CN1553166A - Microscopic multispectral bone marrow and peripheral blood cell automatic analyzer and method - Google Patents

Abstract

The invention discloses a microscopic multispectral bone marrow and peripheral blood cell automatic analyzer and method, which includes a light source, a microscope, a display, an area array CCD and a computer; it is characterized in that: a. There is a filter system, and its core component is The electric tuning filter, which is configured on the microscope with the electric tuning filter controller and the area array CCD, is used to realize the rapid collection of microspectral images, and the computer controls the electric tuning filter controller to adjust the electric Tuning the voltage on the filter to change the center frequency of its passband; b. There is a CCD control and data acquisition system, which is used to convert the image signal detected by the area array CCD into a digital image through the image acquisition card and store it in the In the computer, the computer is used for image processing and analysis; c. There is a three-dimensional mobile control system for automatic focusing and scanning under the control of the computer. The automatic stage is connected to the computer through the three-dimensional mobile control system, and the computer passes through the area array The image collected by the CCD is used to calculate focus, control the automatic stage to move up and down for automatic focus, and move horizontally to realize smear scanning; d. There is also a computer, which is used to control the three-dimensional mobile control system and perform image synthesis on the collected spectral images , Image correction, automatic segmentation, automatic classification and recognition. The automatic analyzer and method of the present invention have high automation degree, accurate analysis result and high reliability, fully utilize microscopic spectrum to carry out spectral identification and classification of blood cells, and provide new spectral basis for diagnosis of blood diseases.

Description

Microscopic multispectral bone marrow and peripheral blood cell automatic analyzer and method

technical field

The invention relates to a microscopic multispectral automatic analyzer and method for bone marrow and peripheral blood cells, belonging to the technical field of biomedical detection and analysis instruments and methods, in particular to a rapid detection analyzer and method for human blood cells.

Background technique

Blood disease is one of the major diseases that seriously threaten human health. Different blood diseases have different qualitative (including shape) and quantitative changes in blood cells of patients. Wright's staining of conventional bone marrow (blood) cells is artificially performed under a light microscope. Recognition and classification are time-consuming (about 2 hours/case) and highly subjective (people with different levels of knowledge, or the same person observes and classifies differently at different times). It is not conducive to accurate diagnosis and treatment of patients, so it is an important topic in the medical field to replace manual classification with intelligent blood cell analyzers and methods.

At present, the methods used in the diagnosis of blood diseases are mainly divided into two types: one is the automatic hematology device designed and manufactured according to the principles of resistance change, laser scattering, radio frequency, histochemistry and flow cytometry, which has been widely used in the past ten years. Analyzers and Methods (AHA). This method is simple, fast and can measure multiple parameters. However, this type of instrument mainly classifies blood cells according to the volume and size of blood cells. It cannot identify the specific shape and structural characteristics of various cells, and there is less available information. Generally, only three classifications can be achieved. or five categories. To be precise, it should be called grouping, not classification. Many studies in the medical system have shown that this type of instrument classification is ineffective for abnormal cells of leukemia and various anemias, covering up many effective disease diagnosis information. Even higher-grade automatic hematology analyzers and methods can only be used as screening for normal specimens. If it is not coordinated with smear microscopy, it is easy to miss or misdiagnose. The other is to stain the smear of bone marrow cells, then manually manipulate the manual mechanical mover under the microscope, observe and find blood cells with the naked eye, and observe and judge the type and number of blood cells. This kind of inspection work is to check cells one by one It is extremely cumbersome and tiring, requires a lot of manpower, and requires specialized medical personnel to identify; due to the small differences in some cell changes and the lack of objective judgment scales, subjective errors cannot be avoided, and effective quantitative description results cannot be formed. Today, when collective censuses are gradually carried out, traditional inspection methods are no longer suitable for the needs, and people need instruments that can automatically complete these tasks, that is, a method for automatic identification of blood cells by computer.

