Methods for locating signal sources in localization microscopy

Abstract

The invention relates to a localization microscopy method for localizing a signal source (3.1, 3.2). In this method, at least once for each pixel (1) of a detector (2), the value of an error parameter is determined and stored in a calibration data record in a manner assigned to the relevant pixel (2). The captured image data are used to identify the origin region (4) of the signal source (3.1, 3.2) and to fit a point spread function to the pixel values ​​of the corresponding origin region (4). The corresponding signal source (3.1, 3.2) is located based on the point spread function. The pixel-specific error parameter of each pixel (1) can be compared with a threshold value. If the threshold value is exceeded, these pixels (1) are ignored when fitting the point spread function or are replaced by interpolation. Additionally or alternatively, the actual noise performance of the pixel (1) is determined and corrected based on the derived pixel-specific error parameter.

Description

Technical Field

[0001] The present invention relates to a localization microscopy method for localizing a signal source. Background Art

[0002] Various localization microscopy methods are known from the prior art. Prominent examples are methods for localizing individual signal sources, in particular individual emitters, according to methods referred to by the abbreviation PALM (Photoactivated Localization Microscopy; e.g. WO 2006 / 127692 A2) or STORM (Stochastic Optical Reconstruction Microscopy; US 2008 / 0032414 A1). As examples, overviews of localization microscopy are given in Klein et al. 2014 (Klein, T. et al., 2014, Eight years of single molecule localization microscopy, Histochemistry and Cell Biology, Vol. 141: 561-575), and in Babcock et al. 2012 (Babcock, H. et al., 2012, A high-density 3D localization algorithm for stochastic optical reconstruction microscopy, Optical Nanomicroscopy, Vol. 1: 6). in the article "Reconstruction Microscopy").

[0003] The known localization microscopy methods have in common that they obtain information on the localization of signal sources (emitters, emitting molecules) from pixel-type 2D surface sensors used as detectors or cameras. Very sensitive detectors are required, since only a few hundred to a few thousand photons are available for each localization step to localize individual signal sources, such as individual emitting molecules.

[0004] As an example, such a sensitive detector is an EMCCD (Electron Multiplying Charge Coupled Device) sensor. However, the CCD architecture underlying the EMCCD sensor has a limited number of pixels and / or a limited readout speed due to the serial readout and gain process.

[0005] As an alternative, in recent years, so-called "scientific CMOS" or sCMOS (CMOS=complementary metal oxide sensor) sensors have been created. They unify the "active pixel sensor" architecture, possibly because CMOS technology has a very low readout noise and a high quantum efficiency.

[0006] The advantages of sCMOS sensors over conventional EMCCD cameras include: high frame rates, a large number of pixels and smaller pixels. In addition, since there is no electron multiplication in sCMOS sensors, there is no so-called excess noise. Ultimately, this leads to a higher effective quantum efficiency. Therefore, sCMOS sensors are also suitable for use in localization microscopy.

[0007] Possible inhomogeneities in the signal behavior of individual pixels are disadvantageous for the localization accuracy of individual emitters as radiation sources. While such signal inhomogeneities would lead to disturbed image impressions as a worst-case scenario in normal imaging, during the localization step of localization microscopy, these signal inhomogeneities could lead to positioning that points to errors. In particular, such inhomogeneities could be due to the already mentioned "active pixel sensor" architecture.

[0008] Patent US 9,769,399 B2 and a technical paper by Huang et al. in 2013 (Huang, F et al., "Video-rate nanoscopy using sCMOS camera-specific single molecule localization algorithms", Nature Methods, Vol. 10, pp. 653-658, 2013) describe the use of parameterized models to model noise performance.

[0009] Here, the noise associated with the pixel is modeled as a Gaussian distribution. As an example, the modeled parameters are the mean (deviation) of the Gaussian distribution per pixel, the variance of the distribution and / or the gain.

