An organoid intelligent analysis method and system based on deep learning

Abstract

The application discloses an organoid intelligent analysis method and system based on deep learning, and relates to the technical field of analysis, and comprises the following steps: acquiring morphological image sequences of organoids at fixed time intervals, establishing a morphological drift baseline of organoids in a spontaneous culture state; after drug action, continuously acquiring morphological image sequences of organoids at the same fixed time intervals, comparing morphological characteristics of each time node with the morphological drift baseline; and early qualitative discrimination of drug sensitivity of organoids is carried out. According to the application, the morphological drift baseline of organoids in the spontaneous culture state is established before drug action, and the morphological image sequences are continuously acquired after drug action for comparison, compared with the lag mode that the traditional end-point detection needs to wait for a fixed time length to obtain the drug sensitivity result, the application can detect the morphological response of organoids in the early stage of drug action, and more time windows are obtained for clinical treatment decision-making.

Description

Technical Field

[0001] This invention relates to the field of analytical technology, and in particular to an organoid intelligent analysis method and system based on deep learning. Background Technology

[0002] Organoids are miniature organ-like structures formed by the self-assembly of adult stem cells or pluripotent stem cells under three-dimensional in vitro culture conditions. They can highly mimic the original organs in terms of histological features, cellular composition, and physiological function. In recent years, organoid technology has been widely used in tumor drug sensitivity testing, personalized medicine guidance, and new drug development. In the context of precision oncology, in vitro drug sensitivity testing using patient-derived tumor organoids can predict a patient's response to candidate drugs before administration, thereby avoiding ineffective treatment and shortening the clinical decision-making cycle.

[0003] Currently, the mainstream technical approach for organoid drug susceptibility testing still relies on endpoint-based detection. Specifically, existing methods typically co-culture tumor organoids with candidate drugs for a fixed duration (e.g., 72 hours). After the drug has fully acted, the viability or metabolic activity data of the organoids at the culture endpoint are obtained through cell metabolic activity assays (e.g., ATP assay, MTT assay), fluorescence staining (e.g., Calcein-AM / PI double staining), or endpoint readings from reagent kits. This allows for the calculation of drug inhibition rates or half-maximal inhibitory concentrations (IC50). While this endpoint-based detection approach provides quantitative drug susceptibility assessment results, it is essentially a "post-hoc" strategy: during the long culture period (several days), researchers cannot know whether the drug has acted on the organoids, when it began to act, or in what manner; they can only passively wait for the predetermined endpoint.

[0004] The inherent lag in the aforementioned endpoint testing model directly leads to significant delays in clinical decision-making. For cancer patients, the time it takes to obtain drug sensitivity test results directly impacts the window for formulating subsequent treatment plans. Current methods, from organoid culture and drug processing to endpoint testing, typically take 10 to 15 days in total. Furthermore, endpoint testing itself cannot provide dynamic information about the drug's action, making it difficult for clinicians to adjust medication strategies before obtaining results. More importantly, if the endpoint test results show the drug is ineffective, the preceding days of culture and testing have irreversibly consumed the patient's optimal treatment window, resulting in a double loss of "wasted time window" and "treatment opportunity cost."

[0005] To shorten the drug sensitivity testing cycle, some studies have attempted to introduce mid-term observation points during drug action, using early acquisition of organoid images to predict drug sensitivity trends. However, organoids exhibit inherent morphological evolution rhythms during spontaneous culture, such as periodic contour contraction, cavity size fluctuations, and edge undulations. These spontaneous morphological drifts are highly similar in appearance to drug-induced morphological changes. Without a dynamic understanding of the inherent morphological fluctuation patterns of organoids during culture, spontaneous drifts can easily be misinterpreted as drug responses, or early true responses can be missed as environmental noise, leading to a serious lack of reliability in early predictions. Summary of the Invention

[0006] The purpose of this invention is to provide a method and system for intelligent organoid analysis based on deep learning, so as to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a deep learning-based intelligent organoid analysis method, comprising the following steps:

[0008] Before drug administration, morphological image sequences of organoids were acquired at fixed time intervals to establish a morphological drift baseline of organoids in spontaneous culture.

[0009] After drug administration, morphological image sequences of organoids are continuously acquired at the same fixed time intervals, and the morphological features at each time point are compared with the morphological drift baseline.

[0010] Identify the continuous range of morphological features that first deviate from the morphological drift baseline, and use the starting time of this continuous range of changes as the earliest detectable change window of the drug response;

[0011] Based on the morphological response characteristics within the earliest detectable change window, the drug sensitivity of organoids can be qualitatively determined at an early stage.

[0012] Preferably, establishing the morphological drift baseline of organoids in spontaneous culture includes:

[0013] Identify the morphological change rhythms of organoids in spontaneous culture and divide the morphological change rhythms into different phase intervals;

[0014] A morphological drift sub-baseline is established for each phase interval. The morphological drift sub-baseline is used to record the inherent morphological fluctuation range of organoids in the corresponding phase interval.