The recent development of this type of instrument in the world is the first way of combining two or more signals to improve the quality of classification. For example, Coulter's VCS automatic white blood cell analyzer and method combine volume analysis, high-frequency conduction analysis and laser scattering. Three techniques are analyzed. Studies have shown that it does help in the differential diagnosis of some blood diseases, but it cannot be used for the classification of bone marrow blood cells. The development of the second way is to use an automatic image analyzer and method to extract the shape and texture parameters of blood cells and nuclei by computer, and use artificial intelligence to automatically classify cells by imitating human eyes. Countries such as the United States and Israel have launched a variety of commercialized equipment, with a price of about 400,000 US dollars. Since the 1980s, the country has been engaged in the research of cell diagnosis, established a cancer cell automatic identification research collaboration group, and achieved some preliminary results. In the early 1990s, Luo Limin of Southeast University used a microcomputer and a monochrome area array CCD to collect white blood cell images, and systematically studied the problem of white blood cell automatic identification. However, due to the use of monochrome grayscale images, the cell segmentation feature description method is difficult. , resulting in unsatisfactory recognition accuracy. Later, with the development of computer technology and color area array CCD imaging technology, Wang Dinghui, Zhang Yong and others applied color image processing and analysis technology to the automatic classification of blood cells, making the system recognize The rate has been improved, but these research results have not been practically applied.

The first method uses the physicochemical composition information and cell size information of cells, but it cannot use the rich morphological information of cells like hematologists; the second method uses the shape and texture information of cells, but cannot use the components of the cells themselves change information. And the first method is only suitable for peripheral blood examination, not for bone marrow blood cells. Therefore, it is currently impossible to replace artificial bone marrow (blood) cell classification, and it can only be used as a screening method. Using this type of instrument, the retest rate of patient smears is between 30-50%.

The emergence of spectral imaging technology fundamentally broke this situation for the first time. It allows us to obtain the spectrogram of bone marrow (blood) cells while obtaining the morphology of bone marrow (blood) cells, so that we can analyze the shape and composition information of bone marrow (blood) cells at the same time. Spectral imaging technology was first applied to remote sensing analysis (also known as multispectral imaging or hyperspectral imaging), which allows us to identify forests, wheat fields, grasslands, and even mineral deposits under the surface based on spectral information while obtaining ground photos. Devices for microspectral imaging have gradually matured since 1998. Many foreign researchers are gradually improving the performance of the device and expanding the application field.

Multispectral imaging technology in biomedicine can be used to increase the information obtained and enhance the computer's ability to identify cells. In principle, when the nature of normal cells changes, the content of some substances in the nucleus, such as deoxyribonucleic acid (DNA), also changes, resulting in changes in the shape, size and ratio of nucleoplasm of the nucleus, and The type of change varies according to the type of cell and the degree of malignancy; therefore, the analysis of the grayscale and morphology of general cell images can only obtain 60%-70% of the identification information of cells at most; it is far less effective than multispectral images in observing cells. At the same time, through the analysis of imaging spectrum, the information of the composition change in the cell is obtained (the wavelength of the absorbed light is directly related to the material composition of the absorbed light), and the quantitative analysis information of the chemical composition of these cells makes the original computer image analysis method have Very broad application prospects. Some domestic scholars have used fiber optic spectrometers to study the spectrum of human blood leukocytes on a microscope. They have noticed that lymphocytes and granulocytes in patients have significantly different spectral characteristics compared with normal people, and the same kind of self-cells in different types of patients The spectrum is also not the same. It can be seen that the spectral identification and classification of white blood cells can be carried out by using microspectroscopy, which can provide a new spectral basis for the diagnosis of leukemia.