[0010] As an example, the offset can be determined by capturing a plurality of dark images. These dark images can be used to determine the signal generated by the pixel despite the lack of detected radiation (dark noise; false signal). As an example, an average value can be formed from this data and this average value can be used as the error parameter.

[0011] The gain can be determined by capturing and evaluating multiple images with different known photon numbers or illumination levels. Starting from the captured image data, a distribution function of the pixel values ​​of the individual pixels can be determined, using, for example, the sum of the photon-induced variance ("shot noise") and the Gaussian variance.

[0012] However, such a method of correcting noise performance has the disadvantage that individual pixels that deviate significantly from the corresponding model may lead to significant errors. As an example of such pixels, so-called "scintillators" are specified, which can lead to random output signals (Wang, X. et al., "Random telegraph signals in CMOS image sensor pixels" in 2006, Electronic Devices Meeting; IEDM'06, International IEEE 2006). Summary of the invention

[0013] The object of the present invention is to propose a localization microscopy method for localizing a signal source, thereby reducing the disadvantages occurring in the prior art.

[0014] This object is achieved by a localization microscopy method for localizing a signal source, wherein at least once for each pixel of a detector for capturing the detection radiation, the value of a pixel-specific error parameter is determined and stored in a calibration data record in a manner assigned to the relevant pixel. Image data of the sample are captured pixel by pixel as pixel values ​​in the image data record, and an origin region of the signal source is identified based on the captured image data, the origin region comprising a plurality of pixels. A point spread function (hereinafter also abbreviated as PSF) is fitted to the pixel values ​​of the respective pixels of the respective origin region. The respective signal source is localized, i.e. an origin is determined in 2D and / or 3D within the relevant origin region based on the PSF.

[0015] According to the invention, within the calibration data record and for each pixel, a pixel-specific error parameter is compared with a predefined threshold value. Each pixel whose value of the pixel-specific error parameter is greater than the threshold value is marked in the calibration data record. When fitting the PSF to the pixel values ​​of the corresponding origin region, all or some of the marked pixels in the image data record are ignored or replaced by interpolation.

[0016] Additionally or alternatively, a derived pixel-specific error parameter can be determined for each pixel in the method based on the calibration data record instead of the pixel-specific error parameter and / or an ADU (Analog to Digital Unit) histogram of a plurality of pixel values ​​can be created. The PSF is fitted based on the derived error parameter or based on the corresponding ADU histogram.

[0017] A pixel within the meaning of this specification is to be understood as a detector element of a detector, for example a detector element of a sCMOS sensor. The detector has a plurality of detector elements which are arranged in particular in a two-dimensional matrix or array.

[0018] The pixel values ​​captured during the calibration measurement are stored in a calibration data record and, optionally, the corresponding pixel-specific and / or derived pixel-specific error parameters are stored therein in a manner assigned to the corresponding pixels. The image data record contains image data in the form of pixel values ​​for the individual pixels, which were captured during image capture, for example as raw data in a PALM or STORM method. The individual calibration data records or data records derived therefrom can be attached to the image data record, for example as metadata. In this way, for example, pixels marked in the calibration data record can also be marked in the image data record.

[0019] The interpolation of the pixel values ​​of the image data records can be carried out using known methods, for example by means of regression methods, estimation and / or the formation of a mean value or a median value.

[0020] The pixel-specific error parameter is a signal output by the relevant pixel that is not due to the captured detection radiation but due to different influences due to the environment and / or components. Such an error signal may be caused by the temperature of the detector and the temperature change of the detector due to the surrounding environment. Due to the components, the error signal may be generated, for example, by the presence of defective areas in the semiconductor material of the detector (Wang, X. et al., "Random telegraph signals in CMOS image sensor pixels" at the Electronic Devices Conference in 2006; IEDM'06, International IEEE 2006). The error signal of the pixel transmits the capture of at least one photon even if no photon is actually captured or fewer photons than required for the sensitivity of the relevant pixel are captured.