[0015] Preferably, establishing the morphological drift baseline further includes:

[0016] When the observation duration before drug action does not cover the complete morphological change rhythm, based on the established morphological drift sub-baseline of some phase intervals, the rhythm continuity of the morphological drift sub-baseline of the unobserved phase interval is inferred to obtain the inferred morphological drift baseline that covers the complete morphological change rhythm.

[0017] Preferably, the continuous acquisition of organoid morphological image sequences after drug administration includes:

[0018] Based on the spatial differences of organoids in the well plate, the effective contact time for substantial contact between organoids and drugs in each well position was determined.

[0019] Starting from the effective contact time of each pore, morphological image sequences of organoids within that pore are acquired at fixed time intervals.

[0020] Preferably, comparing the morphological features at each time point with the morphological drift baseline includes:

[0021] In cases where multiple organoids exist within the same pore site, the morphological changes of each independent organoid at each time point are tracked separately.

[0022] Establish individualized comparison sequences between the morphological feature changes of each independent organoid and the morphological drift baseline to avoid confusion between the morphological feature changes of multiple organoids.

[0023] Preferably, the continuous range of changes in the first deviation of the identified morphological feature from the morphological drift baseline includes:

[0024] Distinguish between the instantaneous fluctuation range and the continuous deviation range of morphological features; when morphological features show the same trend of change in multiple adjacent time nodes, and the same trend of change in the same direction does not reverse in subsequent time nodes, the range is determined to be the continuous deviation range.

[0025] The starting time of the first time that is determined to be a continuous deviation from the interval is taken as the starting time of the earliest detectable change window.

[0026] Preferably, the identification of the continuous deviation range further includes:

[0027] After determining the starting time of the continuous deviation interval, the single time node immediately preceding the starting time is traced back to determine whether the single time node has shown deviation precursor features. The deviation precursor features are manifested as morphological features that do not exceed the inherent fluctuation range of the morphological drift baseline and do not meet the continuous deviation confirmation conditions of the multiple adjacent time nodes, but have shown a tendency to change in the same direction as the continuous deviation interval.

[0028] When the aforementioned early warning features are present, the single time node is used as a preliminary extension point for the earliest detectable change window.

[0029] Preferably, the early qualitative discrimination based on morphological response features within the earliest detectable change window includes:

[0030] By analyzing the spatial unfolding order of morphological response features within the earliest detectable change window, it can be determined whether the drug effect gradually penetrates from the peripheral region to the inner region of the organoid or occurs synchronously within the entire organoid.

[0031] When the infiltration is determined to be a gradual process from the periphery to the interior, it is qualitatively classified as membrane permeability-dependent drug sensitivity; when the infiltration is determined to be a synchronous process occurring throughout the body, it is qualitatively classified as metabolic pathway interference-type drug sensitivity.

[0032] Preferably, the method further includes:

[0033] After identifying the earliest detectable change window, an organoid response urgency classification to the drug is established based on the combination of the rate and direction of change of morphological response features within the window.

[0034] Based on the response urgency level, the acquisition time density of subsequent morphological images is adjusted; when the response urgency level is high, the acquisition time density of subsequent morphological images is increased; when the response urgency level is low, the acquisition time density of subsequent morphological images is maintained or reduced.

[0035] A deep learning-based organoid intelligent analysis system, comprising:

[0036] The baseline establishment module is used to acquire morphological image sequences of organoids at fixed time intervals before drug action, identify the rhythm of morphological changes and establish morphological drift sub-baselines by phase, and infer the rhythmic continuity of unobserved phases to obtain the morphological drift baseline.

[0037] The temporal image acquisition module is used to continuously acquire morphological image sequences of organoids at fixed time intervals after drug action, and determine the effective contact time of each well based on the spatial position difference of the organoids in the well plate.

[0038] The individualized comparison module is used to compare the morphological features at each time point with the morphological drift baseline, and to establish individualized comparison sequences by tracking multiple independent organoids within the same pore location.

[0039] The continuous deviation recognition module is used to identify the continuous change range of the morphological feature when it first deviates from the morphological drift baseline, distinguish between instantaneous fluctuations and continuous deviations, and trace back the deviation precursor features that are immediately adjacent to a single time point.

[0040] The window determination module is used to determine the start time of the first time that is identified as a persistent deviation interval as the earliest detectable change window of the drug response;

[0041] The early discrimination module is used to perform early qualitative discrimination based on the morphological response features within the earliest detectable change window, and to analyze the spatial unfolding order to determine the drug action type.

[0042] And an observation adjustment module, used to establish a response urgency level based on the combination of the rate of change and the direction of change of the morphological response features within the earliest detectable change window, and to adjust the acquisition time density of subsequent morphological images according to the response urgency level.