Contents of the invention

The purpose of the present invention is to combine multi-spectral imaging technology with microscopic technology, use a new type of electronically tuned liquid crystal filter (LCTF) to replace the traditional filter to achieve rapid wavelength scanning, and use a high-sensitivity cooling area array CCD to monitor bone marrow. (Blood) Cellular Wright's stained smear images were digitized. While obtaining the morphology of bone marrow (blood) cells, it also obtains the spectrum of bone marrow (blood) cells. Using spectral imaging technology to systematically study the characteristics of different blood cells in normal human bone marrow (blood) smears (the commonality of the same type of cells, the differences of different cells, and the influence and elimination methods of external factors such as sample preparation and light source on the spectrum), The changes in the spectral characteristics of cells after the lesion occurs, the morphological parameters of blood cells, etc., lay a solid foundation for the early diagnosis and automatic diagnosis of blood diseases. Through imaging spectral analysis, information on compositional changes in cells can be obtained, thereby making the analysis and identification of non-diseased cells and diseased cells more accurate and rapid. Compared with the current international blood automatic image analyzers and methods that only use cell morphology, the degree of automation of the instrument and the accuracy and reliability of the analysis results have been greatly improved. Compared with manual inspection, in addition to reducing complex manual labor, the detection accuracy will be greatly improved, and more patients will be able to detect diseases at an early stage and achieve the purpose of cure. The social benefits are huge. Compared with the international blood automatic image analyzers and methods that only use cell morphology, the detection accuracy is significantly improved.

The technical purpose of a kind of microscopic multi-spectral bone marrow and peripheral blood cell automatic analyzer and method involved in the present invention is achieved in that it comprises light source, microscope, display and, area array CCD and computer; It is characterized in that; a, has A filter system, the core component of which is the electric tuning filter, which is configured on the microscope with the electric tuning filter controller and the area array CCD, is used to realize the rapid collection of microspectral images, and the computer controls the electric tuning The filter controller adjusts the voltage applied to the electric tuning filter to change the center frequency of its passband; b. There is a CCD control and data acquisition system, which is used to realize the image signal detected by the area array CCD through the image The acquisition card is converted into a digital image and stored in the computer for image processing and analysis by the computer; c. There is a three-dimensional movement control system for automatic focusing and automatic scanning under the control of the computer. The automatic stage is controlled by three-dimensional movement The system is connected with a computer, which is scanned by the computer; d, there is also a computer, which is used to control the three-dimensional mobile control system and perform image synthesis, image correction, automatic segmentation, and automatic classification and recognition of the collected spectral images. Calculate the focus of the image collected by the area array CCD, control the automatic stage to move up and down for automatic focus, and move horizontally to realize the smear

Technical effect of the present invention: Compared with the prior art, it has outstanding features and remarkable progress. Among them, compared with the current international blood automatic image analyzers and methods that only use cell morphology, the degree of automation, accuracy of analysis results, and reliability have been significantly improved. Microscopic spectroscopy is fully used to identify and classify blood cells. The diagnosis of blood diseases provides a new spectroscopic basis; compared with manual inspection, in addition to reducing complex manual labor, the detection accuracy is greatly improved. The invention enables more patients to discover diseases at an early stage and achieve the purpose of curing, and has huge social benefits.

Description of drawings

Figure 1 is a schematic diagram of the structure and method of the entire instrument, in which: 1-microscope 2-light source, 3-automatic stage, 4-three-dimensional movement control system 5-electrically tuned filter controller, 6-printer and other peripherals , 7-computer, 8-data backup and recovery system, 9-monitor, 10-image monitor, 11-CCD control and data acquisition system, 12-area array CCD, 13, electric tuning filter.

Figure 2 is a flow chart of the system work

Figure 3 is a flow chart of multispectral image acquisition

Figure 4 is a typical bone marrow image,

Figure 5 is an average spectral curve of the background, mature red blood cells, cytoplasm and nucleus within the rectangular area in Figure 4

Figure 6 is a flowchart of the automatic segmentation algorithm

Detailed ways

The present invention will be further described below in conjunction with accompanying drawing:

In Figure 1 is an embodiment of the automated analyzer and method of the present invention. It has a light source 2, microscope 1, display 9 and area array CCD12 and computer 7; Also have: a, have a filter system, its key component is the electric tuning filter 13, and it and electric tuning filter controller 5 and the area array CCD 12 are configured on the microscope 1 to realize the rapid collection of microspectral images, and the computer 7 adjusts the voltage applied to the electric tuning filter 13 by controlling the electric tuning filter controller 5 to change the The central frequency of its passband; b, there is a CCD control and data acquisition system 11, which is used to realize that the image signal detected by the area array CCD12 is converted into a digital image by the image acquisition card and stored in the computer 7, for the computer to carry out the image Processing and analysis; c, there is a three-dimensional mobile control system 4, which is used to realize automatic focusing and automatic scanning under computer control. The automatic stage 3 is connected with the computer 7 through the three-dimensional mobile control system 4, and the computer 7 passes through the area array CCD12 The collected image calculates the focus, controls the automatic stage 3 to move up and down for automatic focus and horizontal movement to realize smear scanning; d, there is also a computer 7, which is used to control the three-dimensional mobile control system (4) and to collect the spectrum Image synthesis, image correction, automatic segmentation, automatic classification and recognition; further speaking: one end of the electric tuning filter 13 is installed on the microscope 1 through a standard C interface, and the other end is connected to the area array CCD12 to stain the smear cells The image is imaged on the area array CCD12 after passing through the electrical tuning filter 13; the electrical tuning device 13 is installed in front of the area array CCD12, and is installed on the microscope 1 through a lens barrel for compensating the optical path length; the target object image is enlarged through the microscope After that, it is imaged on the area array CCD12 through the electric tuning filter 13; the image signal detected by the area array CCD12 is converted into a digital image by the control and data acquisition system 11 of the area array CCD and stored in the computer 7 for the computer 7 for image processing and analysis; the electrical tuning filter 13 is used as a filter device, and its bandwidth is 5-30nm, which can be a liquid crystal tuning filter or other types of electrical tuning filters, which are controlled by a computer The voltage on the electric tuning filter 13 changes the wavelength of light passing through the electric tuning filter 13; the smear is irradiated by continuous light, and the optical signal (transmitted light) generated by cells or other tiny objects is passed through the electric tuning After the optical filter 13, it is imaged on the area array CCD12; under the control of the computer 7, under each field of view, an image can be collected in each of several selected wavebands, and then the collected images are collected by the computer 7. The spectrum images obtained are image synthesis, image correction, automatic segmentation, automatic classification and recognition, etc.; the light source 2 adopts Kohler illumination to ensure the uniformity of the field of view; different area array CCD12 exposure times and different The light source intensity, under the same magnification, adopts the same area array CCD12 exposure time and the same light source intensity to ensure the consistency of the acquisition conditions.

Figure 2 is a flow chart of the system work.

Turn on the light source 2 and the ETF controller 5 of the microscope 1, and start working after the light source 2 is stable for about half an hour. First, collect the background spectrum. For the method, see the following multispectral image collection method. After the bone marrow (blood) smear is placed on the automatic stage 3, manually focus and adjust to a suitable position; start the automatic scanning program of the computer 7; The transmission wavelength of the device 13 is positioned at the middle wavelength of the wavelength to be collected, an image is collected at this wavelength, the collected image is analyzed, and the image processing method is used to judge whether the slide is in focus, if not, then in the computer Under the control of 7, the height of the automatic stage 3 is continuously adjusted until it is focused; when a field of view is focused, a multispectral image can be collected, and after the multispectral image of a field of view is collected, the computer 7 automatically performs image segmentation. Find the cytoplasm and nucleus; after the image is segmented, you can measure the cell characteristics, extract some morphological features and spectral features, and use these features to classify the cells. After one field of view is segmented and analyzed, continue to collect and analyze the next one. Field of view: After the number of fields of view to be collected reaches a certain requirement, the image collection is stopped, and all previously collected cells are analyzed comprehensively, and the disease classification is obtained by using the neural network classifier according to the disease database established under the guidance of the doctor.