[0021] Pixel-specific error parameters, which are also referred to hereinafter as error parameters for short, refer, for example, to the so-called offset, the variance of the pixel values ​​(in particular the time series), and the gain.

[0022] Pixel-specific error parameters compared with the threshold value may also already be corrected, for example with regard to their offset and / or gain.

[0023] The derived error parameter is the quantity derived from the captured pixel value. In particular, in one configuration of the method according to the invention, a Gaussian function can be fitted to the pixel value of the pixel. The width of the Gaussian function (expressed in sigma) can be used as the derived error parameter of the variance. This procedure is advantageous over the prior art, in which the variance of the pixel is determined under the assumption of a Poisson distribution. Variance deviations from a Poisson distribution may have very different physical causes, which are unknown in the individual case; this is why the distribution function is not necessarily analytically describable.

[0024] However, the width of the distribution function can be approximated under the assumption of a Gaussian function and thus a better reproduction of the noise performance can be achieved for those pixels that deviate from Poisson behavior. The determined variance can thus likewise be compared with a threshold value. Pixels whose variance exceeds the threshold value are marked and then replaced by interpolation or completely ignored using the image data record.

[0025] Emitting molecules (emitters) (their two-dimensional or three-dimensional position is called the origin) can act as the signal source. The origin is determined within the region of origin by the PSF. In localization microscopy, the PSF typically has an extent (full width at half maximum) of at least 9 (3×3) to 16 (e.g. 4×4) or 25 (5×5) pixels.

[0026] The core of the invention is to improve the localization of the signal source. If the pixel values ​​whose error parameters exceed a predetermined threshold are ignored or interpolated, the displacement of the origin of the signal source to be determined in the pixel direction, for example, the expanded pixel value, is effectively offset. In addition, the corresponding specific signal behavior of the pixel can be used to correct the error signal.

[0027] According to the present invention, two basic steps are performed to achieve an improved consideration of the actual noise performance of each individual pixel.

[0028] In a first step, calibration of the individual pixels of the detector is performed. Calibration can be performed once or periodically, or can be repeated if necessary (for example if the properties of the detector change during its service life). The calibration data are stored in a calibration data record.

[0029] Calibration of the detector is initially carried out with the aid of all specific filters of the detector platform (= camera), which reduce or suppress unwanted noise and / or flicker when the normal imaging purpose of the detector is disabled. The same applies to reducing or suppressing false signals from so-called hot pixels (i.e. pixels that permanently emit false signals). Subsequently, a plurality of dark images can be recorded and error parameters and calibration data can be determined for individual pixels based on the pixel values ​​of said dark images.

[0030] In a second step, the calibration data are used to take into account the pixel-specific noise properties in order to correct the corresponding pixel values ​​of the image data record and to avoid or at least reduce the disadvantages known from the prior art. In this case, for example according to the PALM or dSTORM method, the calibration data of the calibration data record are appended to the image data record of the measurement performed or are provided differently. As an example, a link to the stored calibration data is written to the metadata instead of the calibration data themselves.

[0031] In one configuration of the method, particular account is taken of the fact that within the scope of localization microscopy, only a limited number of photons per signal source are available in each case. Therefore, even in the case of detectors with ideal noise performance, the accuracy of the localization is limited.

[0032] Therefore, since in the first step of the method the individual noise performance of each pixel is determined for each pixel of the detector, each pixel contributes this number of error signals during the subsequent localization. The number of photons is then determined, thereby determining the localization accuracy to be expected in the planned experiment. Prior to this, the permissible localization inaccuracy was set. Therefore, the set tolerance threshold is used as a threshold.

[0033] The threshold value may be set to a value that should not be exceeded by the variance of pixel values ​​(particularly the variance of pixel values ​​in time series) or the value of the offset of pixel values.