[0043] The technical effects and advantages of this invention are as follows:

[0044] (1) This invention establishes a morphological drift baseline in the spontaneous culture state of organoids before drug treatment and continuously acquires morphological image sequences for comparison after drug treatment. It can identify the continuous change interval of the first deviation of morphological features from the baseline and take the starting time of the first continuous deviation interval as the earliest detectable change window of drug response. Compared with the traditional endpoint detection, which requires a fixed waiting time to obtain drug sensitivity results, this invention can detect the morphological response of organoids in the early stage of drug treatment, thereby significantly shortening the drug sensitivity detection cycle and gaining more time window for clinical treatment decisions;

[0045] (2) This invention establishes a morphological drift baseline for organoids under spontaneous culture conditions, records the inherent morphological fluctuation range of organoids, and compares the morphological characteristics at each time point with the baseline to identify the continuous change range of the first deviation from the baseline, while distinguishing between instantaneous fluctuations and continuous deviations. Compared with the shortcomings of existing technologies that lack dynamic understanding of the inherent morphological fluctuation law of organoids and thus have difficulty distinguishing between spontaneous drift and drug-induced response, this invention can effectively eliminate the interference caused by the spontaneous morphological evolution of organoids, avoid misjudging spontaneous drift as drug response or omitting early real response as environmental noise, and improve the accuracy and reliability of early drug sensitivity assessment.

[0046] (3) By continuously acquiring morphological image sequences at various time points after drug action and establishing individualized comparison sequences, this invention can track the dynamic evolution of organoid responses to drugs, including the start time, trend of change, and spatial unfolding order of the response. Compared with the limitations of traditional endpoint detection, which can only provide static live / dead status or metabolic activity data at the culture endpoint and cannot obtain dynamic information during drug action, this invention can continuously provide dynamic information on the morphological response of organoids during drug action, enabling researchers and clinicians to know in a timely manner whether the drug has taken effect and in what way it has taken effect within the detection period, providing data support for early adjustment of drug use strategies. Attached Figure Description

[0047] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention, but do not constitute a limitation thereof. In the drawings:

[0048] Figure 1 This is a schematic flowchart of the method of the present invention;

[0049] Figure 2 A schematic diagram illustrating the process of establishing a morphological drift baseline;

[0050] Figure 3 This is a schematic diagram of the module architecture of the system of the present invention. Detailed Implementation

[0051] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0052] Example 1: According to Figure 1-2 As shown, this embodiment provides a deep learning-based intelligent organoid analysis method, including the following steps:

[0053] S1. Before drug administration, morphological image sequences of organoids are acquired at fixed time intervals to identify the morphological change rhythm of organoids in spontaneous culture, morphological drift sub-baselines are established by phase, and rhythmic continuity inference is performed on unobserved phases to obtain the morphological drift baseline.

[0054] In this embodiment, before drug administration, morphological image sequences of organoids are acquired at fixed time intervals to establish a morphological drift baseline of organoids in spontaneous culture. Specifically:

[0055] Tumor organoids were cultured in standard well plates, and bright-field morphological image sequences of the organoids were continuously acquired at fixed time intervals of 6 hours within a set observation period before drug application (e.g., 24 to 48 hours). It should be noted that the setting of this fixed time interval should balance the accuracy of capturing organoid morphological changes with the stability of the culture environment, avoiding phototoxic interference caused by overly frequent imaging, or the omission of rhythmic features due to overly sparse imaging.

[0056] Based on the acquired bright-field morphological image sequence, the morphological change rhythm of organoids under spontaneous culture was identified. It should be noted that during spontaneous culture without drug intervention, the outer contour, internal cavity, and edge texture of organoids typically exhibit periodic fluctuations, such as a periodic contraction and expansion of the contour every 6 to 8 hours. By tracking the temporal patterns of organoid contour fluctuations, cavity diameter changes, and edge texture fluctuations, their inherent morphological change rhythms were identified.

[0057] The identified morphological change rhythms are divided into different phase intervals. In this embodiment, a complete rhythmic cycle is divided into three phase intervals: the expansion phase, corresponding to the stage where the organoid outline stretches and the cavity diameter increases; the stable phase, corresponding to the stage where the organoid outline is relatively stable and morphological changes are weak; and the contraction phase, corresponding to the stage where the organoid outline shrinks and the cavity diameter decreases. It should be noted that the phase intervals are divided based on the directional transition nodes of the organoid morphological change trend, rather than fixed time scales. Therefore, the phase duration is allowed to vary for different organoids or under different culture conditions.

[0058] Morphological drift sub-baselines are established for each phase interval. In this embodiment, for the expansion phase, the upper and lower limits of the organoid contour undulation amplitude and the inherent fluctuation range of the cavity diameter increase are recorded; for the stable phase, the range of slight fluctuations in the relatively stable contour state is recorded; for the contraction phase, the inherent fluctuation range of the contour shrinkage amplitude and the cavity diameter decrease is recorded. Each morphological drift sub-baseline is used to record the inherent morphological fluctuation range of the organoid in the corresponding phase interval, serving as a reference benchmark for subsequent drug response determination.

[0059] When the observation duration prior to drug action does not cover the complete morphological change rhythm, rhythmic continuity is inferred for the morphological drift sub-baseline in the unobserved phase intervals based on the established morphological drift sub-baseline for some phase intervals. In this embodiment, if the observation duration only covers the expansion and stable phases but not the contraction phase, the morphological drift sub-baseline for the contraction phase is inferred and completed based on the observed period duration, fluctuation amplitude, and gradual change trend of the expansion and stable phases, combined with historical rhythmic data from organoid culture.

