Methods and systems for processing biological specimens utilizing multiple wavelengths

The present invention relates to systems and methods for characterization or classification of biological specimens.

In the medical industry, there is often a need for experimental technicians, such as cytotechnologists, to review cell specimens for the presence of certain types of cells. For example, there is currently a need to review cervical-vaginal Papanicolaou (Pap) smear slides. Pap smears are a powerful tool for detecting cancerous and precancerous cervical lesions. The reliability and effectiveness of cervical screening and screening of other specimens is assessed by their ability to diagnose precancerous lesions (sensitivity) and at the same time avoid false-positive diagnoses (specificity). As a result, these criteria depend on the accuracy of the cytological interpretation.

Traditionally, pathologists have used single-cell analysis of biological specimens by examining the characteristics of individual cell nuclei, or contextual analysis of biological specimens by exploring characteristic patterns of cell structure appearing on slides. )It can be performed. To facilitate this review process, automated screening systems have been developed that process multiple microscope slides. In a typical system, an imaging device (imager) is operated to obtain a series of images of cell specimen slides, each showing different parts of the slide. The processor or controller then processes this image data to provide quantitative prognostic information about the sample. In providing this diagnostic information, the processor can perform either single cell analysis, context analysis, or both.

In some automated screening systems, the processor uses diagnostic information to draw a line between normal and abnormal or suspicious biological material in each sample. That is, the processor has the ability to focus the cytotechnologist on the cells that are most appropriate and exclude the remaining cells from further review. In this case, the screening device uses the diagnostic information to determine the most appropriate biological object and its position on the slide. This location information is provided to a review microscope, which automatically advances to the identified location and focuses on the biological object for review by a cytotechnologist. The cytotechnologist then electronically labels the most appropriate biological object (eg, an object with properties consistent with malignant or premalignant cells) for further review by a pathologist. Can be done.

For example, in one automated system, an object or "object of interest" (OOI) is identified based on image data. The object or OOI may be in the form of individual cells and cell clusters of the specimen. The system can be configured to classify identified areas or objects, for example, based on the degree of risk that a particular cell or object has an abnormal condition, such as a malignant tumor or premalignant tumor. .. For example, the processor can evaluate an object for nuclear integrated or average optical density and rank the object according to its optical density value. The object, along with its relative rank and coordinates, can be stored for subsequent processing, review, or analysis. Further aspects of known imaging systems and methods for processing image data and OOI are described in Patent Document 1.

In general, the use of automated screening systems is efficient because the technician's attention is focused on suspicious slides, or a limited number of more suitable objects in each slide. However, automated screening systems can be improved. For example, the way automated systems handle artifacts can be improved to reduce the percentage of false positive or "false error" results. The artifact can be regarded as an object that has no diagnostic value. One cause of false positives is the presence of artifacts, which can be abundant in sample samples and can be large dark objects similar to abnormal samples. Artifacts can be ranked higher than objects containing normal cells.

For example, compared to abnormal nuclei, normal nuclei usually have less DNA and less texture. In the absence of artifacts in the top ranked objects, the majority of cells on a normal slide have a tightly distributed amount of DNA. However, a large number of artifacts that resemble abnormal cells are ranked higher than most normal cells, and these artifacts give false warnings in data modeling. Due to these artifacts, the actual cells may be ranked and unable to be correctly presented in the list of cells with the "top" amount of DNA. Therefore, the automated system does not select the cells to be reviewed, but rather mistakenly recognizes the artifact as an abnormal cell and selects an artifact that ranks higher than the abnormal nucleus. This results in fewer selections of objects that actually have cells, and fewer abnormal objects that deserve review by a cytotechnologist, resulting in less accurate analysis and diagnosis and false positives. It can be a technician.

The occurrence of false positives is sometimes due to limited capabilities or configurations of automated imaging devices. That is, the automated imaging device can be limited by the specimens and data provided to it, as well as its programming. For example, for computational reasons, imaging devices typically use monochrome black-and-white images for analysis. Examples of known monochrome systems are Becton Dickinson Company, 1 Becton Drive, Franklin Lakes, New Jersey, and Cytic Corporation, 250 Campus Drive, Marlborough, Massachusetts (Cytyc). Corporation, 250 Campus Drive, Marlborough, Massachusetts), more available. However, the sample may provide a wide range of spectral data and other information that can be used to characterize or classify the sample. However, this other data is not available when using a monochrome imaging and analysis system.

