Method and device for identifying and selecting cavities of a cavity array for optical analysis of biochemical reactions in the cavities, in particular comprising nucleic acid amplifications for microfluidic detection of pathogens

Abstract

The invention relates to a method (500) for identifying and selecting cavities of a cavity array (110) for optical analysis of, in particular, microfluidic biochemical reactions in the cavities (112) using a processor (1300) of a device (1000). The invention further relates to a device (1000) configured to carry out the method (500), in particular an analysis instrument for performing biochemical reactions.

Description

[0001] R. 418023

[0002] - 1 -

[0003] Description

[0004] title

[0005] Method and apparatus for identifying and selecting cavities of a cavity array for optical analysis of biochemical reactions in the cavities, in particular comprising nucleic acid amplifications for the microfluidic detection of pathogens

[0006] State of the art

[0007] The detection of pathogens in a biological sample, such as blood or sputum, can be achieved by identifying fragments of the pathogens' nucleic acids within the sample. This is done using nucleic acid amplification, a process in which nucleic acids are amplified section by section, to specifically search for these fragments. Such methods can be implemented at the point of care and as immediate, near-patient diagnostics using microfluidic systems. In this process, a patient sample is introduced into a microfluidic cartridge, which is then controlled within an analyzer to perform nucleic acid amplification inside the cartridge, as described, for example, in documents DE 102016 222075 A1 and DE 102016 222072 A1.

[0008] Documents DE 102018204624 A1 and DE 102018210069 A1 describe a silicon chip as part of a cartridge, which has several recesses, also referred to as cavities or wells, and can therefore also be called a cavity array. Different nucleic acid amplifications can take place in parallel in these cavities, thus enabling spatial multiplex detection of various pathogens. Due to the high thermal conductivity of silicon, short PCR cycles can be realized in the cavities, particularly via a heat source applied to the chip. Regarding the use of optical probes, see R. 418023.

[0009] - 2 -

[0010] The reactions in the cavities can be read out for real-time analysis. Fluorescence probes, whose excitation requires illumination of the chip, are particularly suitable for this purpose. R. 418023

[0011] - 3 -

[0012] Disclosure of the invention

[0013] Advantages of the invention

[0014] Against this background, the invention relates to a method for identifying and selecting cavities of a cavity array for optical analysis of biochemical reactions in the cavities, in particular for microfluidic detection of pathogens. The biochemical reactions, which occur particularly in microfluidic dimensions, can include, in particular, nucleic acid amplifications, preferably polymerase chain reactions (PCR) or isothermal amplifications. Alternatively, the biochemical reactions can also (only) comprise a hybridization reaction of nucleic acid fragments with optical probes. The optical analysis can, in particular, be performed in real time and include parallel analyses of several cavities, for example, monitoring the parallel execution of quantitative real-time PCR in the cavities.Optical analysis utilizes, in particular, detected luminescence radiation, especially fluorescence radiation, which is emitted by optical probes in the cavities when these probes are optically, chemically, or electrically excited, depending on their nature.

[0015] The presented method advantageously enables the robust identification of cavities captured in an image of the cavity array, starting from an image acquired with an optical sensor, particularly a camera. Since different biochemical reactions can occur in the cavities, especially nucleic acid amplifications of different nucleic acid segments or hybridizations of different entities due to different capture molecules, for example, for the parallel detection of various pathogens, a correct assignment of the cavities to their representations in the captured image is thus possible. The method therefore advantageously allows the cavities to be clearly distinguished and the relevant cavities to be selected for optical analysis. Furthermore, additional images with the same optical sensor settings and arrangement relative to the cavity array can then be advantageously acquired for the R. 418023

[0016] - 4 -

[0017] Optical analysis can be performed without having to locate or identify the cavities again.

[0018] The method uses an input image, wherein the input image represents an illuminated first region of the cavity array, the first region comprising several cavities and a reference mark. The generation of the input image can be part of advantageous embodiments of the method. This generation can include illuminating a first region of the cavity array with a light source, the first region comprising several cavities and a reference mark. In other words, the illuminated first region on the cavity array, hereinafter also referred to as the illumination spot, is selected by adjusting the light source, in particular by choosing the size of the light cone, and by positioning the cavity array relative to the light source such that the reference mark and a predetermined minimum number of cavities are located in the first region. The light source can, in particular, be part of a device according to the invention.The device can be configured to accommodate the cavity array, in particular by including a microfluidic cartridge containing the cavity array, at least partially within the device. The placement of the cavity array relative to the light source is generally predetermined or fixed, but manufacturing tolerances of the cavity array, the light source, and / or the device can lead to uncertainties in the exact position of the cavities. The presented method advantageously allows these uncertainties to be taken into account and at least partially compensated.

[0019] The reference mark is, in particular, a structure on the cavity array, wherein the reference mark differs from the cavities by a different optical feature, especially due to a shape different from the cavities. For example, the reference mark can be designed as a recess or protrusion, preferably in the shape of a cross, preferably at a location between the cavities or instead of a cavity. The cavities can be arranged in a regular pattern, at least in some areas. Alternatively or additionally, a cavity can be missing, and the resulting defect can be used as the reference mark, especially due to a break in the pattern caused by it. R. 418023

[0020] - 5 -

[0021] The first area can be detected with an optical sensor to create the input image. For this purpose, the optical sensor, preferably a camera, is arranged and configured with respect to the light source and the predetermined placement of the cavity array in such a way as to detect the first area illuminated by the light source.

[0022] According to the method, the reference mark and the cavities are localized in the received input image, with the cavities being identified taking into account the localized reference mark. In particular, the cavities are thus localized with respect to the localized reference mark. Furthermore, the first illuminated area on the cavity array in the input image is determined. This determination can, in particular, include determining the position and preferably the intensity distribution of the first illuminated area. Subsequently, one or more of the localized cavities are selected for optical analysis by comparison with the determined illuminated first area.

[0023] The method is preferably fully automated using a suitably configured processor, wherein the processor receives the provided input image and executes the subsequent steps of the method, in particular the localization of the reference mark and the localization and identification of the cavities, the determination of the illuminated first area, the selection of the localized cavities or relevant measurement areas, and preferably steps of the further developments of the method described below. The method can therefore be, in particular, a computer-implemented method.

[0024] The invention therefore also relates to a computer program comprising instructions which, when the program is executed by a computer, cause it to execute the method.

[0025] The invention also relates to a device, in particular an analysis device, which is configured to execute the steps of the method, in particular the computer program. For this purpose, the computer program is preferably stored in a memory of the device, and the device comprises at least one processor as a control unit for executing the computer program. R. 418023

[0026] - 6 -

[0027] The invention further relates to a device, in particular an analyzer for performing an optical analysis of biochemical reactions, especially nucleic acid amplifications. This device comprises a receiving area for receiving a cavity array, in particular a microfluidic cartridge with the cavity array, a light source for illuminating at least one region of the cavity array, an optical sensor for detecting the region, and at least one processor as a control unit. The device, in particular the processor, is configured to execute the method according to the invention and for this purpose comprises a memory in which the computer program for the method is stored. The computer program can include instructions that cause the device to execute the method, in particular the steps of the method.