[0034] Pixels whose pixel values ​​of the calibration data during calibration exceed a threshold value are marked in the image data record. Since the calibration data record is assigned to the image data record, in particular appended to the image data record, appropriately corresponding pixels of the image data record are also marked. Then, in the step of positioning, the marked pixel values ​​can be interpolated and replaced, for example using pixel values ​​of neighboring pixels of the image data record.

[0035] Alternatively, the pixel values ​​of marked pixels may be ignored.

[0036] Thresholds can be set generally or individually for each experiment. For the general setting, for example, a fluorophore that emits the least number of photons can be used (e.g., tdEOS used with PALM; Wang et al., "Characterization and development of photoactivatable fluorescent proteins for single-molecule-based superresolution imaging", PNAS, 111, 8452-8457, 2014).

[0037] If a threshold is set for a single experiment or a group of experiments, the distribution of labeled pixels may change as a result.

[0038] Additionally or alternatively, the threshold value may be set based on an expected or known wavelength of the detected radiation, the intensity of the detected radiation, and / or the temperature of the detector.

[0039] In order to avoid that the accuracy of positioning is negatively affected by a large number of marked pixels, the allowed number of marked pixels of each origin region (in absolute value or average value) is limited to the maximum number of an advantageous configuration of the method. As an example, each origin region allows a maximum of two pixels, preferably a maximum of one pixel.

[0040] If the maximum number allowed is exceeded, the relevant detector is not used, or a warning is provided for the localization data of the origin region. The image data and / or localization data of the origin region may still be used, but the use in subsequent analysis should be achieved by retention.

[0041] The pixel values ​​of unmarked pixels of the image data record can be corrected using the calibration data record, for example with regard to offset and / or gain.

[0042] According to a further configuration of the method, a derivative error parameter is determined for each pixel based on the calibration data record. Here, the variance of the pixel values ​​of the individual pixels can be determined as the derivative error parameter. Unlike the prior art, this is advantageously achieved not under the assumption of a Poisson distribution but under the assumption of a Gaussian distribution (see explanation above).

[0043] The derived error parameter can also be a photon transfer curve, by which the gain can be corrected or calibrated. To this end, multiple (n) images are captured under uniform illumination and different illumination levels (m values), and the mean value Mm and variance Vm are determined for each pixel. Based on the parameters Mm and Vm, the corresponding photon transfer curve is created (see also: Long, F. et al., 2014, "Effects of fixed pattern noise on single molecule localization microscopy" in "Physical Chemistry Chemical Physics", Vol. 16: pp. 21586-21594).

[0044] Regarding the quality of the positioning, it is advantageous that in other configurations of the method, the measured noise performance of each pixel is determined separately in the form of an ADU histogram or corresponding data and stored in a retrievable manner in a calibration data record. To this end, a dedicated set of ADU histograms for different illumination levels is allocated to each pixel. Therefore, a possible multidimensional LUT (lookup table) can be used for each pixel, on the basis of which the positioning algorithm applied can read the real noise performance of each pixel based on the current local conditions. Taking into account the derived error parameters or based on the corresponding ADU histogram, a positioning calculation based on the PSF is implemented. As an example, for example, the pixel value of the pixel is captured as a grayscale value. The real or actual grayscale value can be determined for the relevant pixel with the help of the LUT and used for other programs. This process allows the real noise performance of the corresponding pixel to be determined still before the step of positioning, and allows positioning to be carried out based on the real noise performance. Here, this configuration of the method can be applied to all pixels, or only to marked pixels.

[0045] In order to ensure efficient processing and to obtain a compromise between resolution (thereby accurately reproducing the pixel behavior) and computational overhead or the range of metadata, the data of the corresponding ADU histograms can be combined in a suitable manner (binning). Furthermore, the ADU histograms can still be smoothed and / or interpolated and / or adjusted by functions, and their parameters can be stored.