<p><patcit num="1"><text>U.S. Patent Publication No. 2004/0254738A1</text></patcit></p>

One aspect relates to a method for classifying a biological sample on a specimen carrier and determining if the sample requires further analysis. The method involves obtaining an image of the object in the sample and identifying the object of interest in the image. The method further involves obtaining additional images of the identified object of interest at multiple different wavelengths, extracting cellular features of the identified object of interest from the additional images. And to classify the specimens according to a probabilistic model based on the characteristics of the extracted cells and determine if the specimens require further analysis.

Another aspect relates to a method for automatically classifying a biological sample on a sample carrier to determine if the sample requires further analysis. This method involves acquiring an image of the object in the sample and identifying the object of interest from the acquired image. This method further captures additional images of the identified object of interest at multiple different wavelengths and extracts the nuclear features of the identified object of interest from the additional images. This includes classifying specimens according to a probabilistic model based on the characteristics associated with the extracted nuclei. The probabilistic model includes first and second probabilistic functions. The first probability function indicates the probability that the identified object of interest is an artifact, and the second probability function is based in part on the result of the first probability function. By using a combination of first and second probability functions, a sample is classified and whether the sample needs further analysis is determined.

A further aspect is to obtain an image of an object in a biological sample and identify a target object in the acquired image with respect to a method of processing the biological sample using light of multiple wavelengths. including. The method further captures additional images of the identified target object at multiple different wavelengths and extracts cellular features of the identified target object from the additional images. ,including.

Yet another aspect relates to a method of classifying a biological specimen using light of multiple wavelengths, obtaining an image of an object in the biological specimen, and identifying a target object in the acquired image. That includes. The method further comprises acquiring additional images of the object of interest in the biological specimen at multiple different wavelengths, extracting cellular features of the object of interest from the acquired images, and extracting. Includes classifying biological specimens based on the characteristics of the cells.

A further aspect relates to a method of classifying a biological sample using light of a plurality of wavelengths, obtaining an image of a target object of the biological sample at a plurality of wavelengths, and acquiring cell characteristics of the target object. Includes extracting from the extracted images and classifying biological specimens based on the characteristics of the extracted cells.

According to another aspect, the biological screening system for classifying the biological specimen held on the specimen carrier and determining whether the biological specimen requires further analysis is operably linked to the imaging member and the imaging member. Includes processors. The imaging member is configured to acquire digital image data of the object in the biological sample, and the processor is configured to process and identify the object of interest from the digital image data. The imaging member is also configured to acquire additional images of the identified object of interest at multiple different wavelengths. The processor extracts the cellular features of the identified target object from additional images, and classifies the biological specimens according to a probabilistic model based on the extracted cellular feature carriers, and the biological specimens undergo further analysis. Further configured to determine if necessary.

According to a further aspect, the biological screening system for classifying the biological specimen held on the specimen carrier and determining whether the biological specimen requires further analysis is an imaging member and a processor operably linked to the imaging member. including. The imaging member is configured to acquire an image of the object in the biological specimen, and the processor is configured to process and identify the object of interest from the acquired image. The imaging member is further configured to capture additional images of the identified object of interest at multiple different wavelengths. The processor will extract the nuclear-related features of the identified object of interest from additional images acquired at different wavelengths and classify the biological specimen according to a probabilistic model based on the measured cellular features. Is further configured. The probabilistic model includes first and second probabilistic functions. The first probability function indicates the probability that the selected object is an artifact, and the second probability function is based in part on the result of the first probability function. By using a combination of first and second probability functions, a biological sample is classified to determine if the biological sample requires further analysis.

Yet another alternative aspect relates to a biological specimen classification system that includes an imaging member and a processor operably linked to the imaging member. The imaging member is configured to acquire an image of the target object of the biological sample at multiple different wavelengths, and the processor extracts cell-related features from the acquired image and turns it into the extracted cells. It is configured to classify biological specimens based on relevant characteristics.

In one or more embodiments, the extracted or measured cellular features are those associated with the nucleus, such as standard deviation of optical density in the nucleus, variation in optical density in the nucleus, nuclear. Corrected optical density and the shape of the nuclear boundary.

In one or more aspects, the probabilistic model used for classification includes two probabilistic functions, eg posterior probabilistic functions. One probability function indicates the average probability that the identified object of the biological specimen is an artifact, and the other probability function is based in part on the result of the first probability function. Both probability functions may be based on features related to different numbers and types of extracted nuclei. For example, the first probability function is one of the structure of the nucleus, the standard deviation of the optical density within the nucleus, the variation of the optical density within the nucleus, the corrected optical density of the nucleus, and the shape of the boundary of the nucleus. Alternatively, it may be based on two or more, or all, and the second probability function is the gray value contrast of the pixels of the result of the first probability function, as well as the image of the cell nucleus of the identified object of interest ( gray value It may be based on the average of contrast) and one or more of the range of pixel gray contrasts of the image of the cell nucleus of the identified object of interest. The results of the first and second probability functions are plotted or displayed in a graph format to classify biological specimens and determine if the specimens need further review or which specimens in the group of specimens need further review. Can be decided.

