EP-4738254-A1 - METHOD FOR IMAGE REGISTRATION, METHOD FOR DETECTING AND CALLING BASES FROM IMAGES, AND COMPUTER DEVICE

EP4738254A1EP 4738254 A1EP4738254 A1EP 4738254A1EP-4738254-A1

Abstract

The present application discloses methods for registering an image, methods for detection of an image to call bases, and a computer device. The image comes from a system that enables sequencing through microscopic imaging of patterned surfaces. A surface region reflected by the image is a patterned surface, and comprises multiple adjacent marker zones and reaction zones regularly arranged. Each marker zone comprises at least two non-parallel extending directions, and multiple reactive sites are regularly arranged in each direction. Coarse registration is performed by using global features such as a distribution of marker zones on the surface region or a distribution of fiducial points within marker zones, and then fine registration is performed by using local features like a distribution of spots of marker zones, such that positions of the fiducial points can be accurately and reliably determined, thereby achieving high-precision registration of an image with a fiducial template.

Inventors

ZENG, Zhen
Lan, Chongzhou
LI, LINSEN
JIN, Huan
WAN, Xinchun
DENG, Chengwei

Assignees

GeneMind Biosciences Company Limited

Dates

Publication Date: 20260506
Application Date: 20251030

Claims (15)

A method for image registration, wherein the image is obtained from a detection system for sequencing based on patterned-surface microscopic imaging, the method comprising: (S10) performing coarse registration of the image on the basis of a fiducial template, comprising: determining positions of a plurality of fiducial points in the image on the basis of reference positions corresponding to the fiducial points in the fiducial template, establishing one or more fitting relationships based on the positions of the plurality of fiducial points, and updating the positions of the fiducial points in the image based on the fitting relationship; wherein the fiducial template comprises a reference coordinate system and reflects the same surface region as the image; the surface region comprises a plurality of adjacent marker zones and reaction zones, the marker zone being provided with a plurality of discrete reactive sites regularly distributed in two or more non-parallel directions, and an intersection of the directions being located within the marker zone; the marker zone or any direction thereof, together with the surrounding reaction zones, exhibits a specific intensity variation feature on the corresponding detection signal curve; the fiducial point is located at the intersection of the directions of the marker zone and is positioned according to the intensity variation feature; at least a portion of the reactive sites appear as bright spots in the image; and (S20) performing fine registration of the image based on the coarse registration result of (S10) and the bright spots, comprising: determining the distribution of the bright spots in each of the directions of the marker zone, determining a respective fitting relationship for each of the directions based on the distribution of the bright spots, and further updating the fiducial point positions of the image on the basis of the fitting relationships; and determining coordinates of other areas or positions of the image based on the updated fiducial point positions, to register the image with the fiducial template.
The method according to claim 1, wherein the method comprises at least one of the following features (a) to (d): (a) the size of the marker zones does not exceed 10% of the surface region; (b) at least a portion of the reaction zones are discrete, and a marker zone is present between two adjacent reaction zones; (c) at least a portion of the marker zones are discrete, and a reaction zone is present between two adjacent marker zones; and/or (d) each reaction zone comprises a plurality of regularly distributed discrete reactive sites, and at least a portion of the reactive sites in the reaction zone appear as bright spots in the image; optionally wherein the reactive sites in the reaction zone and the marker zone are each independently presented as depressions, and/or the reactive sites in the reaction zones are more densely distributed than the reactive sites in the marker zones, and/or the size of individual reactive sites in the reaction zone is smaller than that of individual reactive sites in the marker zone.
The method according to any of claim 1 or 2, wherein (S10) comprises: (a) calculating reference positions corresponding to the fiducial points in the fiducial template; (b) determining a detection signal curve of a predetermined region in the image where the reference position is located, wherein the predetermined region comprises the intersection of the directions of the marker zones and a portion of reaction zones adjacent to the directions; and (c) identifying the intensity variation feature of the detection signal curve, and determining the position of the fiducial point based on the position corresponding to the intensity variation feature.
The method according to claim 3, wherein the plurality of marker zones are regularly distributed in the surface region along two or more non-parallel linear directions, and (S10) further comprises: (d1) establishing a fitting relationship in each corresponding direction by using position components of the fiducial points in the respective directions; (e1) classifying the fiducial points into two categories according to a preset condition, and updating the fitting relationship of each direction based on position components of the fiducial points belonging to a designated category, wherein the preset condition is related to a distance between the fiducial point and the fitting relationship; and (f1) correcting the position components of the fiducial points in other category on the basis of the updated fitting relationship.
The method according to claim 4, wherein the preset condition comprises a plurality of distance thresholds that progressively converge from large to small, including a maximum distance threshold D1 and a second-largest distance threshold D2, and (e1) comprises, (e11) classifying the fiducial points into two categories according to the distance threshold D1, comprising categorizing fiducial points whose distance to the fitting relationship is less than or equal to D1 as the designated category, and categorizing fiducial points whose distance to the fitting relationship is greater than D1 as the other category; (e12) updating the fitting relationship in the corresponding direction by using position components of the fiducial points in the designated category obtained from (e11) in each of the directions; (e13) classifying the fiducial points in the designated category obtained from (e11) into two categories according to the distance threshold D2, comprising categorizing the fiducial points whose distance to the updated fitting relationship obtained from (e12) is less than or equal to D2 as the designated category, and categorizing the fiducial points whose distance to the updated fitting relationship obtained from (e12) is greater than D2 as the other category; and (e14) updating the fitting relationship in the corresponding direction by using position components of the fiducial points in the designated category obtained from (e13) in each of the directions.