[0028] Before locating the reference mark, the input image is preferably normalized. Normalization advantageously helps to compensate for tolerances in the overall illumination intensity, variations in illumination, particularly local inhomogeneous illumination, and differing reflections due to tolerances in the positioning of the cavity array, especially the cartridge, or the light source. Furthermore, normalization can also support the suppression of local image disturbances such as pixel defects or reflections due to dust on the cavity array, the cartridge, or the optics of the optical sensor. This normalization can include histogram transformation, filtering, background subtraction, and / or morphological operations on the input image.

[0029] To locate the reference mark, a stencil image is preferably compared with different areas of the input image, thereby creating a correlation map. The spatially resolved correlation values ​​in the correlation map represent a measure of the correspondence between the light intensity desired according to the stencil image and the light intensity actually detected. The comparison can be performed, in particular, by overlaying the stencil image onto the input image. (R. 418023)

[0030] - 7 -

[0031] The template image includes target zones and / or penalty zones for defining correlation values ​​in the correlation map. The target zones reward the presence of the desired light intensity, while the penalty zones penalize the presence of undesired light intensity. To account for tolerances, particularly scaling and rotation tolerances, at least some of the target zones are preferably larger and / or shaped differently than the cavities. The position of the reference mark can then be estimated from the correlation map by determining a maximum, and in particular an absolute, value of the correlation values. Such a correlation map essentially comprises several local maxima, the granularity of which results from the discrete movement of the template image over the input image. This results in several correlation values ​​located around and at a (local) maximum appearing as step-shaped peaks (correlation peaks) in two dimensions.

[0032] For subsequent processing steps, the input image can be cropped to a region around the defect, known as the cropped area, a process also called masking. Cropping the image generally improves performance and reduces the process's susceptibility to errors, as it allows for the removal of distracting optical artifacts outside the cavity area. Furthermore, it reduces the computational load on the processor, since fewer image values ​​need to be processed.

[0033] Localizing the cavities preferably includes estimating the distances between the cavities (pitch). With a regular arrangement of the cavities (apart from tolerances), the distances can advantageously be estimated using a Fourier transform of at least part of the input image, wherein the pitch is preferably determined by averaging over pitch value candidates estimated from the fundamental spatial frequencies in Fourier space, the pitch value candidates being derived from the amplitudes of the fundamental spatial frequencies.

[0034] The localization of the cavities preferably includes an estimation of the rotation of the input image relative to the optical sensor. This estimation R. 418023

[0035] - 8 -

[0036] This is preferably achieved by comparing the input image with a template image, particularly for creating a correlation map for the rotational correlation between areas of the input image and the template image. The determination of the local correlation values ​​in the correlation map can be carried out using target zones and penalty zones in the template image, similar to the localization of the reference mark. To facilitate the determination of the correlation map, the input image and the template image can be transformed into polar coordinates, which then allows for translational, rather than rotational, guidance of the template image over the input for the creation of the correlation map (translative correlation).

[0037] For locating the cavities, a layout of cavities on the cavity array can be used, wherein the layout comprises a basic arrangement of cavities, particularly with respect to the reference mark, on the surface of the cavity array. Preferably, the cavities on the cavity array are arranged in the form of a grid, i.e., a regular pattern. The layout thus represents a map for these cavities.

[0038] In an advantageous embodiment, the method can include the determination of a confidence value, also referred to as the overall confidence value, wherein the confidence value is derived from an individual confidence value for locating the reference mark and / or an individual confidence value for locating the cavities. This advantageously allows an estimation of the correctness of the localizations. The confidence value corresponds, for example, to the mean of the individual confidence values. Preferably, one or both of the individual confidence values ​​and / or the overall confidence value can be compared with predefined limit values ​​in order to trigger a measure, particularly by the processor, if one of the limit values ​​is undershot. The measure can, for example, include terminating the method or re-executing the method, either with the same input image or, preferably, with a new input image.For the acquisition of a new input image with the optical sensor, optical parameters can be changed compared to the acquisition of the original input image, in particular a changed, for example higher R. 418023.

[0039] - 9 -

[0040] Lighting intensity or altered, especially longer, exposure time due to the light source, color filtering, for example by inserting a physical filter between the light source and the cavity array.

[0041] Alternatively or additionally, the image area can also be changed, in particular enlarged or reduced, for example by adjusting an aperture in the device.

[0042] To determine, and in particular calculate, the individual confidence value for locating the reference mark, the ratio of the amplitude of the largest peak, i.e., the largest local maximum, to the amplitude of the second largest peak, i.e., the second largest local maximum, is preferably determined from the generated correlation map. The individual confidence value for locating the reference mark corresponds, for example, to this ratio or, preferably, to a value derived from this ratio, in particular a normalized value.

[0043] For example, normalization is carried out in such a way that the ratio corresponds to a value between 0 and 1, where the value 1 corresponds to the optimal case of a maximum largest correlation value and a minimum second largest correlation value, and thus to a high confidence of the estimate.

[0044] Determining the individual confidence value for cavity localization can preferably include determining an individual confidence value for estimating the distances between the cavities, i.e., the pitch, and / or determining an individual confidence value for estimating the rotation of the input image relative to the optical sensor. For determining, and in particular calculating, the individual confidence value for pitch estimation, the difference between the maximum and minimum amplitudes of the peaks belonging to each pitch value candidate is preferably calculated. The individual confidence value for pitch estimation then preferably corresponds to an average of these differences. Preferably, all amplitudes are normalized to a value between 0 and 1 before calculating the differences.For the determination, in particular the calculation of the individual confidence value for the estimate of the rotation, the difference between the largest and smallest correlation values ​​from the generated correlation map is preferably determined, since a correlation map generated in this way R. 418023.

[0045] - 10 -

[0046] It usually only exhibits one correlation peak. The individual confidence value for the rotation estimate corresponds, for example, to this difference value, whereby the correlation map is preferably normalized first. For example, normalization is carried out by normalizing the correlation values ​​contained in the correlation map to a range between 0 and 1 before calculating the difference value.

[0047] It is particularly advantageous if measurement areas relevant for optical analysis, also known as regions of interest (ROIs) of the individual cavities, are defined and used. This definition can be achieved, in particular, using the localized reference mark, the localized cavities, and a predefined layout, especially the layout of the cavity array described above. Preferably, the relevant measurement areas are corrected with respect to the light intensities captured in the input image, whereby the correction can, in particular, include a positional correction of the relevant measurement areas. The selection of the cavities for the optical analysis then particularly includes a selection of the defined relevant measurement areas or is replaced by them.

[0048] The selection of one or more localized cavities, in particular the relevant measurement areas of these cavities, can advantageously be carried out simply by selecting all localized cavities that are located within the defined illuminated first area. In a preferred embodiment, the selection is carried out by determining an illumination metric for the cavities, in particular for the relevant measurement areas. Determining the illumination metric comprises comparing the cavities, in particular the relevant measurement areas, with the position-defined first area and a brightness of the respective cavities or relevant measurement areas, which is captured, in particular, in the as-yet-unnormalized input image.