Here we refer to the drawings in which the same signs represent the corresponding parts throughout:

<figref num="1">FIG. 1 is a block diagram of a system for classifying biological specimen slides and deciding whether to analyze or review the slides according to one aspect.</figref><figref num="2">FIG. 2 is a plan view of the biological sample slide.</figref><figref num="3">FIG. 3 is a block diagram of a system for classifying biological specimen slides by a probabilistic model that follows one embodiment using light of multiple wavelengths.</figref><figref num="4">FIG. 4 is a flow chart of a method for classifying biological specimen slides by a probabilistic model that follows one embodiment using light of multiple wavelengths.</figref><figref num="5">FIG. 5 is a diagram generally showing a portion of an object of a biological specimen containing cells and artifacts.</figref><figref num="6">FIG. 6 is a diagram showing a probability model including a plurality of probability functions used in various aspects.</figref><figref num="7">FIG. 7 is a flow chart generally showing a first probability function used in various aspects to determine the probability that an object is an artifact.</figref><figref num="8">FIG. 8 is a flowchart of the first probability function according to another aspect.</figref><figref num="9">FIG. 9 is a flow chart generally showing a second probability function used in various embodiments to determine the probability that a biological sample is normal.</figref><figref num="10">FIG. 10 is a flowchart of the second probability function according to another aspect.</figref><figref num="11">FIG. 11 is a diagram that generally shows how the data obtained by the first and second probability functions can be plotted on a graph.</figref><figref num="12">FIG. 12 shows test data points representing normal and abnormal slides plotted in the graph shown in FIG.</figref><figref num="13">FIG. 13 is a table showing the objects identified during the test shown in FIG.</figref><figref num="14">FIG. 14 is a diagram showing an alternative biological specimen analysis system that can realize the embodiment and includes an imaging station, a server, and a review station.</figref><figref num="15">FIG. 15 shows an alternative biological specimen analysis system that can implement aspects and includes integrated imaging and review capabilities.</figref>

Referring to FIGS. 1 and 2, the multi-wavelength biological specimen screening systems 100 and method according to the embodiment are carried on slide 110 or other suitable carrier (eg, cells). Specimen) 112 images are taken, an object of interest (OOI) as an example is identified, and then multiple images of the selected object or OOI (generally referred to as "OOI") are displayed. Acquired at different wavelengths. Data related to cell characteristics, such as those associated with the cell nucleus of OOI, are determined, extracted, or measured from multi-spectral images. To simplify the description, the extraction of cellular features of OOI will be described. This data can be used for classification purposes and for other applications.

According to one embodiment, screening cis Temu 100 in accordance with the probabilistic model 120 using data on the characteristics of the nuclei that have been extracted from the image of OOI acquired at a plurality of wavelengths, "normal" or "doubtful" a biological specimen 112 Configured or programmed to classify as. For example, as shown in FIG. 1, when the number of biological sample slides 110 to be processed is 1000, the aspect of the sample classification system 100 uses a plurality of wavelengths and automatically classifies the slides 110 according to the probability model 120. This allows certain slides, eg slides 1-300 and 700-1000, to be classified as normal slides 130 and other slides, eg slides 301-699, as suspicious slides 140. Not surprisingly, the numbers and groupings of normal and suspicious slides 130, 140 are provided for explanatory purposes. Thus, in the case of FIG. 1, the "normal" 130 sample or slide is the sample 112 or slide 110 that does not require further review or analysis by a cytotechnologist. Generally, slide 110 is classified as normal 130 if it has a non-cancerous or non-cancerous sample 112. The "suspicious" 140 specimens or slides may or may be abnormal and may contain cancerous or precancerous cells.

Classification as "normal" 130 or "suspicious" 140 is shown in FIG. 1 as the final determination or result of the embodiment, that is, whether specimen 112 or slide 110 requires further analysis or review. The cytotechnologist can then focus his attention on the suspicious slide 140 instead of the normal slide 130. By this method, valuable information about the sample, slide 110, is obtained by using multiple wavelengths, which is used to reduce the frequency of false positives due to artifacts and classify the sample 112, thereby further. The result is that slide 110 is more accurately and efficiently categorized and selected for review.