[0060] It should be noted that the specific rules for rhythm continuation inference include: based on the sum of the period durations of the observed phase intervals, the expected duration of the unobserved phase intervals is allocated according to the typical duration ratio of each phase in the organoid culture history data; or based on the gradual slope of the fluctuation boundary of the observed adjacent phase intervals, the fluctuation boundary range of the unobserved phase interval is linearly extrapolated. For example, if the ratio of the period duration of the expansion phase to the stable phase is 2:1, and the typical duration ratio of the contraction phase to the stable phase in the historical data is 1.5:1, then the expected duration of the contraction phase is extrapolated according to this ratio; if the upper limit of the contour fluctuation of the expansion phase decreases as it transitions to the stable phase, then the upper limit of the contour fluctuation of the contraction phase is linearly extrapolated according to this decreasing slope, thereby obtaining the inferred morphological drift baseline that covers the complete morphological change rhythm.

[0061] S2, after drug action, continuously acquire morphological image sequences of organoids at the same fixed time intervals, and determine the effective contact time when the organoids and the drug make substantial contact in each well based on the spatial position differences of the organoids in the well plate.

[0062] In this embodiment, after drug administration, morphological image sequences of organoids are continuously acquired at the same fixed time intervals, specifically:

[0063] Based on the spatial differences in organoid positions within the well plate, the effective contact time for substantial contact between the organoid and the drug in each well was determined. It should be noted that in a standard well plate, differences in liquid surface tension, temperature gradient, and evaporation rate between edge and center wells lead to differences in the diffusion rate and coverage sequence of drug droplets within the wells. Edge wells typically experience faster drug diffusion due to the liquid edge effect, while center wells may experience a slight delay in drug coverage due to impeded liquid surface tension.

[0064] In this embodiment, the expansion process of the drug droplet coverage area within the well plate is automatically identified using bright-field image sequences. When the edge of the drug surface completely covers the area where the organoid is located, the time corresponding to that frame of the image is recorded as the effective contact time when the organoid and the drug make substantial contact. It should be noted that this effective contact time is later than the drug addition operation time, and the duration of this difference depends on the spatial location of the well, the volume of the drug droplet, and the surface characteristics of the well plate. For example, for edge wells, the effective contact time may be delayed by several minutes to more than ten minutes compared to the drug addition time; for center wells, the delay time may be slightly longer.

[0065] Starting from the effective contact time of each well site, morphological image sequences of organoids within that well site are acquired at fixed time intervals (e.g., every 6 hours). It should be noted that using the effective contact time as the time zero point, rather than the drug administration time, can eliminate timing deviations caused by drug penetration delays and ensure the comparability of drug response timings of organoids within different well sites.

[0066] S3, compare the morphological features at each time point with the morphological drift baseline, and establish individualized comparison sequences by tracking multiple independent organoids within the same pore location;

[0067] In this embodiment, the morphological features at each time point are compared with the morphological drift baseline, specifically as follows:

[0068] For cases where multiple organoids exist within the same well, the morphological changes of each individual organoid are tracked at each time point. It should be noted that multiple independent organoids are typically cultured simultaneously within a single well of a standard plate, and the response of each organoid to the drug exhibits heterogeneity. Achieving a group average of the morphological features of multiple organoids would mask the true differences in individual organoid responses. Therefore, this embodiment uses an image segmentation algorithm to obtain the contour mask of each individual organoid, extracting morphological features such as contour roughness, cavity structure, and edge texture for each organoid. An individualized comparison sequence of the morphological feature changes of each individual organoid and the morphological drift baseline is established to avoid confusion between the morphological feature changes of multiple organoids.

[0069] S4 identifies the continuous change interval of the first deviation of morphological features from the morphological drift baseline, distinguishes between instantaneous fluctuations and continuous deviations, traces back the deviation precursor features adjacent to a single time node, and takes the starting time of the first interval determined to be a continuous deviation as the earliest detectable change window of the drug response.

[0070] In this embodiment, identifying the continuous range of change in morphological features where they first deviate from the morphological drift baseline is specifically as follows:

[0071] It is important to distinguish between transient fluctuations and persistent deviations in morphological characteristics. It should be noted that organoids may experience transient morphological fluctuations during culture due to factors such as environmental temperature disturbances, culture medium evaporation, or mechanical vibration; these fluctuations are transient. In contrast, drug effects manifest as persistent, directional changes in morphological characteristics, which are persistent deviations.

[0072] In this embodiment, when morphological features show a consistent trend of change across multiple adjacent time points, and this trend does not reverse in subsequent time points, the interval is determined to be a continuous deviation interval. It should be noted that multiple adjacent time points typically refer to three or more consecutive time points. For example, with a fixed time interval of 6 hours, if the roughness of the organoid contour shows an increasing trend at the 6th, 12th, and 18th hours after the effective contact time, and no reversal trend of roughness reduction appears at the 24th hour, then the interval is determined to be a continuous deviation interval. Conversely, if the roughness increases only at the 6th hour and returns to its original level at the 12th hour, it is determined to be an instantaneous fluctuation interval.