[0046] To a large extent, the derived error parameter represents the real noise performance of the relevant pixel, advantageously allowing improved modeling and more accurate localization. In contrast to the approximation proposed according to, for example, Huang et al., 2013 (Huang, F. et al., 2013, Video-rate nanoscopy using sCMOS camera-specific single molecule localization algorithms, Nature Methods, Vol. 10: pp. 653-658), the real noise performance of the pixel is reproduced to the best possible extent. Advantageously, this opens up the possibility of better utilizing the capabilities of the detector, for example by fully taking into account the noise and flicker effects of the detector, which is almost impossible in methods according to the prior art.

[0047] The configuration of the method according to the invention advantageously reduces or avoids the disadvantages known in the prior art. Therefore, according to the prior art, the calibration and positioning of the detector requires high computing power. In addition to the measurement and computing overhead of the correction by, for example, 25 million frames (US 9,769,399 B2 and Huang et al. in 2013; see above), the noise of each pixel must also be considered as a convolution of shot noise (Poisson distribution) and pixel-related noise (Gaussian distribution) under the likelihood function in the "maximum likelihood estimation" based on the calibration measurements. Since this must be implemented for each pixel in each positioning process, the computing overhead is very high. Therefore, the noise distribution can be described by an analytical approximation; however, the latter is only accurate enough for those pixels that behave according to the assumed model. Pixels that deviate significantly from the model cannot be taken into account by adjusting the parameterized model, and then the pixel leads to a completely wrong prediction of the model.

[0048] The invention has been explained in an exemplary manner based on the characteristics of sCMOS sensors. However, this can also be applied to different types of pixel-based detectors. For example, the above problems and effects are also applicable in principle to CCD sensors and EMCCD sensors, but the effects are usually less prominent. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] The present invention is explained in more detail below based on the accompanying drawings and various configurations. In the accompanying drawings:

[0050] Figure 1a An exemplary schematic diagram showing captured pixel values ​​with bad pixels and emitting molecules as signal sources, and

[0051] Figure 1b Shows Figure 1a An exemplary diagram of a histogram of the determined PSF width of a signal source.

[0052] Figure 2 A time series diagram showing the gray value distribution of three different pixels of the sCMOS sensor under constant illumination;

[0053] Figure 3 A schematic diagram showing a method of using a threshold value according to the present invention;

[0054] Figure 4 A schematic diagram showing the calibration of a detector with different possible calibration parameters and the generation of a calibration data record;

[0055] Figure 5 A schematic diagram of an exemplary embodiment of processing pixel values ​​and calibration data records is shown. DETAILED DESCRIPTION

[0056] exist Figure 1aAn exemplary schematic diagram of a two-dimensional arrangement of pixels 1 of a detector 2 (e.g., an sCMOS sensor) is provided in the indicated perspective illustration. The different captured grayscale values ​​of the individual pixels 1 (three of which are highlighted by the additional frame) are apparent. Two particularly bright pixels 1 are selected as potential signal sources 3.1 and 3.2, and the origin region 4 of the signal sources is visualized in each case by a circle.

[0057] Figure 2 The temporal noise performance of three different pixels 1 of an sCMOS sensor at a constant illumination level is illustrated in an exemplary manner. In each case, the left column plots the grayscale value over time. The right column correspondingly presents the associated histogram of the frequency of the individual grayscale values.

[0058] The first row shows the noise performance of pixel 1, which is well described by models known in the art, such as the convolution of a Poisson distribution and a Gaussian distribution.

[0059] The noise behavior illustrated in the last row can only be described unsatisfactorily by the corresponding model due to its greater width and approximately triangular form with a large base width.

[0060] The middle row shows the noise performance of pixel 1, whose distribution - as is apparent on the right in the associated histogram - cannot be captured by known models.