According to one embodiment, the multi-wavelength biological screening system 100 is configured to process a series of microscope slides 110 having biological or cellular specimens 112 such as cell neck or vaginal specimens (usually found on Pap smear slides). To. In this case, the cells are abnormal, malignant or premalignant tumors such as mild intraepithelial neoplasia (LGSIL) or severe intraepithelial neoplasia (HGSIL), and any other cytological division (eg, infection, cytolysis). May be reflected. Specimen 112 is typically placed on slide 110 as a thin cytological layer. Preferably, a cover glass (not shown) is adhered to the specimen 112, whereby the specimen 112 is secured in an appropriate position on the slide 110. Specimen 112 may be stained with any suitable dye, such as ThinPrep® nuclear dye.

In addition, using aspects, blood, urine, semen, milk, sputum, mucus, pleural fluid, pelvic fluid, saliva, ascites, body cavity washes, eye brushing, skin scrapings. ), Buccal wipes, Intravaginal wipes, pap smears, Rectal wipes, Aspirations, Needle biopsy specimens, such as tissue fragments obtained by surgery or autopsy, plasma, serum, spinal fluid, lymph, external skin "Normal" of other types of biological specimens 112, including secretions, airways, intestines, and urogenital tracts, tears, saliva, tumors, organs, microbial cultures, viruses, and samples of cell culture components in vitro. Tracing or classification as 130 or "suspicious" 140 can also be performed using multiple wavelengths. This specification describes one method in which an embodiment can be achieved with reference to a cell neck or vaginal specimen 112 (on a Pap smear slide), the embodiment being applied to various types of tissues and cells. Please understand that it is possible. Further aspects of aspects of the system and method will be described with reference to FIGS. 3-15.

Referring to FIG. 3, in one embodiment, the biological specimen screening system 300 for classification or characterization of the biological specimen slide 110 is lighted through a camera or other imaging member 310, a microscope 320, a biological specimen slide 110 and a microscope 320. Includes a light source 330 that provides the 332 to the camera 310, an electric stage 340 that supports the biological specimen slide 110, an image processor 350, and an attached memory or storage device 360 (commonly referred to as the memory 360). The memory 360 may be a part of the image processor 350 or may be a separate component.

The image processor 350 and / or the memory 360 can store or access the probability model 120 and use it to classify the biological specimen 112 as "normal" 130 or "suspicious" 140. The probabilistic model 120 may be in the form of hardware, software, or a combination thereof. For example, the probabilistic model 120 may be in the form of a series of programmed instructions and / or data stored in memory 360 and executed by the image processor 350 together with sample data acquired by the image processor 350. The probabilistic model 120 may be run on a separate processor or part of the controller.

The camera 310 may be one of a variety of known digital cameras, and the light source 330 may include a single light source or a plurality of individual light sources, as shown in FIG. FIG. 3 shows three light sources 330a-c (generally referred to as light sources 330) to generally illustrate that the system 300 may include a plurality of light sources 330. As discussed in more detail below, different numbers of light sources 330 can be used as needed to achieve the desired multi-wavelength analysis of OOI, and Figure 3 shows multiple systems 300. It should be understood that the light source 330 of the is provided to generally explain that it may be included.

According to one aspect, the screening and classification system 300 includes a plurality of light sources 330, each of which emits light 332 of a different wavelength. A suitable light source 330 for this purpose would be a light emitting diode (LED). In another aspect, the light source 330 has one or more prisms (not shown) and / or one or more so that the light 332 transmitted through the filter 334 has the desired wavelength or wavelength range. It may be paired with one or more other optics, such as an optical filter 334. Examples of filters that can be used in embodiments include dichroic filters, interference filters, and filter wheels. The filter 334 can be adjusted or selected to vary the wavelength of the light 332 provided to the microscope 320 and the camera 310 for imaging a portion of the specimen 112. A liquid crystal tunable filter can also be used. Further aspects of the components of a suitable system 300 are described in US Pat. No. 2004/0253616 A1.

Referring to FIG. 4, in method 400 according to one embodiment, light 332 is directed from light source 330 through stage 340 and microscope 320 to camera 310, which camera is a digital image of biological specimen 112 in step 405. To get. More specifically, the slide 110 with the cell specimen 112 is mounted on an electric stage 340 that moves or scans the slide 110 with respect to the field of view of the microscope 320, while the camera 310 emits light emitted by the light source 330. 332 captures an image of the whole or part of the biological specimen 112. For example, each pixel of each image captured by camera 310 can be converted to an 8-bit value (0 to 255) depending on its light transmittance, where "00000000" is the least amount of light transmitted through the pixel. Is assigned to the amount of light, and "11111111" is assigned to the amount of light that passes through the pixel most. The shutter speed of the camera 310 is preferably relatively fast so that the scanning speed and / or the number of images acquired can be maximized. The biological specimen 112 held on the slide 110 may be housed in a cassette (not shown), in which case the slide 110 is taken out of each cassette in a contiguous manner to obtain a digital image. Then it is returned to the cassette.