Figure 3 is a flow chart of multispectral image acquisition. Since the spectral image is composed of a series of monochromatic images, it is necessary to collect an image at each wavelength selected by us in each field of view; turn on the light source 2 and control the electric tuning filter Device 5, after the light source is stable for about half an hour, put the bone marrow (blood) smear on the automatic stage 3, adjust the light source 2 to a suitable position, and put it in the same position every time for the same magnification , so that the illumination intensity is the same; adjust the light source of the microscope 1 to Kohler illumination to ensure that the brightness of the field of view is uniform; the method of image acquisition is to first position the transmission wavelength of the electric tuning filter 13 to the position we want in each field of view. The middle wavelength of the collected wavelength is automatically focused by using an image processing algorithm under the control of the computer 7 to make the field of view the clearest, and then the electronically tuned filter 5 is switched to a selected waveband through the computer 7 to collect a picture Image, perform grayscale correction on the collected image, then switch to the next selected band and then collect an image and then perform grayscale correction, and so on, until all the selected bands are collected, and then perform color synthesis And data storage, use computer 7 image processing technology to automatically find the nucleus and cytoplasm of blood cells, measure the characteristic parameters of the cells, use the neural network method to obtain the category of the cells, and store them in the database together with the characteristic parameters. Then collect the next field of view until the number of scanning fields of view meets the requirements; when all the regions of interest in a smear are collected, the measured data can be automatically analyzed to give whether the lesion information, percentage, benign malignancy and other information;

Since the incandescent light source presents lower blue light in the spectrum and higher near-infrared (NIR) in the spectral output; the quantum efficiency (QE) curve of the area array CCD is usually less in the blue region and the highest in the red light. The throughput properties (throughput properties) of the electrically tuned filter 13 are also shown as certain spectral curves. For example, LCTF filters have less transmittance in blue light and higher transmittance in NIR. Thus, we run into trouble with the combined effect of low performance on Blu-ray. This problem can be alleviated by choosing an area CCD with a higher QE in the blue portion, such as a thin backilluminated area CCD. But in this case, a high-performance camera is required, which will lead to an increase in price. The present invention solves the problem of the incandescent light source well through the method of software correction, and effectively eliminates the influence of the excitation light source 2, the microscope 1, the electric tuning filter 13, and the detector on the imaging spectrum, and ensures the consistency of imaging detection;

The method of grayscale correction is: focus on the smear to make the cells clear best; move the smear to a blank area, under an empty field of view, slightly defocus, control the electric tuning filter 5 and the area array CCD12 through the computer 7, Use the same integration time to collect an image in each selected band, and then calculate the next appropriate integration time for each band based on these images. A long integration time can be given for the band with a relatively low gray value. Give a short integration time to the band with high gray value to compensate for the balance of the whole band range. According to the calculated integration time, another image is collected under each selected band to obtain an enhanced background spectrum image. In the future, for the image of each band, it will be collected with the same integration time as the enhanced background spectrum image, and the band image corresponding to the enhanced background spectrum image will be divided and then multiplied by a scaling factor to adjust the gray value It can be improved to ensure that the background area has consistent gray values in different bands after processing, which is convenient for image processing and analysis. Since the correction of the background is processed in units of pixels, it can not only effectively remove the influence of the different sensitivity of the imaging system on different bands, but also compensate and improve the blue-ray bands, effectively eliminating some slight field of view inconsistencies. Uniformity and photosensitivity difference of each pixel of area array CCD12.

Figure 4 is a typical bone marrow image, and Figure 5 is the average spectral curve of the background, mature red blood cells, cytoplasm and nucleus within the rectangular area in Figure 4. The abscissa in Figure 5 is the wavelength, and the ordinate is the intensity of the transmitted light. An image is collected every 10 nm, and each point represents the average gray level in the rectangular area of the monochromatic image in each waveband in Figure 4 .

It can be seen from Figure 5 that different substances have different spectral transmittances. We can use these differences to automatically segment them. Because the nuclei, cytoplasm and background regions have different spectral characteristics, using their spectral differences, the image segmentation method based on pixels is used to segment the image, and a good segmentation effect is obtained.