[0049] When selecting the localized cavities, especially the relevant measurement areas, for optical analysis, the expected shape, for example a rounded shape or a circular boundary, of the cavities, especially the relevant measurement areas, can also be taken into account. R. 418023

[0050] - 11 -

[0051] This allows cavities or relevant measurement areas to be advantageously disregarded during optical analysis, as these may deviate from the expected shape due to disturbances such as contamination.

[0052] Furthermore, the selection of the localized cavities, in particular the relevant measurement areas, can depend on the detection of fluidic anomalies. Preferably, the first area of ​​the cavity array is re-captured by the optical sensor after the cavities have been filled with a liquid, in order to obtain a second image. This second image is then compared with the captured input image to identify such optical anomalies, for example, caused by bubbles or foam, and to identify the affected cavities.

[0053] To exclude or disregard measurement areas from the analysis. This adjustment can preferably, and in particular exclusively, be performed in areas of the localized cavities within the defined illuminated first area, particularly in the relevant measurement areas, preferably preselected, so that a re-localization and selection of the cavities or relevant measurement areas is not necessary. The second image can be normalized before the adjustment, preferably in the same way as the input image.

[0054] According to an advantageous embodiment, a further input image can be received, which is compared with the captured input image and / or with the second image in order to verify the expected shape of the cavities, in particular the relevant measurement areas, and / or to detect fluidic anomalies, as described above. The comparison can preferably, and in particular exclusively, be performed in areas of the localized cavities within the defined illuminated first area, particularly in the relevant measurement areas, which were preferably preselected using the illumination metrics and determined by the method based on the input image.The reception or control of the device for recording the further input image can occur after a predetermined period of time, for example a few seconds or a few minutes, and / or after, in particular, refilling or rinsing of the cavity array with R. 418023.

[0055] - 12 -

[0056] Liquid is introduced, in particular by the processor controlling a pump in the device, especially the cartridge. The liquid can be, for example, a sealing fluid, especially an oil, which, for instance, covers the cavities previously filled with a sample liquid, especially an aqueous one. Alternatively, the liquid can be sample liquid for refilling or further refilling the cavities. This has the advantage that the optical or fluidic conditions in the cavities and on the cavity surface can be improved for optical analysis by the, in particular initial or subsequent, rinsing or refilling.Preferably, such flushing and / or filling, in particular initial or repeated flushing, is only carried out if an absolute or relative minimum number, for example 10 or 20%, of the selected cavities, in particular relevant measuring areas, do not have the expected shape and / or fluidic abnormalities affect such a minimum number. R. 418023.

[0057] - 13 -

[0058] Brief description of the drawings

[0059] Exemplary embodiments of the invention are shown schematically in the drawings and explained in more detail in the following description. The same reference numerals are used for the elements shown in the various figures that have a similar effect, thus avoiding a repeated description of the elements.

[0060] They show

[0061] Figure 1 shows an embodiment of the device according to the invention as well as a microfluidic cartridge,

[0062] Figure 2 is an illustration of tolerances to be taken into account by the method according to the invention.

[0063] Figures 3, 4 cavity arrays with illuminated areas,

[0064] Figure 5 shows a flowchart for an embodiment of the method according to the invention.

[0065] Figure 6 shows a flowchart with sketchy images of a normalization step of the exemplary embodiment of the method.

[0066] Figure 7 illustrates the difference between quantile normalization (Figure 7b) and a standard histogram transformation.

[0067] Figure 8 shows a schematic overview of the process of detecting a defect as a reference marker on a cavity array.

[0068] Figure 9 shows an example of a template image used for defect detection.

[0069] Figure 10 is a sketch of a masked entrance image, R. 418023

[0070] - 14 -

[0071] Figure 11 shows an exemplary process for estimating distances between cavities on the cavity array,

[0072] Figure 12 shows a sketch of a Fourier-transformed input image,

[0073] Figure 13 shows a schematic excerpt from the image in Figure 12.

[0074] Figure 14 shows an exemplary process for estimating the rotation of the input image relative to the optical sensor.

[0075] Figure 15 illustrates the difference between translational and rotational correlation.

[0076] Figure 16 shows a section of an input image with relevant measurement areas marked on it,

[0077] Figure 17 shows an exemplary division of an input image into tiles for correcting the positions of relevant measurement areas, as well as

[0078] Figure 18 shows an exemplary procedure for determining the position of an illuminated area on a cavity array.

[0079] Embodiments of the invention

[0080] An exemplary embodiment of the method and device according to the invention is described below.

[0081] Figure 1 shows, in sub-figure 1a, an analyzer 1000 as an embodiment of the device according to the invention and a microfluidic cartridge 100 which can be processed by the analyzer 1000, for example based on a cartridge and an analyzer as described in documents DE 102016222075 A1 and DE 102016 222072 A1, wherein the cartridge, as for example in DE 102018R. 418023

[0082] - 15 -

[0083] German patent applications DE 204624 A1 and DE 102018210069 A1 disclose and are explained in more detail below, a cavity array for the parallel execution of biochemical reactions in the cavities comprises. As indicated in partial figure 1a, the cartridge 100 can be received into the device 1000 for processing and optical analysis via a shaft 1010. A result of the analysis can be displayed via a display 1020.

[0084] Part 1b shows a schematic top view of the cartridge 100. A biological sample, for example, comprising blood, sputum, urine, or a swab, can be introduced into the cartridge via a sample entry chamber 151. The cartridge 100 includes chambers 152, 153, 154, and 155 for pre-storing buffers and reagents for sample purification and carrying out the biochemical reactions. The cartridge 100 also features a microfluidic network 130 for sample processing, which is connected to the reagent chambers 152–155 and the sample entry chamber 151 via channels (not shown). Actuation of the cartridge 100 is achieved, for example, via deflectable diaphragms integrated into the cartridge 100 in diaphragm valves and diaphragm pumps, with the deflection being effected by applying different pressure levels via a pneumatic interface 140 with multiple ports 141.

[0085] Performing an analysis of the biological sample, particularly an examination for specific pathogens in the sample, can, for example, involve a two-phase nucleic acid amplification. In the first phase, PCR pre-amplification is carried out in three differently temperature-controlled chambers 161, 162, 163 as so-called shut-PCR. In a second phase, the pre-amplification product is applied to a surface 111 of a chip 110 arranged in the cartridge 100, hereinafter also referred to as an array or cavity array, wherein the surface has recesses (cavities or wells) 112 for receiving a portion of the product each. Further reagents, in particular a PCR master mix with primers and probes, can be placed in the cavities for the individual amplification of the second phase. For example, the array 110 is a silicon substrate with eleven by eleven cavities 112 arranged in a grid.For example, a cavity 112 has a diameter of 375R. 418023.

[0086] - 16 -

[0087] The diameter (pm) and the distance (pitch) between the cavities is nominally 500 pm, excluding manufacturing tolerances. The temperature control of chambers 161, 162, 163 and the cavity array for PCR can be achieved by heaters (not shown) of the device 1000, which contact the cartridge 100 in defined areas after it has been inserted into the analyzer 1000.

[0088] As indicated in Figure 1a, the device 1000 includes a light source 1100 for illuminating the cavity array 110 when the cartridge 100 is inserted into the analyzer. Furthermore, the device 1000 comprises an optical sensor 1200, for example a camera with a CMOS chip, and a processor 1300 with, for example, integrated memory, such as a single-board computer (SBC), for controlling the components of the device 1000, in which, in particular, the method according to the invention may be installed as software.