[0061] exist Figure 3 A possible configuration of the method using threshold values ​​according to the invention is presented in an overview diagram in FIG. First, the threshold value to be applied is set. Here, the experiment to be performed is taken into account. As an example, the properties of the sample and emitter to be used, the illumination wavelength and (multiple) illumination levels to be used, the (multiple) detection wavelengths to be used, and the known specifications of the detector and / or the remaining experimental device are taken into account.

[0062] Starting from this, the expected positioning accuracy and the tolerance to be accepted are determined or estimated. The threshold value is set according to these agreements and specifications and is stored, for example, in the evaluation unit 5. For example, the threshold value is set as an offset or as a variance of the pixel value so that the threshold value can be compared with a pixel-specific error parameter determined from the pixel value.

[0063] In an alternative configuration of the method, the threshold value can also be determined empirically by means of a localization algorithm.

[0064] Furthermore, calibration data records are created and stored in such a way that they are assigned to the individual pixels 1 of the detector 2 (see Figure 1a). For this purpose, the detector 2 records a number of dark images and the pixel values ​​of the individual pixels 1 captured in the process are stored in an allocated manner as a calibration data record. Subsequently, for example, a shift or a variance of the pixel values ​​is determined from the pixel values ​​as pixel-specific error parameters. These can also be stored in a manner allocated to the pixels 1 in the calibration data record.

[0065] The value of the pixel-specific error parameter thus determined is compared with a threshold value for the respective pixel 1. If the value of the pixel-specific error parameter exceeds the threshold value, the relevant pixel 1 is marked in the calibration data record, for example by means of its pixel coordinates (x, y) listed in the metadata. The marked pixel 1 should not be taken into account in the subsequent evaluation and is therefore "masked", ie marked as no longer being taken into account (pixel masking).

[0066] In a further step, it is checked whether there are any obscured pixels 1 in the cluster. For this purpose, the maximum allowed cluster is predefined (specified) and also stored. As an example, the proximity of two marked pixels 1 is set to a minimum value that must not form a cluster.

[0067] If the conditions for an allowed cluster are met, a transition is made to the localization of the signal sources 3.1, 3.2.

[0068] In contrast, if the distribution of shaded pixels 1 does not satisfy the allowed clustering, it is checked whether the allowed clustering (clustering specification) can be achieved by modified settings of the operating parameters (specifications) of the detector 2. If this is not the case, the detector 2 (=sensor) is unsuitable.

[0069] In contrast, if the operating parameters of the detector 2 can be set such that allowed clusters are acquired, these settings are performed and transition is made to the localization of the signal sources 3 . 1 , 3 . 2 .

[0070] As an example, a calibration data record and information about the marked pixels 1 (pixel masking) are provided as metadata and appended to the image data record to be created.

[0071] For the sample to be evaluated, a time series of image data (time series for localization microscopy) is captured and the pixel values ​​assigned to the corresponding pixels 1 are stored as image data records. To the obscured pixels 1, pixel values ​​determined by interpolation based on the pixel values ​​of the neighboring pixels 1 are assigned, or no pixel value is available for the obscured pixels 1 and is ignored in the subsequent localization of the signal sources 3.1, 3.2. Alternatively, the corresponding captured pixel values ​​can also be assigned to the marked pixels 1, for example in order to be able to carry out a separate error evaluation. However, for the actual localization, these pixels 1 are ignored, or their pixel values ​​for localization are previously determined by interpolation and assigned.

[0072] Based on the calibration data record and optionally by taking into account the metadata, the pixel values ​​of the image data record are corrected (corrected according to the metadata). Based on the image data (in particular the pixel values), the origin region 4 in which the signal sources 3.1, 3.2 are located is determined within the array of pixels 1. As an example, this is achieved based on the maxima of the captured and corrected pixel values. A localization algorithm is applied to the corrected pixel values ​​of the pixels 1 of the corresponding origin region and a PSF is fitted. The position of the maximum of the PSF is determined and stored as the origin of the signal sources 3.1, 3.2.