The main segmentation method used is the spectral ratio operation. Because the color difference of different types of bone marrow cells is small, and because the staining depth is difficult to be completely consistent, it is difficult to accurately distinguish the nucleus, cytoplasm and background by simply using the difference in a certain band or the absolute gray value of the image. end. Therefore, we use the spectral ratio method to increase the difference between them, and the ratio is a relative value, which greatly reduces the influence of the dyeing depth and the sensitivity of the photoreceptor unit on the image segmentation results. The specific method is to select two bands, and for each image pixel, use the gray levels of the two bands to perform a division operation, and then multiply the obtained result by a fixed magnification factor to generate a new monochrome image. This monochromatic image is utilized to carry out automatic threshold value segmentation, what the present invention uses is the maximum variance threshold value method, has obtained good effect. The present invention has found several groups of wavelengths, which can separate the cytoplasm and nucleus respectively;

Due to the massive data of spectral images, band selection is a necessary link in processing spectral images. On the one hand, band selection can effectively reduce the amount of calculation, and on the other hand, it can also remove redundant information. The search algorithm of feature selection can be divided into optimal search and suboptimal search. The optimal algorithm includes exhaustive method and branch and bound algorithm; the suboptimal algorithm includes simulated annealing algorithm, Tabu algorithm and genetic algorithm, etc. From the perspective of obtaining the optimal result and simple implementation, the present invention adopts an exhaustive method. The method for finding the wavelength of the present invention is to use the exhaustive method to traverse all possible wavelength groups, corresponding to different wavelength groups, respectively find the ratio of the corresponding gray levels of these two kinds of substances to form a feature space, and then find out the ratio of the two types of substances. Separability criteria (such as intra-class and inter-class distance, Bhattacharyya distance and divergence), compare the error rate of classification, and a group of wavelengths with the smallest error rate is the desired one.

Fig. 6 is a flow chart of automatic segmentation (detection) of bone marrow (blood) smear cell image

The present invention studies the automatic segmentation (detection) algorithm of bone marrow (blood) smear cell image, and finds out a series of automatic fast segmentation algorithms. Image segmentation mainly uses the spectral ratio method. Since the nuclei, cytoplasm, mature red blood cells, and background have different spectral properties, their spectral properties can be used for fully automatic image segmentation. The specific segmentation process is: select several bands, use the band information to find the blank area; select several bands, use the band information to find the mature red blood cell area; delete the red blood cell area and the blank background area to get the nucleated cell area . Then perform binary morphological operations such as erosion and expansion to remove some small particles, and obtain region S1. Select two bands and perform spectral ratio operation. For the obtained ratio image, use the maximum variance threshold method to find a global threshold in the area marked by S1, which can separate the cytoplasm and nucleus to obtain the nucleus area S2. The effect is rough, but fast. Delete the small area in S2, fill the deleted part with cytoplasmic markers, and then delete the area that is not connected to the nucleus to get the area S3, then S3 is a cell one by one. Since the previous segmentation is a rough segmentation, several bands are selected, and each cell is used as a segmentation unit, and the band information is used to accurately segment it, so that the nucleus and cytoplasm of each cell can be accurately segmented. After the cells are segmented, they can be classified and identified.

Use the visual method to see the effect of image segmentation. Accuracy of nucleus segmentation for monocytes, lymphocytes, neutrophil rods, neutrophils, and eosinophils, for peripheral blood smears, for segmentation of approximately 8,000 cells stained by Wright At above 98.7%, the accuracy of basophil nuclei segmentation was 90.2%. The accuracy of the cytoplasmic segmentation is slightly worse, mainly because the spectral difference of some cytoplasmic cells is very small compared with other anucleated cells, but the segmentation accuracy rate has reached more than 90%, which is much higher than the existing literature. reported value. For the bone marrow cell smear stained by Wright, about 9000 cells were divided, common metamyelocytes, myelocytes, promyelocytes, neutral segmented nuclei, neutral rod-shaped nuclei, neutrophil myelocytes, The accuracy of nucleus segmentation in neutrophil metamyelocytes and lymphocytes is above 98.5%. For relatively few eosinophils and basophils, the accuracy of nucleus segmentation is above 95.1%. For the segmentation of cytoplasm, except for late Except for young red blood cells, the accuracy rate is above 94.3%. Since late young red blood cells become mature red blood cells when they grow up, the spectral characteristics of their cytoplasm are very close to those of mature red blood cells, so it is easy to misclassify. Judging from the current situation, the accuracy rate has reached 85.7%.