[0089] The light source 1100 is designed to illuminate a first region of the array 110. Figure 3 shows two examples of the illuminated first region 120 in the two sub-figures 3a and 3b. The illuminated first region 120, hereinafter also referred to as the illumination spot, has, for example, a nominal diameter of 5 mm and a positional tolerance in the mm range. Although the first regions 120 differ in their position (see left in the sub-figures), practically identical illuminated images 1, 1 can result after detection by the optical sensor 1200 (see right in the sub-figures), which therefore initially do not allow for a unique determination of the positions and thus also no unique identification of the individual cavities 112.The cavity array 110 therefore also includes a reference mark 113, which, for example, is arranged as a cross 113 or defect 113 instead of one of the cavities 112 on the surface 111 of the array 110, as shown in the two sub-figures 4a and 4b of Figure 4. The light source 1100 and the reference mark 113 are arranged, taking tolerances into account, in the device 1000 and on the array 110 fixed in the cartridge 100, respectively, such that when the cartridge 100 is inserted into the device 1000, the illuminated first area 120 always includes the reference mark 113. For example, the tolerances are those specified in R. 418023.

[0090] - 17 -

[0091] The values ​​given in Table 1 below are explained below with reference to Figure 2:

[0092] Table 1:

[0093] Tolerance type Value Displacement error Array in xy dimension -5 mm < x, y < +5 mm Scaling error from nominal pitch -10% < z' < +10% Rotation error Array -5° < <p < +5° Verschiebungsfehler des Beleuchtungsspots -1,2 mm < x, y < +1,2

[0094]

[0095] mm

[0096] In Figure 2a, the cavity array 110 is schematically arranged within the field of view of the optical sensor 1200, for example, the camera 1200, particularly when the cartridge 100 is received and positioned at a predetermined location in the device 1000. A detectable area 1201, extending from the sensor 1200, is indicated by two solid lines and a dashed rectangle and encompasses, for example, the entire surface 111 of the cavity array 110. In particular, the detectable area 1201 is selected such that the entire surface, or alternatively at least selected cavities, are located within the area 1201, taking tolerances into account.As indicated in Table 1 and shown in Figure 2a, a lateral tolerance can exist in the x and y directions (as indicated in Figure 2a by the straight, oppositely oriented arrows), i.e., a displacement of the cavity array in a plane normal to the optical axis between the sensor and the cavity array, where the cavity surface 111 is nominally located in this plane. Furthermore, a tolerance, i.e., an uncertainty, can exist in the distance z between this plane and the optical sensor 1200 (as indicated in Figure 2b by the two straight, oppositely oriented arrows), which manifests itself as a scaling error in the images captured by the sensor. In other words, the distance between the cavities, denoted by z' in the table, varies in the captured images depending on this distance z, for example by up to 10%.Furthermore, the cavity array 110 can be rotated relative to the optical sensor (rotational error), both within the plane (xy-rotation) and tilted relative to the plane (xz- and / or yz-rotation), for example by an R. 418023.

[0097] - 18 -

[0098] Rotation angles of up to + / -5° each, as indicated by the semicircular double arrows in subfigures 2a and 2b (xy-rotation in subfigure 2a and xz-rotation in subfigure 2b). Furthermore, there may also be an error in the position of the lighting spot 120 (displacement error), which, as shown in Table 1, can be up to 1.2 mm in the x and y directions.

[0099] within the plane. By means of the method according to the invention, which is explained below by way of example, these tolerances can advantageously be at least partially taken into account and compensated for.

[0100] According to this embodiment, the method 500 according to the invention comprises the following steps, which are illustrated as a flowchart in Figure 5. The method 500 is not limited to this embodiment and is generally applicable to methods and devices in which the illuminated analysis areas can be distinguished from the other areas of the array or substrate by an optical sensor.

[0101] After the cavity array 110 has been positioned at the predetermined location in the analyzer 1000 relative to the light source 1100 by receiving the cartridge 100, the first area 120 of the cavity array 110 is illuminated by the light source 1100 in a first step 510, whereby, as described above, the first area 120 comprises several cavities 112 and the reference mark 113, which for this example will be referred to as

[0102] The missing position 113 is trained.

[0103] During illumination, the first area 120 is detected by the optical sensor 1200 of the analyzer 1000 in order to obtain an input image 201, which is shown in sketch form in Figure 5.

[0104] Alternatively, the first step 510 can simply comprise the provision of such a previously captured input image 201. In the first step 510, the processor 1300 of the device, or alternatively another processor of a computer, receives this input image and executes the subsequent steps of the procedure 500.

[0105] In a preferred second step 520, the input image 201 is normalized, in particular to ensure the most uniform conditions possible for the R. 418023

[0106] - 19 -

[0107] to create subsequent image processing steps. Normalization step 520 can include histogram transformation, filtering, background subtraction and / or morphological operations.

[0108] For example, normalization step 520 may have the following sub-steps, which are shown in the flowchart of Figure 6 with sketchy images.

[0109] In a first step 521, the input image 201 is normalized using a histogram transformation to obtain a normalized input image 202, which is referred to below as the first normalized input image 202. Preferably, a so-called quantile normalization is performed. Quantile normalization is advantageously more robust against outliers in intensity, for example, due to strong reflections or pixel defects, and thus provides a significant contrast improvement in the actually relevant intensity range compared to normal histogram transformations. Figure 7 shows in subfigure 7a how, in the usual histogram transformation, the entire gray value distribution 20 is normalized so that the maximum pixel value 21 of the gray value distribution 20 in the input image 201 is mapped to the maximum value 31 of the normalized gray value distribution 30.In contrast, in the quantile normalization indicated in subfigure 7b, the mapping 51 of a quantile value 41 to be defined, which determines the actually relevant intensity range, of the original distribution 40 is set as the maximum value 51 of the normalized gray value distribution 50, and values ​​42 greater than the quantile value 41 are also mapped (dipped) to this defined maximum value, and not to their true value 52 according to the figure (hatched area 52 in the normalized distribution 50).

[0110] In a second step (522), the background of the input image 201 is estimated and subtracted from the first-normalized input image 202. Initially, a ranking filter is used, specifically a median filter, although other ranks can also be used. The size of the filter kernel, which is applied to the first-normalized input image 202, is parameterizable, for example, a 20x20 rectangular or circular filter. The result of this filter operation is a highly blurred image, which is optionally R. 418023

[0111] - 20 -

[0112] This can be further optimized, particularly by using a Gaussian filter (also called a Gaussian blur). The blurred image 203 corresponds to the background estimate of the input image 201 and is subtracted from the initially normalized input image 202 in a third step 523 to obtain a difference image 204. Subtracting the background ensures a consistent light intensity across the difference image 204 and thus helps to avoid effects caused by illumination gradients.

[0113] After subtracting the background, a fourth sub-step 524 involves further normalization using quantile normalization to obtain a first-normalized difference image 205.