The method according to claim 3, wherein the plurality of marker zones are distributed in the surface region along a regular curve, and (S10) further comprises: (d2) establishing a corresponding curve fitting relationship by using the positions of the fiducial points; (e2) classifying the fiducial points into two categories according to a preset condition, and updating the curve fitting relationship on the basis of the positions of fiducial points in a designated category, wherein the preset condition is related to a distance between the fiducial point and the curve fitting relationship; and (f2) correcting positions of fiducial points in the other category on the basis of the updated curve fitting relationship.
The method according to claim 6, wherein the preset condition comprises a plurality of distance thresholds that progressively converge from large to small, including a maximum distance threshold D1 and a second-largest distance threshold D2, and (e2) comprises: (e21) classifying the fiducial points into two categories according to the distance threshold D1, comprising categorizing the fiducial points whose distance to the curve fitting relationship is less than or equal to D1 as the designated category, and categorizing the fiducial points whose distance to the curve fitting relationship is greater than D1 as the other category; (e22) updating the curve fitting relationship according to positions of the fiducial points in the designated category obtained from (e21); (e23) classifying the fiducial points in the designated category obtained from (e21) into two categories according to the distance threshold D2, including categorizing the fiducial points whose distance to the updated curve fitting relationship obtained from (e22) is less than or equal to D2 as the designated category, and categorizing those whose distance to the updated curve fitting relationship is greater than D2 as the other category and (e24) updating the curve fitting relationship according to positions of the fiducial points in the designated category obtained from (e23).
The method according to any of claims 1 to 7, wherein (S20) comprises: (S22) determining the bright spot distribution of each direction of the marker zones on the basis of the positions of the fiducial points obtained from (S10) and the fiducial template, comprising identifying bright spots in multiple regions of each direction; (S24) determining a fitting relationship for each direction according to the bright spot distribution of the respective direction, and filtering the bright spots to optimize the fitting relationship; (S26) further updating the fiducial point positions of the image based on the fitting relationships obtained from (S24), and determining coordinates of other regions or positions of the image based on the further updated fiducial point positions, to register the image with the fiducial template.
The method according to claim 8, wherein (S22) comprises: determining, by using position components of the fiducial points obtained from (S10) in the respective direction and the relative positional relationship between the fiducial points and the bright spots included in the fiducial template, a pixel range k1 × k2 where each of the bright spots in multiple regions of each direction is possibly located, wherein k1 and k2 are each independently an odd number greater than 1; and identifying a bright spot in the pixel range k1 × k2 where a bright spot is possibly located based on a difference in pixel intensity between a central pixel and surrounding pixels.
The method according to claim 8 or 9, wherein the preset condition in (S10) is a first preset condition, and (S24) comprises filtering the bright spots according to a second preset condition and updating the fitting relationship of the respective direction on the basis of the bright spots retained after filtering, wherein the second preset condition is related to a distance between the bright spot and the fitting relationship.
The method according to claim 10, wherein the second preset condition comprises a plurality of distance thresholds that progressively converge from large to small, including a maximum distance threshold D21 and a second-largest distance threshold D22, and (S24) comprises: (S241) filtering the bright spots according to the distance threshold D21, comprising retaining bright spots whose distance to the fitting relationship is less than or equal to D21, and optionally discarding bright spots whose distance to the fitting relationship is greater than the distance threshold D21; (S243) updating the fitting relationship by using bright spots retained after (S241); (S245) filtering the bright spots retained after (S241) according to the distance threshold D22, comprising retaining bright spots whose distance to the updated fitting relationship obtained after (S243)is less than or equal to the distance threshold D22, and optionally discarding bright spots whose distance to the updated fitting relationship is greater than the distance threshold D22; and (S247) updating the fitting relationship based on the bright spots retained after (S245).
The method according to any of claims 3 to 11, wherein at least one of the following steps or processes (i) to (iv) is performed in parallel: (i) the process of determining each of the positions of the plurality of fiducial points in the image in (S10) of claim 3; (ii) the process of updating respective fitting relationship of each direction in (S10) of claim 4 or 5; (iii) the process of optimizing respective fitting relationships of each direction in (S20) of claim 8; (iv) the process of determining each of the coordinates of the other regions or positions of the image on the basis of the further updated fiducial point positions in (S26) of (S20) of claim 8.
A method for detecting and calling bases from an image, wherein the image is obtained from a detection system for sequencing based on microscopic imaging of a patterned surface, and wherein a surface region reflected by the image comprises a plurality of adjacent marker zones and reaction zones, the method comprising: determining signal intensity values at positions corresponding to chemical features in the reaction zones of the image, wherein the image has been registered by using the method according to any of claims 1 to 12; and calling the type of base introduced at the positions on the basis of the signal intensity values; wherein optionally the signal intensity values are corrected signal intensity values.
A computer device, comprising: a memory configured to store an application and data generated by execution of the application; at least one processor configured to execute the application to implement the method for image registration according to any of claims 1 to 12, and/or the method for detecting and calling bases from an image according to claim 13.
The computer device according to claim 14, wherein the processor comprises at least one set of processing units, and each processing unit in the set is capable of parallel execution, the set of processing units being configured to perform at least one of the following steps or processes (i)-(iv): (i) determining positions of the plurality of fiducial points in the image when performing (S10) of the method for image registration according to claim 3; (ii) updating respective fitting relationships of respective direction when performing (S10) of the method for image registration according to claim 4 or 5; (iii) optimizing respective fitting relationships of respective direction when performing (S20) of the method for image registration according to claim 8; (iv) determining coordinates of other regions or positions of the image according to the further updated fiducial point positions when performing (S26) in (S20) of the method for image registration according to claim 8.

