阅读说明：本技术 一种基于光谱特征参数的***前病变诊断方法 (Cervical cancer pre-lesion diagnosis method based on spectral characteristic parameters ) 是由 钱志余 刘文文 晋晓飞 李韪韬 李怡燃 刘洋洋 张欢 李俊俊 于 2019-10-24 设计创作，主要内容包括：本发明公开了一种基于光谱特征参数的宫颈癌前病变诊断方法,属于宫颈癌前病变筛查方法领域。本发明利用光谱采集系统获取宫颈组织漫反射光谱,从光谱数据中计算特征参数,并基于特征参数实现正常和病变组织的区分。对实验采集的宫颈组织光谱数据进行分析,首先进行筛选和预处理,在特征波段计算特征参数,后利用特征参数建立分类模型。本发明方法完全基于光谱特征模型的预测结果,不需要依赖于医生的主观判断,同时也不需要高水平的技术操作人员,为医疗资源匮乏的地区提供了更合适的筛查手段。(The invention discloses a cervical cancer precancerous lesion diagnosis method based on spectral characteristic parameters, and belongs to the field of cervical cancer precancerous lesion screening methods. The invention utilizes the spectrum acquisition system to obtain the diffuse reflection spectrum of the cervical tissue, calculates the characteristic parameters from the spectrum data, and realizes the distinction between normal tissue and lesion tissue based on the characteristic parameters. Analyzing the cervix tissue spectral data collected by the experiment, firstly screening and preprocessing, calculating characteristic parameters in characteristic wave bands, and then establishing a classification model by using the characteristic parameters. The method is completely based on the prediction result of the spectral feature model, does not need to rely on the subjective judgment of doctors, does not need high-level technical operators, and provides a more appropriate screening means for the areas with deficient medical resources.)

1. A cervical cancer pre-lesion diagnosis method based on spectral characteristic parameters is characterized by comprising the following steps:

(1) data acquisition: performing spectrum acquisition by using a spectrum acquisition system, and acquiring diffuse reflection spectrum data of cervical tissues of a patient;

(2) data screening and preprocessing: screening out unqualified spectral data, and smoothing and normalizing the qualified spectral data;

(3) feature extraction: extracting the characteristics of the spectrum, and selecting characteristic parameters for modeling;

(4) establishing a classification model: taking the characteristic parameters as input, establishing a classification model, and distinguishing a normal tissue spectrum from a pathological tissue spectrum;

(5) type diagnosis: the type of the tissue spectrum is judged based on the existing model, and a diagnosis result is output.

2. The cervical cancer pre-lesion diagnosis method based on spectral characteristic parameters of claim 1, wherein the spectral collection system in step (1) comprises a halogen light source, a fiber optic spectrometer and a dual fiber optic probe, wherein the dual fiber optic probe is respectively connected with the halogen light source and the fiber optic spectrometer.

3. The cervical cancer precancerous lesion diagnosis method based on spectral feature parameters of claim 1, wherein the collection wavelength range of the spectral collection system in the step (1) is 200-1100 nm.

4. The cervical cancer precancerous lesion diagnosis method based on spectral feature parameters of claim 1, wherein the spectral collection in step (1) comprises the following processes:

a) reserving a background spectrum of the detection environment;

b) the probe is slightly close to the cervical moving belt and slowly slides;

c) the collection time of the suspected lesion area is prolonged, the probe is moved after the probe slides for a circle on the moving belt, and the collection is finished.

5. The method for diagnosing cervical cancer lesion based on spectral characteristic parameters of claim 1, wherein the qualified spectral data in step (2) is divided into a training set and a verification set, the training set trains the prediction model, and the test set is used for verifying the accuracy of the model.

6. The cervical cancer precancerous lesion diagnostic method based on spectral characteristic parameters of claim 1, wherein the characteristic extraction method in the step (3) is as follows: and finding out characteristic wave bands by comparing spectral data of different tissues in an experiment, and searching for available characteristic parameters in the characteristic wave bands.

7. The cervical cancer precancerous lesion diagnostic method based on spectral characteristic parameters of claim 6, wherein the characteristic parameters of the step (3) are divided into three types of spectral peak-to-peak area and slope parameters and optical parameters.