In step 410, the acquired image or image data is provided to the image processor 350, which uses the most appropriate or highest ranked object, also known as the object of interest (OOI). Perform various operations on the image or image data 312 to identify. For example, the image processor 350 can identify about 20 or 40 objects that appear to be most relevant, or any other appropriate number of objects. To this end, the image processor 350 performed the primary and secondary segmentation described in US 2004/0253616 to measure various characteristics for each of the individual and clustered objects. It can be determined or extracted and subsequently the object score for each object can be calculated based on the measured values of these features. Based on this score, the image processor 350 can identify or select objects that are considered to be objects of interest (OOI) and clustered objects, the location of which is in memory for future reference. Can be memorized in.

Referring to FIG. 5, the image processor 350 attempts to identify an OOI (a portion thereof or which is shown in FIG. 5) containing cell 500 containing cytoplasm 502 and nucleus 504, but the selected OOI is , Rather, it could be artifact 510. Selection of artifact 510 may result in exclusion of other cell-containing objects to be identified and analyzed as OOI. Aspects favorably determine how likely the OOI is to be artifact 510 for the purpose of classifying sample 112 as normal 130 or suspicious 140. Aspects can use this information to improve the review of biological specimen 112 and replace artifact 510 with a cell-containing object, eg, a next-ranked object that contains cells 500 and is not an artifact.

More specifically, with reference to FIG. 4 again, after imaging and identifying multiple OOIs from the first sample image in step 415, the system 300, as discussed above with reference to FIG. Proceed to the acquisition of additional images of OOI at multiple different wavelengths using a separate light source 300, or one or more filters 334. According to one aspect, each OOI can be imaged at about 3 to about 30 different wavelengths, for example 19 different wavelengths. According to one aspect, the wavelength range used for OOI imaging may be from about 410 nm to about 720 nm, for example from about 440 nm to about 720 nm. In step 420, the image processor 350 performs various operations on the multi-wavelength digital image of the OOI to measure, determine, or extract various nucleus-related features from the multi-wavelength OOI image. Further details regarding the acquisition of additional images of OOI at different wavelengths and the extraction of features from these images are provided in US Patent Publication No. 2006/0245630A1. In step 425, biological sample 112 is classified or characterized as normal 130 or suspicious 140 according to a probabilistic model 120 that uses nuclei-related features extracted from multi-wavelength OOI images.

Referring to FIG. 6, according to one embodiment, the biological sample 112 is classified or characterized according to a probability model 600 using a plurality of probability functions (eg, a plurality of posterior probability functions). According to one aspect, the probability model 600 can also be used to show how likely it is that the OOI is artifact 510 and that the biological sample 112 is normal 130. This data is then used in combination to classify whether sample 112 is normal 130 or suspicious 140.

In the aspect shown in FIG. 6, the probability model 600 includes two different probability functions 610 and 620, which may be posterior probability functions according to one aspect. The probability functions 610 and 620 both use nuclear-related features 612 and 622. According to one aspect, the probability functions 610, 620 use different features 612, 622 related to the nucleus, as generally shown in FIG.

In the illustrated embodiment, the second probability function 620 uses the results of the nucleus-related feature 622 and the first probability function 610. That is, the first probability function 610 is based on the nucleus-related feature 612 and is independent of the second probability function 620, while the second probability function 620 is the nucleus-related feature 622 and the second. One probability function 610 is used. 7 and 8 further show the first probability function 610 according to the aspect, and FIGS. 9 and 10 further show the second probability function 620 according to the aspect.

Referring to FIG. 7, according to one embodiment, method 700 for classifying or characterizing biological specimen 112 using a probabilistic model is used to determine the probability or likelihood that the selected OOI is artifact 510 rather than cell 500. Includes the first probability function 610 used. An artifact among the top-ranked objects by determining the probability of how likely it is that the top-ranked object or OOI is artifact 510 rather than cell 500. You can determine the percentage that 510 is present.