[0073] The starting time of the first identified persistent deviation interval is taken as the starting time of the earliest detectable change window. In this embodiment, if the persistent deviation interval begins 6 hours after the effective exposure time, this time is taken as the starting time of the earliest detectable change window of the drug response.

[0074] After determining the starting time of the persistent deviation interval, the system traces back to the immediately preceding single time point to determine whether any early signs of deviation have appeared at that single time point. It should be noted that the immediately preceding single time point refers to the observation node preceding the starting time of the persistent deviation interval. For example, if the fixed time interval is 6 hours, and the persistent deviation interval starts at the 6th hour, then the system traces back to the immediately preceding observation node (i.e., the 0th hour node); if the fixed time interval is shortened to 3 hours due to a high response urgency level, and the persistent deviation interval starts at the 6th hour, then the system traces back to the immediately preceding observation node (i.e., the 3rd hour node).

[0075] The early signs of deviation are manifested as follows: although the morphological features do not exceed the inherent fluctuation range of the morphological drift baseline and do not meet the continuous deviation confirmation condition for multiple adjacent time nodes, they show a tendency to change in the same direction as within the continuous deviation interval. In this embodiment, for example, at the immediately preceding observation node (hour 0 node), the organoid contour roughness is slightly increased compared to the baseline, but this slight increase is still within the inherent fluctuation range of the stable phase, and only a single time node shows this tendency, not yet meeting the continuous deviation confirmation condition of three consecutive time nodes changing in the same direction. At this time, it is determined that the hour 0 node has early signs of deviation and is used as a preliminary extension point for the earliest detectable change window. It should be noted that this preliminary extension point can be used to improve the sensitivity of early response identification. In scenarios where it is necessary to capture very early drug responses, this preliminary extension point can be included in the time domain range of the earliest detectable change window.

[0076] S5, based on the morphological response characteristics within the earliest detectable change window, analyzes the spatial unfolding order to determine the type of drug action, and performs early qualitative identification of the drug sensitivity of organoids;

[0077] In this embodiment, early qualitative discrimination is performed based on the morphological response features within the earliest detectable change window, specifically as follows:

[0078] The spatial unfolding order of morphological response features within the earliest detectable change window is analyzed. In this embodiment, using the organoid centroid as the reference origin, the organoid interior is divided into several concentric radial layers according to radius, such as the peripheral region (radius 0.5 to 1.0 times the maximum radius), the intermediate region (radius 0.2 to 0.5 times the maximum radius), and the inner region (radius 0 to 0.2 times the maximum radius). The order of appearance of morphological response features within each radial layer is tracked.

[0079] The determination is made as to whether the drug effect penetrates gradually from the peripheral region to the inner region of the organoid, or occurs synchronously throughout the entire organoid. In this embodiment, if, within the earliest detectable change window, the peripheral region first shows an increase in edge roughness, followed by changes in the cavity structure in the middle region, and finally a response in the inner region, then the drug effect is determined to be a gradual penetration from the periphery to the inner region. Conversely, if the peripheral, middle, and inner regions all show morphological responses synchronously within the same time window, and there is no obvious spatial order of development, then the drug effect is determined to occur synchronously throughout the entire organoid.

[0080] When the infiltration is determined to be a gradual process from the periphery to the interior, it is qualitatively classified as membrane permeability-dependent drug sensitivity. It should be noted that membrane permeability-dependent drugs (such as certain antibiotics or membrane disruptors) usually act first on the cell membrane of the peripheral cells of organoids, causing the morphological features of the peripheral region to respond first, and then gradually penetrate into the interior, thus exhibiting a spatial unfolding sequence from the outside to the inside.

[0081] When global synchronization is identified, the qualitative classification indicates sensitivity to metabolic pathway-disrupting drugs. It should be noted that metabolic pathway-disrupting drugs (such as certain chemotherapeutic drugs or signaling pathway inhibitors) may rapidly penetrate organoids and act on intracellular metabolic pathways, causing cellular synchronization to be affected throughout the organoid, thus exhibiting spatial characteristics of a global synchronous response.

[0082] S6. Based on the combination of the rate of change and the direction of change of morphological response features within the earliest detectable change window, an organoid response urgency classification to drugs is established, and the acquisition time density of subsequent morphological images is adjusted according to the response urgency classification.

[0083] In this embodiment, after identifying the earliest detectable change window, the urgency grading of organoid response to the drug is established based on the combination of the rate and direction of change of morphological response features within that window. Specifically:

[0084] The rate of change of morphological response features within the earliest detectable change window is analyzed. In this embodiment, the amplitude of change of morphological features within a unit time interval is compared with the inherent fluctuation amplitude of the morphological drift baseline. When the amplitude of change of morphological features significantly exceeds the inherent fluctuation range of the baseline, it is determined to be a high rate of change; when the amplitude of change is close to the inherent fluctuation range of the baseline, it is determined to be a low rate of change. It should be noted that "significantly exceeding" generally means that the amplitude of change reaches or exceeds the upper limit of the inherent fluctuation range of the baseline by several times; "close to" generally means that the amplitude of change is at a level close to the upper limit of the inherent fluctuation range of the baseline.