The ultimate goal of image segmentation is to perform classification and recognition. In order to be able to identify effectively, it is necessary to establish a feature library first. The method of establishing the feature library is to first collect 300 case slides of normal people and various stages of leukemia, collect their microscopic multi-spectral images, and perform manual classification under the guidance of experts, and compile programs for manual and accurate segmentation and measurement. Perform feature extraction. Due to the large number of extracted parameters and the multi-spectral image has many bands, its spectral features are several times more than ordinary grayscale images. Too many features will not only slow down the analysis speed, but also may cause more errors in the analysis results. Therefore, it is necessary to perform feature selection and feature screening to reduce the dimensionality of features. For the training samples, the present invention extracts a total of 438 features including morphological parameters, optical density parameters, texture parameters, and spectral ratio parameters. The present invention uses a genetic algorithm to select features and analyze the spectra and morphological characteristics of different types of leukemia and different types of cells , analyze their specificity and correlation, select the characteristic parameters of cell classification, and achieved good results. After the genetic algorithm, 132 optimal features are finally obtained. After the feature selection is completed, we use the selected features to establish the original feature database.

After the original feature library is established, build a 3-layer neural network, use the neural network to train these samples, and store the trained weight data in the database after the network training is completed.

With the original feature library and the results of neural network training, we can identify unknown samples. The invention adopts the neural network to automatically identify and classify the cells to be tested, and obtains good results, and the speed is extremely fast. From the perspective of recognition effect, for peripheral blood smears, about 8000 cells stained by Wright were classified, the correct rate of monocytes was 97.2%, the correct rate of lymphocytes was 98.7%, and the correct rate of neutrophils was 95.3%. %, the correct rate of neutrophils was 96.6%, the correct rate of eosinophils was 97.6%, and the correct rate of basophils was 92.1%. For the bone marrow smear stained by Wright, about 9000 cells were classified, and the accuracy rate was: 94.6% for common metamyelocytes, 93.2% for mesoerythroblasts, 90.7% for promyelocytes, 95.5% for neutral segmented nuclei, and 95.5% for mesenchymal 87.4% rod-shaped nuclei, 89.4% neutrophil myelocytes, 88.3% neutrophil metamyelocytes, 92.0% lymphocytes, less eosinophils and basophils and other cells are above 80.3%.

Claims (10)

1, a kind of micro-multispectral marrow and peripheral blood cells automatic analyzer and method, it comprises light source (2), microscope (1), display (9) and area array CCD (12) and computing machine (7); It is characterized in that: a, a filter system is arranged, its core component is electric tuning light filter (13), it and electric tuning control device of the light filter (5) and area array CCD (12) are configured on the microscope (1), be used to realize the quick collection of microspectrum image, computing machine (7) is then regulated the voltage that is added on the electric tuning light filter (13) by control electric tuning control device of the light filter (5), changes the centre frequency of its passband; B, CCD control and data acquisition system (DAS) (11) are arranged, be used for realizing that converting the detected picture signal of area array CCD (12) to digital picture by image pick-up card is stored in computing machine (7), carries out Flame Image Process and analysis for computing machine; C, a three-dimensional mobile control system (4) is arranged, be used under computer control, realizing automatic focus and scanning, automatic carrier (3) is connected with computing machine (7) by three-dimensional mobile control system (4), and the image calculation of being gathered by area array CCD (12) by computing machine (7) focuses on, control automatic carrier (3) moves up and down and carries out automatic focus and move horizontally realizing smear scanning; D, also have computing machine (7), be used for to the control of three-dimensional mobile control system (4) and to the spectrum picture that collects carry out that image is synthetic, image rectification, cut apart automatically, Classification and Identification automatically.

2, by described analyser of claim 1 and method, it is characterized in that, described electric tuning light filter (13) one ends are installed on the microscope (1) by the standard C interface, the other end is connected with area array CCD (12), is imaged on the area array CCD (12) behind the stain smear cell picture process electric tuning light filter (13).