The method 700 according to one embodiment comprises extracting or measuring the nucleus-related features 612 of each OOI using images of OOIs acquired at multiple different wavelengths. Measuring nuclei-related features extracted from multi-wavelength images provides more detailed information about cells in OOI compared to features extracted from a single gray-level or monochrome image. In one embodiment, this involves measuring or determining nuclear-related features 612, including, for example, nuclear structure, optical density, and shape, based on images of OOI obtained at multiple wavelengths. Structure means the value of any pixel compared to adjacent pixels. The optical density is a measured value of absorbance. Fluctuations in optical density in multi-wavelength images provide important information used, for example, to determine how likely OOI is to be artifact 510. The shape means the irregularity of the outer shape of the nucleus. It is possible that certain features may be extracted from a single wavelength image and other features may be extracted from a multi-wavelength image. For example, structural features can be extracted from single-wavelength images (eg, about 570 nm), and optical density features can be extracted from multi-wavelength images (eg, about 520 nm and 630 nm).

In step 710, the extracted feature measurements 612 are used to determine the probability that the OOI is artifact 510. Do this for each OOI. Thus, the result of step 710 is a set of probability values, with each OOI associated with a particular probability value (eg, a decimal or a percentage). Next, in step 715, the average probability is determined by calculating the average of the probability values obtained in step 710. For example, if the probability that the first OOI is artifact 510 is 0.4, the probability that the second OOI is artifact 510 is 0.8, and the probability that the third OOI is artifact 510 is 0.5, then the OOI The average probability that (or any other object) is an artifact (also known as "APA") is (0.8 + 0.4 + 0.5) / 3, or about 0.57. The result of step 715 is the average probability that OOI for any sample 112 consists of artifact 510 with more than actual cells 500.

Referring to FIG. 8, according to one embodiment, method 800 for determining APA comprises measuring multiple nuclear-related features 612 for each OOI. In one embodiment, this involves measuring or determining nuclear-related features 612, including nuclear structure, optical density, and shape, based on images of OOI obtained at multiple wavelengths. For example, in one embodiment, it measures, determines, or extracts the nuclear structure of each OOI in step 805, and measures, determines, or extracts the standard deviation of the optical density in the nucleus in step 810. To measure, determine, or extract the variation in optical density within the nucleus in step 815, to measure, determine, or extract the corrected optical density of the nucleus in step 820, and to step 825. Includes measuring, determining, or extracting the shape of the nuclear boundary. After having this data, in step 830, the first probability of how likely the selected OOI is artifact 510, eg, the first posterior probability, is calculated. In step 835, the average probability of how likely the selected OOI is artifact 510 is calculated based on the individual probability values determined in step 825.

In the illustrated embodiment, Method 800 comprises five types of nuclear-related features 612. In another aspect, Method 800 may include features 612 associated with a different number of nuclei, such as 612 features associated with less than 5 types of nuclei or 612 associated with more than 5 types of nuclei. In addition, features 612 related to different nuclei other than the five nuclei related features listed in steps 805-825 can also be used. Therefore, FIG. 8 is provided to show an example of the first probability function 610.

Referring to FIG. 9, method 900 to determine how likely it is that biological specimen 112 is normal 130 is how likely it is that the selected OOI in step 905 is artifact 510. Includes determining the average probability of (eg, using the APA discussed with reference to Figures 7 and 8) and then determining the nuclear-related feature 622 in step 910. .. In step 915, the probability that the biological sample 112 is normal 130 is determined based on the combination of the mean probability (step 905) and the measured nucleus-related features (step 910). In step 920, the combination of data obtained by the first and second probability functions is then used to classify the biological sample 112 as normal 130 or suspicious 140.

FIG. 10 shows one aspect of method 1000 for determining how likely it is that biological specimen 112 is normal 130, where the APA, or selected OOI, in step 1005 Includes determining the average probability of how likely it is to be artifact 510, and determining the nucleus-related features 622 that correspond to the gray values of the pixels in the image of the nucleus. This may include determining the average gray value contrast in the nucleus in step 1010 and determining the range of gray value contrast in the nucleus in step 1015. After having this data, in step 1020, the probability that the biological sample 112 is normal 130 is determined based on the first probability function 610 (APA) and the determined gray value contrast data (steps 1010 and 1015). To do. The combination of data obtained by the first and second probability functions is then used to classify biological specimen 112 as normal 130 or suspicious 140.

Referring to FIG. 11, appropriate graphs of the data obtained by the first probability function 610 and the data obtained by the second probability function 620 for the purpose of classifying sample 112 as normal 130 or suspicious 140. Can be plotted at 1100. In the illustrated embodiment, graph 1100 is a two-dimensional graph including the x and y axes. The y-axis 1102 represents the first probability function 610, the average probability or "APA" that the object is an artifact. The x-axis 1104 of graph 1100 represents the second probability function 620, i.e., how likely it is that a biological sample is a normal 130 sample. Therefore, the values on the x and y axes, 1102 and 1104 are expressed as decimal or percentage values (eg 20%, 60%). Thus, the first probability function 610 includes confirming that the normal 130 specimens or slides on the border are most likely due to artifacts, and for the second probability function 620 this Use to classify cases of normal 130 on the borderline (high number of artifacts) as normal 130 and ensure that they can be safely sorted so that they do not require further review or analysis.

