CN110874548A - Lung cancer cell and normal cell recognition method based on combination of Raman spectrum and SVM - Google Patents

Abstract

A method for identifying lung cancer cells and normal cells based on Raman spectroscopy combined with SVM, comprising the following steps: (1) culturing two kinds of cells into cell clusters; (2) culturing cells into cell clusters and directly placing them on a glass slide, and confocal laser light Raman spectrum is measured by Raman spectrometer; ③ After the Raman spectrum is obtained, some spectra are processed to remove cosmic rays in combination with Project FOUR 4.1 software; ④ Feature extraction is performed on the spectrum preprocessed in step ③, and the extracted features are characteristic peaks ⑤ The features extracted in ④ are combined with the machine learning method SVM to classify and identify the spectral data; ⑥ The remaining samples are selected for testing to obtain the accuracy of cell identification. The present invention utilizes the laser confocal Raman spectrometer to obtain the characteristics of the spectrum and combines the latter feature extraction to eliminate the low recognition rate caused by the error generated in the experiment or the sample culture process.

Description

Identification of lung cancer cells and normal cells based on Raman spectroscopy combined with SVM method

technical field

The invention belongs to the field of computer vision and pattern recognition, in particular to a method for identifying lung cancer cells and normal cells based on Raman spectroscopy combined with SVM.

Background technique

Recognizing and locating objects is an important research content in the field of computer vision and pattern recognition. As a branch of object detection, the classification of cancer cells is a special case of object detection. Cells are a special class of substances, which are not only universal but also diverse in particularity. Therefore, biometric detection has broad scientific research value and application prospects, and has very important research significance in medicine.

Currently, fluorescent labeling methods are mainly used to identify cell types due to their specificity. Fluorescent labeling is based on the specific binding of antigens and antibodies. This method is prone to damage the original physiological activity of cells and is prone to false positive results of specific binding proteins. In addition, there are many drawbacks in clinical application due to the complex, high cost and low efficiency of sample processing.

Raman spectroscopy is a molecular inelastic scattering fingerprint spectroscopy technology. It is a non-contact technology that can specifically identify cancer cells at the physical level. It can not only maintain integrity and cell activity, but also effectively solve biological problems. Complexity and efficiency of sample pretreatment efficiency. Raman spectroscopy is highly specific and reflects the changes in the biochemical composition of living cells in aqueous solution without any labeling and fixation. Therefore, Raman spectroscopy has been widely used in clinical diagnosis, toxicology detection and tissue engineering. In addition, compared with other techniques, Raman spectroscopy has the characteristics of rapidity, simplicity, good repeatability, and more importantly, it is a non-destructive qualitative and quantitative analysis method. It does not require sample preparation and the sample can be measured directly with a fiber optic probe or through glass or silica and fiber optic measurements.

Laser confocal Raman spectrometer is an effective spectroscopic analysis method for analyzing the composition and structure of substances. The principle is that incident laser light causes molecules (or lattices) to vibrate. Losing (or gaining) part of the energy, changing the scattered light frequency to scattered light analysis, ie Raman spectroscopy, can explore the composition, structure and relative content of known molecules. Micro Raman technology can focus the spot of the excitation light to the micron level, and then accurately analyze the micro area of the sample. The exact part where the laser acts on the sample can be clearly displayed by a CCD identification device and a TV monitor. come out. Confocal Raman spectroscopy can select any part of any sample of interest for analysis, and the entire analysis and identification process is very intuitive, easy to observe and control.

SUMMARY OF THE INVENTION

The purpose of the present invention is to provide a method for identifying lung cancer cells and normal cells based on Raman spectroscopy combined with SVM. The method utilizes a laser confocal Raman spectrometer to obtain spectral features and combines the following feature extraction to eliminate experimental or The low recognition rate is caused by errors in the sample culture process.

In order to achieve the above object, the scheme of the present invention is: a method for identifying lung cancer cells and normal cells based on Raman spectroscopy combined with SVM, which is characterized in that: comprising the following steps:

①Cultivate two kinds of cells into cell clusters;

② After culturing the cells into cell clusters, place them directly on a glass slide, and measure the Raman spectrum with a laser confocal Raman spectrometer;

③ After obtaining the Raman spectrum, combine with the Project FOUR 4.1 software to remove cosmic rays from some spectra;

④ Perform feature extraction on the preprocessed spectrum in step ③, and the extracted feature is the position of the characteristic peak and the intensity ratio of the characteristic peak;

⑤ The features extracted in ④ are combined with the machine learning method SVM to classify and identify the spectral data;

⑥ Select the remaining samples for testing to obtain the accuracy of cell identification.

The adenocarcinoma cell line A549 and the pleural mesothelial cell line Met-5A were used for the above two cells.

The SVM classifier used by the above machine learning method SVM is LIBSVM.

The above laser confocal Raman spectrometer is a WITec spectrometer produced in Germany.