[0085] The stability of the change direction of morphological response features within the earliest detectable change window is analyzed. In this embodiment, the change direction of morphological features is tracked within multiple consecutive time points. When the change direction remains consistent across multiple consecutive time points (e.g., the contour roughness continuously increases or the cavity diameter continuously decreases), it is determined to be stable; when the change direction fluctuates between different time points (e.g., the roughness increases and then decreases, then increases again), it is determined to be unstable.

[0086] Based on the combination of change rate and change direction, an urgency grading of organoid response to drugs is established. In this embodiment, a high urgency grading is established when the change rate is high and the change direction is stable; a low urgency grading is established when the change rate is low and the change direction is stable, or when the change rate and direction are in an intermediate state.

[0087] Based on the response urgency level, the acquisition time density of subsequent morphological images is adjusted. In this embodiment, when the response urgency level is high, the acquisition time density of subsequent morphological images is increased, for example, the fixed interval is shortened from every 6 hours to every 2 hours, to capture rapidly evolving drug response details; when the response urgency level is low, the acquisition time density of subsequent morphological images is maintained or decreased, for example, maintaining the fixed interval of every 6 hours or increasing it to every 12 hours, to avoid resource waste caused by over-observation. It should be noted that the above adjustment of the increased or decreased density should be set with a minimum maintenance duration (e.g., at least two observation periods) to avoid frequent changes in the observation strategy due to short-term fluctuations.

[0088] Example 2: According to Figure 3 As shown, this embodiment provides an organoid intelligent analysis system based on deep learning, including:

[0089] The baseline establishment module is used to acquire morphological image sequences of organoids at fixed time intervals before drug action, identify the rhythm of morphological changes and establish morphological drift sub-baselines by phase, and infer the rhythmic continuity of unobserved phases to obtain the morphological drift baseline.

[0090] The temporal image acquisition module is used to continuously acquire morphological image sequences of organoids at fixed time intervals after drug action, and determine the effective contact time of each well based on the spatial position difference of the organoids in the well plate.

[0091] The individualized alignment module is used to compare the morphological features at each time point with the morphological drift baseline, and to establish individualized alignment sequences by tracking multiple independent organoids within the same pore location.

[0092] The continuous deviation identification module is used to identify the continuous change range of morphological features when they first deviate from the morphological drift baseline, distinguish between instantaneous fluctuations and continuous deviations, and trace back the deviation precursor features that are immediately adjacent to a single time point.

[0093] The window determination module is used to determine the start time of the first time that is identified as a persistent deviation interval as the earliest detectable change window of the drug response;

[0094] The early discrimination module is used to perform early qualitative discrimination based on morphological response features within the earliest detectable change window, and to analyze the spatial unfolding order to determine the type of drug action.

[0095] And an observation adjustment module, used to establish a response urgency level based on the combination of the rate of change and the direction of change of morphological response features within the earliest detectable change window, and to adjust the acquisition time density of subsequent morphological images according to the response urgency level.

[0096] In this embodiment, the functions and internal implementation logic of each of the above modules are as follows:

[0097] The baseline establishment module includes: a rhythm recognition unit, used to extract the periodic contraction and relaxation features of the organoid's outer contour through a contour tracking algorithm to identify the rhythm of morphological changes; a sub-baseline construction unit, used to statistically analyze the amplitude of contour fluctuations and the inherent fluctuation range of cavity diameter in each phase interval to establish morphological drift sub-baselines; and an inference unit, used to infer the continuation of morphological drift sub-baselines in unobserved phase intervals based on the period duration ratio or the gradual slope of the fluctuation boundary of the observed phase intervals.

[0098] The time-series image acquisition module includes: a liquid surface coverage recognition unit, which is used to automatically identify the expansion process of the coverage area of ​​drug droplets in the well plate through bright field image sequences; and a contact time recording unit, which is used to record the time corresponding to the frame image as the effective contact time when the edge of the drug liquid surface completely covers the area where the organoid is located.

[0099] The continuous deviation identification module includes: a same-direction change detection unit, used to compare whether the change direction of morphological features at adjacent time nodes is consistent; a reversal detection unit, used to determine whether a reverse change occurs at subsequent time nodes; and a precursor backtracking unit, used to backtrack to the immediately preceding observation node after confirming continuous deviation and detect precursor features of deviation.

[0100] The observation adjustment module includes: a rate comparison unit, used to compare the magnitude of morphological feature changes within a unit time interval with the inherent fluctuation magnitude of the morphological drift baseline; a direction stability tracking unit, used to track whether the direction of morphological feature changes remains consistent across multiple consecutive time points; and a hierarchical output unit, used to output a response urgency level based on a combination of change rate and change direction, and generate adjustment instructions for the temporal density of subsequent image acquisition.