FIG. 12 shows a graph in which the data obtained in the tests performed using the first and second probability functions 610 and 620 plotted against each other are input to the graph 1100 shown in FIG. Each test data point in FIG. 12 represents slide 110 with biological specimen 112. Data points were obtained by identifying the OOI. The different types of OOI identified are summarized in Chart 1300 in Figure 13. The first column 1301 of the chart identifies the type of object and the second column 1302 contains a particular type of artifact 510 (identified by grouping 1301) identified as the object. Indicates the number of objects in.

In this particular test, OOI images were obtained at 19 different wavelengths in the range of approximately 440-720 nm. Using this multi-wavelength image, 10 different nuclei-related features were then analyzed to obtain 190 different nuclei-related features. A nucleus-related feature 612 containing shape, optics, and structure was used for the first probability function 610, and a nucleus-related feature 622 containing shape, optics, and structure was used for the second probability function 620. ..

The selection of features may be based on various criteria. In this particular example, feature selection was based on correlating feature values to a pre-assigned group, for example using Pearson's product moment correlation. The pre-assigned groups in this example are "cells" and "artifacts", which process the covariance matrix for these selected features for the cells and artifacts. The Mahalanobis distance from the group average between the test objects is calculated, and if the distance from the group average is the minimum value, the test object belongs to that category. The posterior probability of how likely each object is to belong to the "artifact" category is used to determine the first probability function, or the average probability that the object is an "artifact" ("APA"). calculate.

The resulting data points representing individual slides 110 are plotted on Graph 1100 as shown in FIG. Graph 1100 is divided into four quadrants, 1211, 1212, 1213, 1214 by straight lines 1201 and 1204. The horizontal straight line 1202 is defined by the value of the first probability function 610, and the vertical straight line 1204 is defined by the value of the second probability function 620. Graph 1100 includes data points representing 299 sample slides 110. Normal slides or specimens 130 correctly classified as normal 130 are represented by "X" 1220 and not classified as normal 130, normal slides or specimens 130 classified as suspicious 140 are represented by "circle" 1221 and are normal. Abnormal or suspicious slides or specimens 140 not classified as 130 are represented by "enclosed X" 1222, and abnormal or suspicious slides or specimens 140 classified as normal 130 are represented by "enclosed circles" (1223). Will be done. Therefore, based on the embodiments, in the illustrated example, the specimen corresponding to data point 1220 is reviewed by a cytotechnologist, while the specimen corresponding to data points 1221, 1222, and 1223 is reviewed by a cytotechnologist. Do not receive.

More specifically, of the 299 sample slides 100, 225 slides contained samples that were suspicious or abnormal 140, and the remaining 74 slides contained samples that were normal 130. Aspects were tested to determine how many of these 74 normal 130 sample slides could be correctly classified as normal 130 based on the first and second probability functions 610, 620.

More specifically, since the x-axis 1104 represents the probability that the sample 112 is normal 130, which is determined using the second probability function 620, the sample slide 110 having a sufficiently high x-axis 1104 value. It can be classified as normal 130. Similarly, slide 110 with a sufficiently high y-axis value 1102 is also normal, as artifact 510 often resembles abnormal cells but is not abnormal cells and can therefore be classified as normal 130. It can be classified as 130. Thus, depending on the aspect, slide 110 is of a sufficiently high x-axis 1104 or first probability function (APA) along the x-axis 1104 so that the corresponding data points fall within the range of the upper right quadrant 1211. Based on the value and the corresponding data point with a sufficiently high second probability function (probability of being normal) value along the y-axis 1102, it can be advantageously classified as normal 130.

In the illustrated embodiment, slide 110, which is most likely to be normal 130, is greater than the first value 1202, eg, the value of the first probability function 610 or APA greater than about 0.3, as well as the second value 1204. A slide with a value of a second probability function 620 greater than, eg, greater than about 0.4. Using the sample slide 110 corresponding to these data points within the upper right quadrant 1211 defined by the intersection of straight lines extending through the x and y axes at these points 1202, 1204, the corresponding slide is set to normal 130. Can be categorized (identified by "X" 1220). Therefore, the cytotechnologist does not need to review or analyze the slides corresponding to data point 1220 (X), and the embodiment advantageously excludes these slides from further review.