[0073] By way of example, Figure 4 The calibration of the detector with different possible calibration data records and the generation of an overall calibration data record are illustrated. For this purpose, a time series with different illumination levels I1 to Ik is captured. Here, the illumination level I1 is equal to zero and corresponds to a time series of dark images. Figure 3 As explained, the pixel values ​​captured at illumination level I1 are used to generate a calibration data record [Cal Interp(mask)] with marked and masked pixels.

[0074] Furthermore, the pixel values ​​captured at the illumination level I1 can be used to generate a calibration data record (Cal offset) in which the offset of the individual pixels 1 is determined and stored. The pixel values ​​can be corrected by subtracting the mean value of the pixel values ​​from the individual pixel values. In order to avoid negative pixel values ​​in the process or if, depending on the data format, negative values ​​are not allowed (so-called "zero clipping"), a constant and known value ("NoiseMargin") is optionally added to all pixels and stored.

[0075] Starting from the individual time series, the mean and variance of the pixel values ​​can also be determined in each case within the time series. The mean and variance are used to create and fit (photon transfer curve fitting) a photon transfer curve (PTC). The created photon transfer curve is stored in a calibration data record (Cal gain correction).

[0076] Furthermore, ADU histograms [Histogram (I1), Histogram (I2), ...; Histogram (Ik)] can be created in each case from the pixel values ​​of the time series and stored. These ADU histograms are then optionally combined, smoothed, approximated and / or filtered in a suitable manner (binning) before being stored as a calibration data record (Cal (histogram)).

[0077] The aforementioned calibration data records can be combined to form an overall calibration data record. Alternatively, they can also be generated and / or stored and used individually.

[0078] Figure 5The application of various calibration data records to the captured image data of an image data record is shown in an exemplary manner. The image data of the image data record represented by T(raw data) are corrected by means of an offset calibration data record regarding the offset (Cal offset) and by means of a gain calibration data record regarding the photon transfer (Cal gain correction). The pixel values ​​of the captured image data thus corrected are subsequently corrected using a CalInterp(masked) calibration data record (masked). As already explained above, the variance of pixel 1 exceeds the threshold value set for the Cal Interp(masked) calibration data record, which pixel 1 is marked in the Cal Interp(masked) calibration data record and in the image data record and optionally masked.

[0079] Based on the corrected pixel values, possible signal sources are identified (peak finder) and the associated region of origin 4 (extraction ROI; ROI = Region of Interest).

[0080] The origin of the signal sources 3 . 1 , 3 . 2 in the respective origin region 4 is determined by fitting the PSF [PSF-Fit (Gaussian)] and the quality of the fit is checked.

[0081] Furthermore, a filtering step (Filtering) can be implemented after the actual positioning (Positioning). Using this step, "bad pixels" can be identified, whose grayscale values ​​and noise performance do not correspond to those of the emitters used in the experiment. Bad pixels are understood to be hot pixels, scintillators and dead pixels or other types of pixel behavior that lead to apparent photon detection events. Such bad pixels could have been identified and positioned as ordinary signal sources 3.1, 3.2, for example as fluorescent molecules, despite the previously applied calibration and positioning steps.

[0082] As examples, the number of captured photons, the signal-to-noise ratio and the blinking behavior can be used as filter parameters. The width of the PSF is also a suitable filter parameter, since the latter is largely independent of the respective experiment.

[0083] In particular, the latter filtering is based on the fact that the localization of a single signal source requires a spread of the PSF over a plurality of pixels 1, which occurs through the effects of the optical design of the system (e.g. objective lens, tube lens and pixel size). If a significantly narrower PSF width occurs at an individual pixel 1 localized by the algorithm, which corresponds, for example, to one pixel, it can be identified as a bad pixel by appropriately selecting the filter threshold for the PSF width and can be excluded.