Description

The present application claims priority under the Paris Convention to Chinese patent application Nos. 202411549366.4 and 202511423061.3, filed with the China National Intellectual Property Administration on October 31, 2024, and September 29, 2025, respectively. The entire contents of each of the above applications are incorporated herein by reference in their entirety. TECHNICAL FIELD The present application relates to the field of image processing technology, and in particular to a method for registering an image, a method for calling bases by performing detection on an image, and a computer device. BACKGROUND The subject matter discussed in this section should not be construed as prior art merely by virtue of its inclusion in this section. Similarly, technical problems mentioned in this section or associated with the subject matter provided as background art should not be assumed to have been previously recognized in the prior art. The subject matter in this section merely represents different methods, which themselves may also correspond to specific implementations of the technical solutions included in the claims. In the related art, so-called sequencing generally refers to a process of determining a primary structure or sequence of a biological polymer, including a nucleic acid such as DNA and RNA, comprising determining the order of nucleotide bases (adenine A, guanine G, thymine T/uracil U, and cytosine C) of a given nucleic acid fragment. Such methods generally include calling bases at one or more positions in a nucleic acid, i.e., performing base calling, to determine the sequence of at least a segment or moiety of the nucleic acid molecule. A signal and/or signal intensity variation directly or indirectly generated by binding or attachment of a nucleotide/base to a specific position of a nucleic acid molecule to be tested (template) can indicate the type of a base at that position on the nucleic acid molecule. For example, different fluorescent marker molecules may be used to call different incorporated or attached nucleotides/bases. Said nucleotide/base binding or attaching to a specific position of the nucleic acid molecule to be tested is also called nucleotide/base incorporation into the nucleic acid molecule to be tested or base extension, which may be achieved, for example, by polymerization, attachment, hybridization, etc. Specifically, in a detection system for nucleic acid molecule sequencing based on surface microscopy, a nucleic acid molecule is usually attached to a solid surface. A solid surface containing a chamber that can hold a solution is often called a chip or flow cell. According to a layout design or production process of the solid surface or a distribution of the nucleic acid molecule to be tested on the solid surface, the solid surface may be classified as a patterned surface (or regular array, patterned surface) or a random surface (random array, random surface). A process of acquiring and detecting signals from any surface array, such as capturing images, usually includes moving related hardware such as moving an objective lens and/or moving a solid surface, so as to acquire images of the same object (nucleic acid molecules in the same surface region/field of view) after a relevant biochemical reaction at different moments, and then processing the plurality of images captured at different moments, including signals at positions of corresponding chemical features in the images (such as a nucleic acid molecule to be tested that has undergone an extension reaction), to determine at least a part of sequence information of the nucleic acid molecule to be tested in the field of view. Relatively speaking, it is generally believed that image registration on patterned surfaces is simpler than on random surfaces, because regular patterns have been introduced into the former during a chip design or preparation stage. This means that a spatial layout of relevant regions or sites is known before imaging, and may be used as a reference image or template or reference frame (or global coordinate system). Therefore, by identifying pattern features on the image and aligning the image with the reference image accordingly, one or more images of the surface field of view can be mapped or unified to the same coordinate system as the reference image to achieve registration. However, during an actual image acquisition process, since movement of related hardware has a set precision and application or reception of a force may cause a position offset or shape change of a related mechanical structure, and/or a related sequencing biochemical reaction process may involve a temperature increase or decrease, etc. affecting the shape or surface properties of the solid surface, or the number, form or position of some nucleic acid molecules to be tested on the surface, it is inevitable that at least a portion of the nucleic acid molecules to be tested in the same field of view will be shifted or inconsiste