[0101] In this embodiment, the data flow of each of the above modules is as follows:

[0102] The baseline establishment module acquires organoid morphological image sequences before drug administration, identifies the rhythm of morphological changes, and outputs the morphological drift baseline to the individualized comparison module. The time-series image acquisition module acquires time-series image sequences after drug administration and outputs them to the individualized comparison module. The individualized comparison module compares the morphological features at each time point with the morphological drift baseline and outputs the individualized comparison sequence to the persistent deviation identification module. The persistent deviation identification module identifies persistent deviation intervals and deviation precursor features and outputs them to the window determination module. The window determination module determines the earliest detectable change window and outputs it to the early discrimination module and the observation adjustment module. The early discrimination module performs early qualitative discrimination based on the morphological response features within the window and outputs the drug action type discrimination result. The observation adjustment module establishes a response urgency grade based on the response features within the window, adjusts the time density of subsequent image acquisition, and feeds it back to the time-series image acquisition module.

[0103] Example 3: This example provides experimental verification data for an organoid intelligent analysis method based on deep learning to demonstrate the technical effectiveness of the present invention.

[0104] Taking cholangiocarcinoma organoids as an example, they were cultured in standard 96-well plates, with each well containing 3 to 5 independent organoids. Before drug administration, 48 hours of bright-field morphological image sequences were continuously acquired at fixed time intervals of 6 hours. The inherent morphological change rhythm cycle of the organoids was identified by the baseline establishment module as approximately 8 hours, and divided into three phase intervals: expansion phase, stable phase, and contraction phase. Morphological drift sub-baselines for each phase were established.

[0105] After drug application, the effective contact time was defined as the moment when the drug liquid level completely covered the organoid region in each well (approximately 5 minutes for edge wells and approximately 12 minutes for center wells). From the effective contact time, bright-field morphological image sequences were acquired every 6 hours.

[0106] Individualized alignment sequences were established for five independent organoids within the same pore location. Six hours after the effective contact time, the contour roughness of three of these organoids showed an increasing trend; twelve hours after the effective contact time, the contour roughness of these three organoids continued to increase; and eighteen hours after the effective contact time, the contour roughness of these three organoids still showed an increasing trend, and the cavity diameter began to decrease, with the same trend of change not reversing. This met the criteria for confirming continuous deviation at three consecutive time points, and the continuous deviation interval was determined to have started at the 6th hour. Looking back at the immediately preceding observation node (hour 0 node), it was found that the contour roughness of these three organoids at hour 0 node had already shown a slight increasing tendency, but the amplitude was still within the inherent fluctuation range of the stable phase, indicating the presence of deviation precursor characteristics. Hour 0 node was then used as a preliminary extension point for the earliest detectable change window. It should be noted that in this embodiment, the hour 0 node was not included in the window for conservative judgment, but in scenarios requiring the capture of very early responses, this preliminary extension point can be included in the temporal range of the earliest detectable change window.

[0107] Analysis of the spatial unfolding sequence of the three organoids within the earliest detectable change window (6 to 18 hours) revealed that the peripheral region first showed increased edge roughness, followed by changes in the cavity structure in the central region, while the inner region began to respond later in the window. Based on this, an early qualitative assessment concluded that the three organoids were sensitive to the drug, and that the action type was membrane permeability dependent.

[0108] Meanwhile, the rate of change of these three organoids within the window significantly exceeded the baseline's inherent fluctuation range, and the direction of change was stable. The urgency level of the response was high, and the observation adjustment module increased the subsequent image acquisition time density from every 6 hours to every 2 hours.

[0109] As a control, traditional endpoint testing determines the drug sensitivity of these three organoids by fluorescent staining 72 hours after drug treatment, which is consistent with the early prediction results of this invention from the 6th to the 18th hour. This demonstrates that this invention can achieve reliable drug sensitivity prediction in the early stages of drug treatment, significantly shortening the detection cycle.

[0110] The specific implementation of each step in the above embodiments has been described in detail in Embodiment 1. The hardware implementation of each module can be carried out by a computing device with image acquisition, storage and processing functions, and its software implementation can be completed by computer program instructions. Those skilled in the art can implement it by modular programming according to the method logic in Embodiment 1, which will not be repeated here.

[0111] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A deep learning-based intelligent analysis method for organoids, characterized in that, Includes the following steps: Before drug administration, morphological image sequences of organoids were acquired at fixed time intervals to establish a morphological drift baseline of organoids in spontaneous culture. After drug administration, morphological image sequences of organoids are continuously acquired at the same fixed time intervals, and the morphological features at each time point are compared with the morphological drift baseline. Identify the continuous range of morphological features that first deviate from the morphological drift baseline, and use the starting time of this continuous range of changes as the earliest detectable change window of the drug response; Based on the morphological response characteristics within the earliest detectable change window, the drug sensitivity of organoids can be qualitatively determined at an early stage.

2. The organoid intelligent analysis method based on deep learning according to claim 1, characterized in that, The establishment of the morphological drift baseline of organoids in spontaneous culture includes: Identify the morphological change rhythms of organoids in spontaneous culture and divide the morphological change rhythms into different phase intervals; A morphological drift sub-baseline is established for each phase interval. The morphological drift sub-baseline is used to record the inherent morphological fluctuation range of organoids in the corresponding phase interval.