[0084] As an example, in Figure 1a It is obvious that there are two signal sources 3.1 and 3.2, whose corresponding origin areas 4 are drawn with circles. The size of the pixel 1 in the sample is 100nm. Figure 1b The PSF widths (full width at half maximum, Gaussian fit) of the two signal sources 3.1, 3.2 for 1000 localizations of each of the two signal sources 3.1, 3.2 are shown in an exemplary manner in FIG. Figure 1a The PSF width of the signal source 3.1 on the upper left is only about 40nm (range: 30-50nm), but it is located Figure 1a The PSF width of the signal source 3.2 at the bottom right is about 140 nm, which is typical for fluorophores.

[0085] Therefore, it can be assumed that the left signal source 3.1 is caused by a bad pixel (hot pixel or warm pixel or scintillator), while the right signal source 3.2 is actually a fluorophore.

[0086] In principle, this filtering function can also be performed without the preceding steps of the method according to the invention.

[0087] Reference numerals

[0088] 1 pixel

[0089] 2 Detector

[0090] 3.1 Signal Source

[0091] 3.2 Signal Source

[0092] 4 Origin Region

[0093] 5 Evaluation Units

Claims

1. A localization microscopy method for localizing a signal source (3.1, 3.2), wherein: At least once for each pixel (1) of a detector (2) for capturing the detection radiation, a value of a pixel-specific error parameter is determined and stored in a calibration data record in a manner assigned to the relevant pixel (1); capturing image data of the sample pixel by pixel as pixel values ​​and storing in an image data record; Identifying an origin region (4) of a signal source (3.1, 3.2) based on the captured image data, the origin region comprising a plurality of pixels (1); fitting a point spread function to the pixel values ​​of the corresponding origin region (4); and The corresponding signal sources (3.1, 3.2) are located within the relevant origin region (4) based on the point spread function; It is characterized in that For each pixel (1), comparing the pixel-specific error parameter with a threshold value; marking in the calibration data record each pixel (1) for which the value of the pixel-specific error parameter is greater than the threshold value; and When fitting the point spread function, the marked pixels (1) in the image data record are ignored or replaced by interpolation.

2. The method according to claim 1, characterized in that The threshold value is set based on the wavelength of the detection radiation, the intensity of the detection radiation and / or the temperature of the detector (2).

3. The method according to any one of claims 1 and 2, characterized in that On average, a predetermined maximum number of pixels (1) is marked in each origin region.

4. The method according to any one of the preceding claims, characterized in that The pixel values ​​of the unmarked pixels (1) are corrected using the calibration data record.

5. A localization microscopy method for localizing a signal source (3.1, 3.2), wherein: At least once for each pixel (1) of a detector (2) for capturing the detection radiation, a value of a pixel-specific error parameter is determined and stored in a calibration data record in a manner assigned to the relevant pixel (1); storing image data of the sample captured pixel by pixel as pixel values ​​in an image data record; Identifying an origin region (4) of a signal source (3.1, 3.2) based on the captured image data, the origin region comprising a plurality of pixels (1); Fitting the point spread function to the pixel values ​​of the corresponding origin region (4); and Based on the point spread function, the corresponding signal source (3.1, 3.2) is located in the relevant origin area; It is characterized in that Using the calibration data record, Determine the error parameter of the derivative of (1) for each pixel, and / or creating an ADU histogram of a plurality of pixel values ​​for each pixel (1); and The error parameters derived in the calibration data record and / or the ADU histogram are stored in such a way that they are assigned to the corresponding pixels (1); The point spread function is fitted based on the derived error parameter or the corresponding ADU histogram.

6. The method according to claim 5, characterized in that The variance of the pixel values ​​of the time series, the variance of the pixel values ​​of the time series at different illumination levels and / or the mean value of a respective time series at different illumination levels are determined as the derived pixel-specific error parameter.

7. The method according to claim 6, characterized in that A photon transfer curve is determined for each pixel (1) based on the variance and the mean of the time series at different illumination levels and is used as a derived pixel-specific error parameter.

Images