3, by described analyser of claim 1 and method, it is characterized in that the bandwidth of electric tuning light filter (13) is 5-30nm, can be the liquid crystal tunable filter.

4, by described analyser of claim 1 and method, it is characterized in that, light source (2) adopts kohler's illumination, be used to guarantee the homogeneity of visual field, it is to adopt different area array CCD (12) time shutter and the different intensities of light source under different enlargement factors, under identical enlargement factor, adopt identical area array CCD (12) time shutter and the identical intensity of light source, guarantee the consistance of acquisition condition.

5, by described analyser of claim 1 and method, it is characterized in that its working routine comprises the steps: a, automatic focus; The collection of b, spectrum picture; C, based on the cell segmentation of spectrum picture; D, characteristics of image are measured; E, cytological classification; F, autoscan; G, classification of diseases.

By described analyser of claim 5 and method, it is characterized in that 6, the collection of described spectrum picture and background correction comprise the steps: a, gather a width of cloth background image at each selected wave band earlier; B, calculate integral time of each spectral band according to the background image that is collected; C, again according to integral time of calculating gathering the background image that a width of cloth strengthens at each selected wave band; D, the image that collects later on use identical with the background image that strengthens integral time at each wave band, the image of each wave band all looks like to carry out gray correction according to the back of the body base map that strengthens, to eliminate or to reduce the influence of light source, optical element, area array CCD (12) photosensitive unit, luminous sensitivity and spectral response.

7, by described analyser of claim 5 and method, it is characterized in that, described cell segmentation detects, adopt a kind of full automatic spectrum dividing method, it is to adopt the method for spectral ratio to obtain some gray level images, utilizes automatic threshold dividing method and two-value morphological method to carry out by thick to smart cutting apart again.

8, by described analyser of claim 7 and method, it is characterized in that, described full automatic spectrum is cut apart flow process and is: choose several wave bands, utilize these band class information to find out white space, select some wave bands again, utilize band class information to find out ripe red blood cell zone, just can obtain the karyocyte zone behind deletion red blood cell zone and the blank back bottom area territory; Carry out two-value morphological operations such as corroding expansion again and remove some little particles, obtain region S 1; Select two wave bands, carry out the spectral ratio operation, the ratio image that obtains, utilize the maximum variance threshold method to find the threshold value of an overall situation in the zone of S1 institute mark, can separate endochylema and karyon, obtain karyon region S 2, the effect of cutting apart like this is more rough, but speed is very fast; Deletion S2 medium and small surface area, the part of deletion is filled with the endochylema mark, and then the zone that deletion does not link to each other with karyon obtains region S 3, and S3 is exactly a cell one by one so; Because be rough cutting apart cutting apart of front, therefore, selects several wave bands again, as cutting apart unit, utilizes band class information that it is accurately cut apart with each cell, can be partitioned into the karyon and the endochylema of each cell accurately.After cell segmentation is intact, just can carry out Classification and Identification.

9, by described analyser of claim 6 and method, it is characterized in that, the band selection of described micro-multispectral analysis, it is to utilize the separability criterion, selects one group of wave band can distinguishing two kinds of materials with the method for exhaustion from the wave band of setting.

10, by described analyser of claim 5 and method, it is characterized in that, described classification of diseases: a, set up the initial characteristic data storehouse: it is according to collecting normal population earlier, the case slide of each phase type of leukemia, gather micro-multispectral image, carry out the manual sort, program and manually accurately cut apart and measure, carry out feature extraction, screening, extract the morphology parameter, optical density parameter and parametric texture and spectral ratio parameter attribute, analyze dissimilar leukaemia, the spectrum of dissimilar cells and morphological character, analyze their specificity and correlativity, select the characteristic parameter of cytological classification, obtain optimal characteristics at last, the feature after will selecting is again set up the initial characteristic data storehouse; B, 3 layers of neural network of structure utilize neural network to these sample training, after network training finishes the weights data of training are deposited in database; C, according to the result of primitive character storehouse and neural metwork training, unknown sample is discerned, with neural network cell to be measured is carried out automatic recognition classification.

Images