[0114] In a fifth sub-step 525 of normalization step 520, morphological opening is applied to the grayscale image of the first normalized difference image 205, specifically a so-called grayscale opening, which comprises grayscale erosion followed by grayscale dilation. The kernel size for this is based on the expected cavity size and, for example, corresponds to an expected size of half a cavity, 5*5 pixels, in order to suppress image elements smaller than a cavity. This fifth sub-step 525 is intended to suppress image disturbances such as pixel defects or non-circular structures (especially when using circular cavities 112). The result of this fifth sub-step 525 is subsequently referred to as the normalized input image 206.

[0115] Locating the reference mark in the input image

[0116] In a third step 530 of the procedure 500, the reference mark 113, in this embodiment the defect 113, is located. Figure 8 shows a general overview of the procedure for this defect detection, which is accomplished via pattern recognition and correlation. A 3x3 pattern with a defect 221, as shown in Figure 8, is used as a template image 220 for creating a correlation map 230. The 3x3 pattern with the defect serves as a template for a 3x3R. 418023

[0117] - 21 -

[0118] An arrangement of eight cavities 112 with the reference mark 113 as a defect 113 in the center. The eight areas 222 of the template representing the cavities, which are referred to as target zones 222 for cavities, are larger and partly shaped differently than the cavities to accommodate scaling and rotation tolerances. This is because the cavity array 110, partially captured in the input image, may be rotated relative to the template image 220. Furthermore, the distance between the cavity array and the optical sensor may deviate from the target distance due to the tolerances described above, and thus the size of the cavities captured in the input image 201 may vary. Therefore, the target zones 222 are preferably widened rotationally and linearly. In particular, the four target zones located at the corners of the 3x3 pattern are, for example, linearly stretched and diagonally aligned to the square 3x3 arrangement, as shown.Figure 9 shows an example of such a template image 220 with both the nominal target zones 222 and the stretched and rotated target zones 224, where the stretching and elongation are indicated by a double arrow. Each target zone 222 is assigned a positive numerical value for the creation of the correlation map 230, which results in a higher correlation value being calculated for sections (image samples) of the input image 206 that are compared with the template image 220 and exhibit intensity values ​​in these target zones.

[0119] Furthermore, all areas 225 of the template image 220 that are not being searched for, i.e., are not part of the eight target zones 220, are assigned a negative numerical value. This reduces the calculated correlation value for image samples that exhibit intensity in these areas. These areas 225 with negative numerical values, i.e., preferably all areas of the template image 220 outside the target zones 224, are referred to as penalty zones. Since the defect 113 has little to no intensity in the input image, the corresponding zone 221 of the defect 113 in the template image 220 is also a penalty zone. Using a pattern with target and penalty zones thus ensures that only patterns in the image sample that truly correspond to the searched pattern receive a high correlation value. R. 418023

[0120] - 22 -

[0121] In this embodiment, the template image 220 comprises two different types of penalty zones. All areas 225 outside the target zones 224 are assigned a moderate negative value, while the center of the template image 220, i.e., the penalty zone 221 of the defect 113, is assigned a high negative value. Both values ​​are configurable and can, in principle, be chosen differently. Furthermore, it is possible to define additional penalty and target zones within the template image 220, particularly those with different values.

[0122] Alternatively, instead of a 3x3 pattern, a larger pattern, such as a 5x5 or an asymmetrical pattern, such as a 5x3 pattern, can be used.

[0123] For the creation of the correlation map 230, the template image 220 is preferably moved stepwise, in particular in steps of 1 pixel, over the normalized input image 230. The portion of the input image 206 covered by the template image 220 then corresponds to an image sample, which is compared with the template image 220, in particular whereby the pixel-wise light intensities are assigned to the target and penalty zones and correlation values ​​are calculated for each. For example, the pixel-wise light intensities can be multiplied pairwise by the template values ​​– that is, figuratively speaking, the numerical values ​​that are currently aligned – whereby the sum of the multiplications is taken as the correlation value for the respective image sample.The result of comparing the template image 220 with the template image 220 for all considered image samples yields the two-dimensional correlation map 230, i.e., a 2D correlogram, which ideally contains a correlation peak at the position of the defect. An example of a correlation map 230 is also sketched in Figure 8.

[0124] To determine the defect location, the position of the peak, i.e., the point of highest correlation, must now be estimated. This can be done using standard peak estimation algorithms. In its simplest form, peak position estimation can be accomplished using a maximum value seek. More accurate results are obtained from approaches that consider all peak values ​​above a certain threshold. 418023

[0125] - 23 -

[0126] The limit values ​​are weighted to average in order to determine the center of gravity of the peak.

[0127] To assess the accuracy of the reference mark localization 530, a confidence value (individual confidence value for the localization of the reference mark) can be calculated, as described above, as a normalized ratio of the largest peak to the second largest peak from the correlation map 230. If this normalized individual confidence value is below a threshold, for example below 0.7, the procedure 500 can be terminated at this point or alternatively restarted with a new input image.

[0128] Masking

[0129] After the defect position has been estimated (530), an optional fourth step (540) of the procedure (500) involves cropping the normalized input image (206) to an area around the defect (113). This cropped image (207) is referred to below as the masked image (540). Generally, masking the image improves performance and reduces the error susceptibility of the procedure (500), as it allows for the removal of interfering optical artifacts outside the cavity area. Furthermore, it reduces computational effort because fewer image values ​​are used. The extent of the masking, and thus the resulting size of the masked image (207), can be parameterized, specifically its length and width, for example, by specifying the number of pixels in both directions. An example of such a masked image (207) is shown in Figure 10.Compared to the normalized input image 206, the cropped image 207 was restricted to the cavity area. For example, taking into account the already known position of the defect 113 and prior knowledge of the number and basic arrangement of the cavities in relation to the defect 113, i.e., knowledge of the given layout of the cavity array 110, and utilizing the above R. 418023.

[0130] - 24 -

[0131] The tolerances described, in particular the edge areas of image 206, are cut off.

[0132] Locating the cavities in the input image for cavity identification

[0133] In a fifth step 550 of the procedure 500, the cavities are localized, either with the normalized input image 206 or, if masking 540 has been performed, preferably with the cropped image 207.

[0134] a) Distance estimation

[0135] The localization 550 of the cavities 112 includes, in particular, an estimate 551 of the pitch, i.e., the distance, between the cavities. This distance can be unique for each analysis due to manufacturing tolerances of the array 110 (see also Table 1 above) and scaling effects resulting from the varying distance between the array 110 and the optical sensor 1200, i.e., in this case, the camera. Figure 11 illustrates an example of the substeps for the distance estimation 551.

[0136] First, in a substep 552, a 2D Fourier transform of image 206, 207 is calculated. Using the Fourier transform, the spatial frequency, and thus the spacing of the cavities arranged regularly within the aforementioned tolerances, can be determined. Figure 12 shows an example of the resulting Fourier-transformed image 208. A distinct 3x3 grid 11 is visible in the center of the image. The center point (position 0 / 0) represents DC components of the spatial frequency (in particular, representing the constant background). The eight points around the center point (also visible in image 208 in Figure 11), hereinafter referred to as peaks, are the fundamental spatial frequencies of the periodic cavity pattern in the different image dimensions. These can be used to determine the cavity spacing.The fundamental spatial frequencies of the cavity spacing have nominally, i.e., provided that image 206, 207 is not rotated, the positions (fx / O), (fx / fy), (O / fy), (-fx / fy), (-fx / O),R. 418023.