3. The organoid intelligent analysis method based on deep learning according to claim 2, characterized in that, The establishment of the morphological drift baseline also includes: When the observation duration before drug action does not cover the complete morphological change rhythm, based on the established morphological drift sub-baseline of some phase intervals, the rhythm continuity of the morphological drift sub-baseline of the unobserved phase interval is inferred to obtain the inferred morphological drift baseline that covers the complete morphological change rhythm.

4. The organoid intelligent analysis method based on deep learning according to claim 1, characterized in that, The sequence of morphological images of organoids continuously acquired after drug administration includes: Based on the spatial differences of organoids in the well plate, the effective contact time for substantial contact between organoids and drugs in each well position was determined. Starting from the effective contact time of each pore, morphological image sequences of organoids within that pore are acquired at fixed time intervals.

5. The organoid intelligent analysis method based on deep learning according to claim 1, characterized in that, The comparison of morphological features at each time point with the morphological drift baseline includes: In cases where multiple organoids exist within the same pore site, the morphological changes of each independent organoid at each time point are tracked separately. Establish individualized comparison sequences between the morphological feature changes of each independent organoid and the morphological drift baseline to avoid confusion between the morphological feature changes of multiple organoids.

6. The organoid intelligent analysis method based on deep learning according to claim 1, characterized in that, The continuous range of changes in which the identified morphological features first deviate from the morphological drift baseline includes: Distinguish between the instantaneous fluctuation range and the continuous deviation range of morphological features; when morphological features show the same trend of change in multiple adjacent time nodes, and the same trend of change in the same direction does not reverse in subsequent time nodes, the range is determined to be the continuous deviation range. The starting time of the first point that is determined to be a continuous deviation from the interval is taken as the starting time of the earliest detectable change window.

7. The organoid intelligent analysis method based on deep learning according to claim 6, characterized in that, The identification of the continuous deviation range also includes: After determining the starting time of the continuous deviation interval, the single time node immediately preceding the starting time is traced back to determine whether the single time node has shown deviation precursor features. The deviation precursor features are manifested as morphological features that do not exceed the inherent fluctuation range of the morphological drift baseline and do not meet the continuous deviation confirmation conditions of the multiple adjacent time nodes, but have shown a tendency to change in the same direction as the continuous deviation interval. When the aforementioned early warning features are present, the single time node is used as a preliminary extension point for the earliest detectable change window.

8. The organoid intelligent analysis method based on deep learning according to claim 1, characterized in that, The early qualitative discrimination based on morphological response features within the earliest detectable change window includes: By analyzing the spatial unfolding order of morphological response features within the earliest detectable change window, it can be determined whether the drug effect gradually penetrates from the peripheral region to the inner region of the organoid or occurs synchronously within the entire organoid. When the infiltration is determined to be a gradual process from the periphery to the interior, it is qualitatively classified as membrane permeability-dependent drug sensitivity; when the infiltration is determined to be a synchronous process occurring throughout the body, it is qualitatively classified as metabolic pathway interference-type drug sensitivity.

9. The organoid intelligent analysis method based on deep learning according to claim 1, characterized in that, The method further includes: After identifying the earliest detectable change window, an organoid response urgency classification to the drug is established based on the combination of the rate and direction of change of morphological response features within the window. Based on the response urgency level, the acquisition time density of subsequent morphological images is adjusted; when the response urgency level is high, the acquisition time density of subsequent morphological images is increased; when the response urgency level is low, the acquisition time density of subsequent morphological images is maintained or reduced.

10. A deep learning-based organoid intelligent analysis system, characterized in that, include: The baseline establishment module is used to acquire morphological image sequences of organoids at fixed time intervals before drug action, identify the rhythm of morphological changes and establish morphological drift sub-baselines by phase, and infer the rhythmic continuity of unobserved phases to obtain the morphological drift baseline. The temporal image acquisition module is used to continuously acquire morphological image sequences of organoids at fixed time intervals after drug action, and determine the effective contact time of each well based on the spatial position difference of the organoids in the well plate. The individualized comparison module is used to compare the morphological features at each time point with the morphological drift baseline, and to establish individualized comparison sequences by tracking multiple independent organoids within the same pore location. The continuous deviation recognition module is used to identify the continuous change range of the morphological feature when it first deviates from the morphological drift baseline, distinguish between instantaneous fluctuations and continuous deviations, and trace back the deviation precursor features that are immediately adjacent to a single time point. The window determination module is used to determine the start time of the first time that is identified as a persistent deviation interval as the earliest detectable change window in the drug response; The early discrimination module is used to perform early qualitative discrimination based on the morphological response features within the earliest detectable change window, and to analyze the spatial unfolding order to determine the drug action type. And an observation adjustment module, used to establish a response urgency level based on the combination of the rate of change and the direction of change of the morphological response features within the earliest detectable change window, and to adjust the acquisition time density of subsequent morphological images according to the response urgency level.

Images