characterized in that, Determining the features includes: For each individual image in the images, bounding boxes are identified based on the position of the portion of the tool within the individual image to obtain a plurality of bounding boxes; and The inhalation depth is calculated based on a first bounding box that has the largest distance in a first direction among the multiple bounding boxes.

Images