[0137] - 25 -

[0138] (-fx / -fy), (0 / -fy), (fx / -fy), where fx is the pitch in the x-direction and fy is the pitch in the y-direction. If, however, image 206, 207 shows a rotation, the points are rotated according to the rotation. The points that become progressively fainter towards the outside, which can be seen in figures 11 and 12 of image 208, represent harmonics of the cavity spacing spatial frequency.

[0139] To calculate the cavity spacing spatial frequency, the positions of one or more peaks, i.e., the points of the fundamental spatial frequencies (see the marked area 11 in Figure 12), must now be estimated. It is generally assumed that the pitch, i.e., the distance between the cavities, is identical in the x and y directions. However, in a second substep 553, the positions of the two peaks with nominal positions (fx / O) and (0 / f) are preferably determined. y ) (see the two peaks 12, 13 in Figure 13) determined and an estimate for f x and f yDerived. However, other or multiple peaks can also be used for the calculation. To estimate peaks 12, 13, an image region 14, 15 (region within the dotted rectangles 14, 15 in Figure 13) is first extracted for each peak to be estimated, within which the peak is expected. This region 14, 15 is determined using the nominal pitch, the expected pitch tolerance, and the expected maximum rotation (see Table 1 above). Each peak estimate consists of an x- and y-coordinate. The cavity spacing spatial frequency is calculated from the magnitude of the estimated peak position, i.e., / p. itch , n = + y e 2 st,n - Their reciprocal is

[0140]

[0141] The cavity spacing in pixels. Since the x and y pitches are assumed to be identical, the estimated values ​​(pitch value candidates) are preferably averaged to obtain a pitch estimate. However, it is also generally conceivable to proceed with individual estimates for the x and y directions. The accuracy of the peak estimation can be further improved by interpolation, specifically by zero-padding the image 206, 207 before performing the Fourier transform. Due to the zero-padding of the image, this advantageously leads to a higher frequency resolution in the Fourier transform and thus to interpolation of the values ​​in the frequency domain.

[0142] The peak coordinates allow us to determine not only the pitch but also the rotation of image 206, 207R, due to the influence of possible image rotation. 418023

[0143] - 26 -

[0144] Figure 13 also illustrates, using the formulas given therein, how the rotation can be estimated over a rotation angle cp to be determined, with the help of the two peaks 12, 13, by determining a rotation of the 3x3 grid relative to the horizontal 16 using simple trigonometry. Preferably, however, the rotation of the image 206, 207 is estimated separately as follows.

[0145] To assess the quality of pitch estimation 551, a confidence value (individual confidence value for the pitch estimation) can be calculated, as described above, as the averaged difference between the amplitudes of the pitch value candidates. If this normalized individual confidence value is below a threshold, for example 0.7, procedure 500 can be terminated at this point or alternatively restarted with a new input image.

[0146] b) Rotation estimation

[0147] The localization 550 of the cavities 112 preferably also includes an estimate 554 of the rotation, i.e., how far the captured image 201-207 is rotated with respect to a nominal position, in particular relative to the optical sensor 1200. In particular, correlation is used to estimate the rotation of the image relative to the optical sensor 1200, i.e., in this case, the camera. Figure 14 provides an overview of the exemplary procedure described below.

[0148] Similar to the localization 530 of the reference mark 113, a correlation map 250 is generated here by comparing the input image, in particular image 206, 207, with a template image 240. This map represents the rotational correlation between sections (image samples) of image 206, 207 and template image 240, i.e., the correlation as a function of the rotation angle. Analogous to the defect detection described above, template image 240 can comprise a 3x3 or 5x5 pattern with target and penalty zones, for example, a 5x5 pattern as shown in Figure 14. In principle, however, other pattern dimensions can also be used. DasR. 418023

[0149] - 27 -

[0150] Template image 240 is generated taking into account the above-estimated distance between the cavities 112; that is, the target zones 241 of the template image 240 are arranged at this distance from one another. For the required accuracy of the rotation estimation, it may be sufficient to use a standard shape, in particular a circular shape, for the target zones 241 for the cavities without distortion or magnification. Therefore, in this embodiment, pitch and rotation tolerances are not taken into account via rotational or linear distribution of the target zones (as described above for defect detection). Penalty zones are defined as all areas outside the circular target zones 241 as well as the circular area of ​​the reference mark 242.

[0151] Since the rotation is to be estimated, a standard, "translative" correlation, where the stencil image 240 is linearly shifted over the image 206, 207, as described above for defect detection, is not suitable. Instead, a rotational correlation is required, where the stencil image 240 is rotated over a specific position in the image 206, 207, which serves as the rotation point. Figure 15 schematically illustrates translational and rotational correlation in the two sub-figures 15a and 15b, respectively. In sub-figure 15a, for a translational correlation, the stencil image 240 is shifted linearly, specifically row by row, for example in an S-shape as indicated, over the image 206, 207, while in sub-figure 15b, for a rotational correlation, the stencil image 240 is rotated relative to the image 206, 207 at a rotation point 249. The rotation point 249 preferably serves as the already estimated reference mark, i.e., in this case, the defect 113.

[0152] The technical implementation of the rotational correlation is achieved through the transformation 555 of the template image 240 and the images 206, 207 into polar coordinates, with the nominal reference mark 242 of the template image 240 as its center point. Figure 14 illustrates the transformation 555 for the two images 240, 206, 207 into transformed images 243, 208. In the polar coordinate system, the rotational correlation becomes a translational correlation 556. The result of the correlation 556 of both images 240, 206, 207 is a curve 251 in the correlation map 250, which represents the correlation value as a function of the rotation angle. The maximum value of this curve 251 is R. 418023

[0153] - 28 -

[0154] Assuming an estimate of 557 for the rotation, which in the curve 251 shown is, for example, about 1 degree,

[0155] Preferably, the template image 243 transformed into polar coordinates can be weighted; in particular, different areas in the transformed template image 243 can be weighted differently. Preferably, a weighting with higher weights is applied to areas with larger radial values, for example, a continuous increase in the weight with increasing radial values, to compensate for the effect that areas with small radii are stretched more by the transformation. Figure 14 also illustrates that target zones 244, 245 closer to the center 242 of the template image 240 are depicted larger in the polar-transformed template image 243 than target zones 246, 247 that are further away.

[0156] To assess the quality of the rotation estimate 554, a confidence value (individual confidence value for the rotation estimate) can be calculated, as described above, as the difference between the largest and smallest correlation values ​​from the correlation map 250, preferably normalized to 1. If this normalized individual confidence value, or alternatively, the mean of this individual confidence value and the individual confidence value for the pitch estimate, or in a further alternative configuration, the mean of all determined individual confidence values ​​(i.e., the overall confidence value described above), lies below a threshold, for example, below 0.7, the procedure 500 can be terminated at this point or, alternatively, restarted with a new input image.