The present invention has the following advantages and positive effects:

1. The present invention uses a laser confocal Raman spectrometer with Raman microscopy as the background as an experimental device, and can obtain real-time Raman spectra to determine biological substances through several consecutive exposures, multiple sampling and appropriate sampling time. Whether the molecular components of the test show characteristic peaks, and then obtain a graded Raman spectrum;

2. The laser confocal Raman spectrometer used in the present invention is a WITec spectrometer produced in Germany. At the same time, by using the Project FOUR 4.1 software matched with the spectrometer, the spectrum can be preprocessed to remove the interfering cosmic rays, and the Raman spectrum before feature extraction can be obtained; and by using the Project FOUR 4.1 software, a large amount of processing can be performed in a short time. Raman spectrum obtained by experiment;

3. The SVM classifier used in the present invention after feature extraction is LIBSVM, which is a simple, easy-to-use, fast and effective SVM pattern recognition and regression software package developed and designed by Professor Lin Zhiren of National Taiwan University. There are relatively few parameter adjustments, and many default parameters are provided, which have obvious effects on classification. And it provides a better performance of the interactive test regardless of the large sample or the small sample.

4. The present invention classifies and identifies the preprocessed Raman spectrum by combining the traditional Raman spectrum feature extraction (the position of the feature peak and its intensity ratio) with the machine learning method SVM. In the preprocessing process, since the characteristic peaks of the obtained Raman spectra have biological meanings and the experimentally obtained spectral lines are very clearly visible, only individual Raman spectra have cosmic rays, using Project FOUR 4.1 software To remove cosmic rays, smoothing is not required.

5. The data set used as the experimental sample in the present invention is established by using the WITec spectrometer, and the intensity ratio in the feature is obtained by normalizing one of the peaks;

Description of drawings

Fig. 1 is the experimental flow chart of the present invention.

Detailed ways

A method for identifying lung cancer cells and normal cells based on Raman spectroscopy combined with SVM, comprising the following steps:

1. Cultivate two kinds of cells to form cell clusters: two kinds of cells are: lung adenocarcinoma cell line A549 and pleural mesothelial cell line Met-5A.

The culture conditions are: lung adenocarcinoma cell line A549 using DMEM basal medium (containing 10% fetal bovine serum, 1% penicillin-streptomycin double antibody), pleural mesothelial cell line Met-5A using DMEM high glucose medium (containing 10 % fetal bovine serum) was cultured in a 37°C, 5% CO ² incubator, collected into a 15ml sterile centrifuge tube after culture, washed twice with phosphate buffered saline (PBS), centrifuged to pellet cells, centrifuged at 4500 rpm for 10 min, and discarded. After all the supernatant is removed, the cells will be collected for easy observation;

2. After discarding all the supernatant, the collected cells were evenly spread on a glass slide, and the Raman spectrum was measured with a laser confocal Raman spectrometer;

3. After obtaining the Raman spectrum, combine with Project FOUR 4.1 software to remove cosmic rays for some spectra.

4. Perform feature extraction on the preprocessed spectrum, the extracted feature is the position of the feature peak, and the intensity ratio in the feature is obtained by normalizing one of the peaks; the position of the feature peak includes 1080cm ^-1 and 1128cm ^-1 , 1258cm ^-1 , 1301cm ^-1 , 1342cm ^-1 , 1449cm ^-1 , 1578cm ^-1 , 1617cm ^-1 , 1659cm ^-1 and the intensity ratio of these nine position characteristic peaks, here I use the intensity value of the seventh peak For the normalized value, the eight eigenvalues are (taking the first sample as an example) 1.077, 1.047, 1.180, 1.263, 1.214, 1.468, 1.036, and 1.414.

5. The machine learning method SVM is used to train the selected features to obtain the SVM model, which is obtained by selecting a certain proportion of two cell samples for training, including 242 lung cancer cell samples and 231 normal cell samples. The SVM classifier adopts LIBSVM.

6. Select the remaining samples for testing to obtain the accuracy of cell identification.

The invention can effectively identify cancer cells and normal cells. The data set used as the experimental sample is the cell Raman spectrum data set constructed by using the WITec spectrometer. The method is simple in operation, and the size of the constructed SVM model can be determined according to the size of the corresponding data set. The requirements for machine hardware are not high, and it can be applied to pattern recognition in other fields.

It should be noted that, the above descriptions are only embodiments of the present invention, which are merely to explain the present invention, and are not intended to limit the scope of the patent of the present invention. Changes that belong to the technical concept of the present invention but are only obvious changes also fall within the protection scope of the present invention.

Claims (4)

1. A lung cancer cell and normal cell recognition method based on Raman spectrum combined with SVM is characterized in that: the method comprises the following steps:

① culturing the two cells into cell mass;

② culturing the cells into cell clusters, directly placing on a glass slide, and measuring with a laser confocal Raman spectrometer to obtain Raman spectrum;

③ obtaining Raman spectra, and removing cosmic rays from some spectra by combining Project FOUR 4.1 software;

④, extracting the characteristics of the spectrum pretreated in step ③, wherein the extracted characteristics are the position of a characteristic peak and the intensity ratio of the characteristic peak;

⑤, classifying and identifying the spectral data by combining the features extracted from ④ with a machine learning method SVM;

⑥ the remaining samples are selected for testing and the accuracy of cell identification is obtained.

2. The method for identifying lung cancer cells and normal cells based on the combination of Raman spectroscopy and SVM of claim 1, wherein: the two cells adopt an adenocarcinoma cell line A549 and a pleural mesothelial cell line Met-5A.

3. The method for identifying lung cancer cells and normal cells based on the combination of Raman spectroscopy and SVM of claim 1, wherein: the SVM classifier adopted by the machine learning method SVM is LIBSVM.

4. The method for identifying lung cancer cells and normal cells based on the combination of Raman spectroscopy and SVM of claim 1, wherein: the laser confocal raman spectrometer described above is a WITec spectrometer produced in germany.

Images