In the illustrated example, 37 of the 74 normal 130 slides were correctly classified as normal 130 (identified by the "X" 1220). Most of the remaining 37 normal 130 slides (identified by the "circle" 1221) have low values on the x-axis 1104 or the second probability function 620, with corresponding data points in the upper left quadrant 1212. Initially, it was not classified as normal 130, but as suspicious 140 because it was within the range of. The slide corresponding to data point 1221 (circle) will be classified as suspicious 140 and will therefore be identified as requiring further review and analysis by a cytotechnologist. Thus, by aspect, approximately 50% of normal 130 slides (identified by "X" 1220), and approximately 12% of all slides 110 were advantageously excluded from the pool of slides that could be considered by the cytotechnologist.

In the illustrated example, there was one abnormal specimen (identified as the "enclosed circle" 1223) that was incorrectly classified as normal 130 in the upper right quadrant 1211. As a result, this abnormal sample 1223 was classified as normal 130 and was not tested by a cytological technician. However, of the approximately 47 slides identified as normal 130, only one was erroneously classified as normal 130 (identified as the enclosed "circle" 1223). This low error rate is considered to be better than the error rate obtained in a manual review by a cytotechnologist. Therefore, although a small number of anomalous 140 slides may be classified as normal 130 if they should be classified as suspicious 140, the error rate is low enough that such errors are about 50% of normal 130 slides. It is believed that identifying will reduce the burden on the cytotechnologist and offer an acceptable trade-off for the ability of the cytotechnical engineer to focus his attention on the more appropriate suspicious or abnormal slide 140.

Data points within the remaining three quadrants (upper left 1212, lower left 1213, and lower right 1214 quad) were 130 normal specimens (identified as "circles" 1221) that were not initially classified as normal. And represents an abnormal specimen 140 (identified as the enclosed "X" 1222) correctly classified as "abnormal" or "unnormal". The image processor 350 processes this data to generate instructions as to whether a particular slide 110 should be reviewed by a cytological engineer, or to identify which slide 110 requires a cytological engineer's review. A list of slides 110, which should be reviewed because they are not included in the upper right quadrant 1211 and were not initially classified as normal 130 using the first and second probability functions 610, 620. By creating).

The embodiment will be described with reference to the example of the imaging system shown in FIG. 3, but other imaging system configurations may be used, and the imaging system in which the embodiment is realized is the other It may be used with system components, attached to them, or connected to them. For example, referring to FIG. 14, another system that can realize the embodiment is an imaging apparatus 1410 (eg, as shown in FIG. 3, or another suitable imaging apparatus), an image processor 350, a memory 360 and a probability. It has a model 120, a server 1420 that includes other ancillary components such as the FOI processor 1422 and a routing processor 1424, and a review station 1430 that includes a separate microscope 1432 and electric stage 1434. Further aspects of the system configuration shown in FIG. 14 are described in US Patent Application Publication No. 2004/0253616A1. In addition, embodiments can be implemented in independent or stand-alone imaging systems, eg, as shown in FIG. 14, or are available from Psychic Corporation and are commonly shown in FIG.²It can also be achieved with an integrated system 1500 that includes both imaging and review capabilities, such as an imaging / review system.

In addition, the embodiments can be used to process and analyze various types of specimens other than the cell neck or vaginal specimens provided as examples of how the embodiments can be achieved. In addition, embodiments can include specimens housed or held in various specimen carriers, including slides and vials. Furthermore, it should be understood that the embodiments can be applied to the classification of different types of specimens and may be used for other purposes.

Aspects may include first and second probability functions 610, 620 (eg posterior probability functions) based on data obtained from images acquired with different numbers of wavelengths and features associated with different nuclei. .. Various optical members and combinations thereof can be used to generate light of multiple wavelengths. In addition, features associated with different numbers of nuclei can be used for the purpose of determining values using first and second probability functions. Therefore, the first stochastic function using the features related to the five nuclei, and the second stochastic function using the features related to the two nuclei relate to how the aspects can be realized. Provided to illustrate by example, other aspects may include the use of features associated with different types and numbers of nuclei. Further, the probabilistic model may include a modification of the probabilistic model described above.

Further, the embodiment can be used in conjunction with a biological specimen classification system and can be embodied as a computer program product that embodies all or part of the functions described herein. Such an implementation can be fixed on a tangible medium such as a diskette, CD-ROM, ROM, or computer readable medium such as a hard disk, or transmitted to a computer system via a modem or other interface device. May include a set of computer readable instructions.