[0157] Calculation of the relevant measurement ranges

[0158] After the cavities 550 have been located, in the sixth step 560 relevant measurement ranges 210 of the individual cavities 112 are determined, in particular calculated, using the localized reference mark 113, the localized cavities 112 and a predefined layout of the cavity array 110. R. 418023

[0159] - 29 -

[0160] The specified layout of the cavity array 110 includes in particular the grid or pattern of the arrangement of the cavities on the array 110, the size of the cavities and the nominal pitch, whereby the nominal pitch is replaced by the pitch estimated during localization 550, 551.

[0161] The size of the relevant measurement areas 210 can be determined, in particular, by two parameters concerning the length and width of the respective measurement areas 210. The estimated cavity spacing and the nominal ratio can be taken into account in this regard. Thus, for each cavity, a window with a defined number of pixels is defined as the relevant measurement area 210, which is used for optical analysis to extract intensity values. Figure 16 shows a section of the input image 201 with the relevant measurement areas 210 marked on it. The relevant measurement areas 210 can be bounded, as shown, by squares 211 with a border thickness of one pixel.

[0162] Correction of the relevant measurement ranges

[0163] Preferably, the relevant measurement ranges 210 are corrected with respect to the light intensities captured in the input image 201. This correction includes, in particular, a positional correction of the relevant measurement ranges 210 to reduce inaccuracies in the individual estimates used to locate the cavities 112, i.e., the estimates of the defect position, rotation, and pitch. Due to the orientation of the pitch and rotation estimates to the reference mark, even small errors can accumulate into a larger positional uncertainty for cavities 112 located further away from the reference mark 113, which can be substantially reduced by the procedure described below.

[0164] Based on the cavity positions estimated in step 550 and the relevant measurement areas 210 defined in step 560, the image is divided into tiles 275, as schematically shown in Figure 17, using both the input image 201 and preferably a further processed image, in particular the first normalized image R. 418023

[0165] - 30 -

[0166] 202, the first normalized difference image 205, or the normalized image 206. Within each tile 275, a local position optimization is now performed via a center of gravity calculation. The center of gravity calculation can be performed using a formula commonly used for calculating physical centers of mass, where the pixel intensities are interpreted as masses and where detected bright light intensities are, in this sense, "heavier." The positions of the relevant measurement areas 210 are thus corrected in the direction of the detected light intensity, without changing their size. This is shown schematically in an enlarged representation of a tile 275: The luminous area 276 within an arbitrarily selected tile 274 is simplified and uniformly represented as a filled circle 276. Furthermore, the uncorrected relevant measurement area 277 and the corrected relevant measurement area 278 of this tile 274, determined by the center of gravity calculation, are shown.The corrected relevant measurement area 278 is located, as expected, in the center or, in the case of non-isotropic light intensity, near the center of area 276.

[0167] Estimation of the lighting spot

[0168] In a seventh step 570 of the method 500, the position and preferably also the intensity distribution of the illuminated first area 120 (illumination spot) on the cavity array 110 are determined. With the help of this information, the degree of illumination of the cavities can be determined in particular. This advantageously helps to exclude unilluminated cavities from further processing, especially from subsequent analysis, or to compensate for inhomogeneous illumination.

[0169] The determination 570 of the position of the lighting spot can be carried out parallel to or after the detection or reception 510 of the input image 201 and, in particular, also parallel to one or more of the subsequent steps 520-560. Figure 18 outlines the sequence of the following sub-steps.

[0170] In a first step 571, the input image 201 is to be normalized using histogram transformation, in particular by means of an R. 418023

[0171] - 31 -

[0172] Quantile normalization. According to a second substep 572, the background of this normalized input image 301, 202 is to be estimated. Since these two substeps 571, 572 were already performed during the input image normalization described above, they do not need to be repeated, and the already generated blurred image 302, 203 from the background estimation performed above can be used directly for a third substep 573. In the third substep 573, this image 302, 203 is morphologically opened with a kernel the size of the expected illumination spot, thereby suppressing image elements that do not correspond to this size. This results in an image 303 with an approximately circular intensity, which corresponds to the illumination spot in its position and shape.Therefore, in a fourth sub-step 574, a circle 304 can be assigned to this intensity via Hough transformation, the position and size of which are subsequently used as the position and boundary of the illumination spot 120, as shown by way of example in Figure 18 on the normalized input image 206. Alternatively to morphological opening, the image 302, 203 can be subjected to a simple thresholding procedure in the third sub-step 573 to obtain a binary image, on which the circle 304 is then determined as the position and boundary of the illumination spot via Hough transformation.

[0173] Calculation of the illumination metrics of the cavities

[0174] According to a preferred eighth step 580 of procedure 500, metrics are derived for each cavity, which can be considered for the further procedure, in particular for classifying the cavities. For this purpose, information from the preceding determination of the illumination spot and the estimated cavity positions is used. Furthermore, additional information from the captured input image 201 can, in principle, be used to determine the metrics.

[0175] In particular, step eight 508 can determine an illumination metric of the cavities by comparing the relevant measurement ranges with the position-determined illumination spot and a recorded brightness R. 418023

[0176] - 32 -

[0177] The relevant measurement areas must be included. The illumination metric can include a Boolean value (True / False) for the illumination of each relevant measurement area. In a simple implementation, a percentage value can initially be assigned to all relevant measurement areas. Relevant measurement areas that lie entirely within the illumination spot are assigned 100%, and relevant measurement areas that lie entirely outside the illumination spot are assigned 0%. Relevant measurement areas that lie partially within the illumination spot are assigned a percentage of 100% corresponding to the area within the illumination spot. Only relevant measurement areas with a percentage value above a predefined threshold, for example, above 70%, are then assigned the Boolean value "True".The illumination metric can alternatively or additionally include a real value representing the illuminance level averaged or summed across the relevant measurement range. This value can be expressed as a percentage of a predefined maximum intensity value. The predefined maximum intensity value can, for example, correspond to the highest intensity value recorded across all relevant measurement ranges. Alternatively, the predefined maximum intensity value can correspond to the maximum resolution assigned to the pixels, such as 255 for an 8-bit resolution. This allows for the identification of differences in illumination intensity between cavities, which can then be advantageously used to normalize or compensate for signal intensities in subsequent analysis.

[0178] Preferably, only relevant measurement areas lying within the position-determined illumination spot are considered. For example, the aforementioned Boolean value can also be set to False for all relevant measurement areas lying outside the position-determined illumination spot. The illumination metric can also include a two-dimensional quantity to represent a spatially resolved intensity distribution in absolute or relative terms within the respective relevant measurement areas, thus revealing inhomogeneities in the illumination. R. 418023

[0179] - 33 -

[0180] Selection of relevant measurement ranges for optical analysis

[0181] In a ninth step (590) of procedure 500, one or more of the relevant measurement areas for optical analysis are selected. This selection depends on the specific illumination metric for each relevant measurement area. As described above, relevant measurement areas, and thus the associated cavities, can be excluded from the analysis if the illumination metric falls below a defined threshold. This serves to exclude cavities with insufficient illumination from further processing. Alternatively, only those relevant measurement areas that lie within the position-determined illumination spot can be selected. When using an illumination metric such as the one described above, the relevant measurement areas that only partially lie within the illumination spot can be assigned a percentage value of 0.

[0182] The selection in step nine (590) can include a classification of the cavities, in particular the associated relevant measurement ranges. A first class, for example designated BELOW_ILLUMINATION_THRESHOLD, can be assigned to those cavities or relevant measurement ranges whose illumination metric, as described above, lies below the defined limit value.

[0183] In a simplified alternative of the procedure 500, in the ninth step 590, instead of determining the illumination metric, all relevant measurement areas for the optical analysis can be (pre)selected which are located within the position-determined, in particular circularly limited, illumination spot 120.

[0184] Furthermore, the selection of the relevant measurement ranges for optical analysis may depend on the expected shape of the cavities. For example, rounded cavities are expected. However, artifacts are possible due to disturbances such as fibers or dust on the array 110, which make the cavities 112 appear non-round when illuminated by the light source 1100. R. 418023

[0185] - 34 -

[0186] This can be done, for example, by calculating the circularity of the illuminated area within the respective relevant measurement range. First, the relevant measurement ranges can be subjected to a thresholding procedure with a defined or dynamic threshold (for example, using Otsu's Method) to obtain binary images for the measurement ranges. On the binary image of the respective measurement range, which in the undistorted case shows a white circular area corresponding to the cavity on a black background, the circularity of the white area can then be calculated. The circularity is calculated by multiplying the area of ​​the white area by 4*n and dividing by the square of the perimeter of the white area, where the area is the sum of the white pixels and the perimeter is the sum of the pixels bounding the white area (i.e., the edge pixels).If a predefined circularity threshold is not met, for example 75%, the affected cavities, and in particular the associated relevant measurement areas, can be disregarded during further processing or analysis. Such cavities can then be excluded from further processing, as it is assumed that a defect exists in these cavities. In the case of classification, such cavities or relevant measurement areas can be assigned to a second class, for example, designated as UNCIRCULAR.

[0187] The selection of relevant measurement areas for optical analysis can depend, in particular, on the detection of fluidic anomalies. This serves to exclude cavities that are covered by fluidic artifacts such as bubbles or foam. Such cavities, or relevant measurement areas, can be assigned to a third class, for example, labeled COVERED_BY_BUBBLE. To visualize and detect such fluidic artifacts, the first area 120 can be re-captured by the optical sensor after the cavity array has been filled with liquid, in order to obtain a second input image, provided the first input image 201 was acquired before filling. The intensity distributions in the relevant measurement areas of the first image 206, 207 are then compared with the intensity distributions in the same relevant measurement areas to analyze any resulting differences. R. 418023

[0188] - 35 -

[0189] to identify the artifacts. This can be achieved by exploiting the fact that bubbles generally cause darkening, while liquid films are associated with brightening. Additionally, edge detection algorithms can be used to identify more sharply defined objects. Furthermore, data-driven approaches, particularly machine learning, as well as pattern recognition approaches using templates, can be employed, especially for foam detection.

[0190] In the course of the selection in the ninth step 590, all cavities or relevant measurement areas can then preferably be excluded from the analysis which, although they lie within the position-determined illumination spot, belong to at least one or alternatively several of the classes.

Claims

R. 418023 - 36 - Claims 1. Method (500) for identifying and selecting cavities of a cavity array (110) for optical analysis, in particular of microfluidic biochemical reactions in the cavities (112), with a processor (1300) of a device (1000), comprising the steps: • Receiving (510) an input image (201), wherein the input image represents an illuminated first area (120) of the cavity array (110), the first area (120) comprising several cavities (112) and a reference mark (113). • Locating (530) the reference mark (113) in the input image (201, 206, 207) • Locating (550) the cavities (112) in the input image for cavity identification, taking into account the localized reference mark • Determining (570), in particular a position and preferably an intensity distribution, of the illuminated first area (120) in the input image • Selecting (590) one or more of the localized cavities for optical analysis by comparison with the determined illuminated first area (120) 2. Method (500) according to claim 1, wherein the input image (201) undergoes normalization (520) prior to localizing the reference mark (113), in particular comprising a histogram transformation, filtering, background subtraction and / or morphological operations on the input image.

3. Method (500) according to one of the preceding claims, wherein for locating (530) the reference mark, a template image (220) is compared with different areas of the input image (201, 206) for creating a correlation map (230), in particular by passing it over the input image, wherein the template image comprises target zones (221) and / or penalty zones (221) for defining correlation values ​​in the correlation map, wherein preferably at least some of the target zones are larger and / or different to take into account tolerances. R. 418023 - 37 - shaped like the cavities.

4. Method (500) according to any of the preceding claims, wherein the localization (550) of the cavities comprises an estimation (551) of distances between the cavities and / or an estimation of a rotation (554) of the input image relative to the optical sensor.

5. Method (500) according to claim 4, wherein the estimation (551) of the distances is carried out via a Fourier transformation (552) of at least a part of the input image.

6. Method (500) according to claim 4 or 5, wherein the estimation (554) of the rotation is carried out by comparing (556) the input image with a template image (240), in particular for creating a correlation map (250) for rotational correlation between areas of the input image and the template image, and wherein preferably before the comparison the input image and the template image each undergo a transformation into polar coordinates for translational correlation.

7. Method (500) according to one of the preceding claims, wherein measurement areas (210, 211) relevant for optical analysis of the individual cavities are determined using the localized reference mark, the localized cavities and a predetermined layout of the cavity array.

8. Method (500) according to claim 7, wherein the relevant measuring ranges (210, 211) are corrected with respect to the light intensities detected in the input image, in particular comprising a position correction of the relevant measuring ranges.

9. Method (500) according to one of the preceding claims, wherein the selection of the localized cavities (112), in particular the relevant measurement areas (210, 211), is carried out by determining an illumination metric for the cavities, wherein the determination of the illumination metric comprises comparing the cavities with the position-determined first area and a detected brightness of the respective cavities. R. 418023 - 38 - 10. Method (500) according to one of the preceding claims, wherein the selection (590) of the localized cavities, in particular the relevant measurement areas, for the optical analysis depends on an expected shape of the cavities (112), in particular the relevant measurement areas (210, 211).

11. Method (500) according to one of the preceding claims, wherein the selection (590) of the localized cavities (112), in particular the relevant measurement areas (210, 211), for the optical analysis depends on the detection of fluidic anomalies 12. Method (500) according to one of the preceding claims, wherein the method (500) comprises determining a confidence value, wherein the confidence value is derived from a single confidence value for locating (530) the reference mark (113) and / or a single confidence value for locating (550) the cavities (112), wherein preferably, if the confidence value falls below a predetermined limit value, a measure is triggered by the processor, in particular a termination or a re-execution of the method (500).

13. Device (1000), in particular an analytical device for carrying out biochemical reactions, in particular comprising nucleic acid amplification, comprising a receiving area for receiving a cavity array (110), in particular a microfluidic cartridge (100) with a cavity array (110), a light source (1100) for illuminating at least a first area (120) of the cavity array, an optical sensor (1200) for detecting the area and a processor (1300), wherein the device is configured to carry out a method (500) according to one of the preceding claims.

14. Computer program comprising instructions which, when the program is executed by a computer, cause it to execute a method (500) according to any one of claims 1 to 12.

15. Computer-readable medium on which the computer program according to claim 14 is stored.

Images