8. The cervical cancer pre-lesion diagnostic method based on the spectral characteristic parameters of claim 1, wherein: and (4) adopting a neural network model for modeling in the step (3).

Technical Field

The invention relates to a cervical cancer precancerous lesion diagnosis method based on spectral characteristic parameters, and belongs to the field of cervical cancer precancerous lesion screening methods.

Background

In China, cervical cancer is the second most common malignancy among young and middle-aged women, seriously jeopardizing women's life health. The cervical cancer is the only cancer with definite etiology at present, and can be cured, and the development of the cervical cancer can be effectively blocked by early screening and prevention. Infection with high-risk Human Papillomavirus (HPV) lasts for more than two years and can develop into Cervical Intraepithelial Neoplasia (CIN) without treatment, high-grade cervical lesions can further develop into cervical cancer, and the time from infecting HPV to developing into cervical cancer generally lasts for 10-20 years. The cervical intratumoral lesion can be divided into three grades, namely CIN I, CIN II and CIN III, and the higher the grade is, the larger the diffusion area of heterotypic cells is.

The clinical cervical cancer screening method mainly comprises acetic acid (or iodine) staining observation, cytology examination, electronic colposcopy, tissue biopsy and the like. The above methods have problems of low sensitivity and specificity, long waiting period, pain in the examination process, and the like, and other methods depend on the technical level of operators and the subjective judgment of doctors, have unstable accuracy, and are not suitable for small hospitals with shortage of resources.

The existing instrument for screening early cervical cancer lesions is mainly an electronic colposcope, and the electronic colposcope can amplify images and observe focus details which cannot be seen by naked eyes, thereby being beneficial to judging early cancers. The diagnosis result of the electronic colposcope depends on the subjective judgment of a doctor, and the accuracy is determined by the doctor level.

During the process of lesion development, the optical parameters, absorption coefficient, reduced scattering coefficient, etc. of the cervical tissue will change, and the changes of these parameters will be reflected in the diffuse reflectance spectrum of the tissue. Research on the application of diffuse reflectance spectroscopy to the detection of cancer has been conducted in various fields such as oral cancer, skin cancer, and cervical cancer.

Disclosure of Invention

The invention provides a cervical cancer precancerous lesion diagnosis method based on spectral characteristic parameters, which has the advantages of being noninvasive, rapid, real-time and the like, and makes up for the defects of the existing clinical application technology. The pathological state of the tissue is diagnosed through the tissue spectrum characteristics of the patient, and the diagnosis result has objectivity and is not influenced by the technical level of operators.

The invention adopts the following technical scheme for solving the technical problems:

a cervical cancer pre-lesion diagnosis method based on spectral characteristic parameters comprises the following steps:

(1) data acquisition: performing spectrum acquisition by using a spectrum acquisition system, and acquiring diffuse reflection spectrum data of cervical tissues of a patient;

(2) data screening and preprocessing: screening out unqualified spectral data, and smoothing and normalizing the qualified spectral data;

(3) feature extraction: extracting the characteristics of the spectrum, and selecting characteristic parameters for modeling;

(4) establishing a classification model: taking the characteristic parameters as input, establishing a classification model, and distinguishing a normal tissue spectrum from a pathological tissue spectrum;

(5) type diagnosis: the type of the tissue spectrum is judged based on the existing model, and a diagnosis result is output.

The spectrum acquisition system in the step (1) comprises a halogen light source, an optical fiber spectrometer and a double optical fiber probe, wherein the double optical fiber probe is respectively connected with the halogen light source and the optical fiber spectrometer.

The collection wavelength range of the spectrum collection system in the step (1) is 200-1100 nm.

The spectrum acquisition in the step (1) comprises the following processes:

a) reserving a background spectrum of the detection environment;

b) the probe is slightly close to the cervical moving belt and slowly slides;

c) the collection time of the suspected lesion area is prolonged, the probe is moved after the probe slides for a circle on the moving belt, and the collection is finished.

And (3) dividing the qualified spectral data in the step (2) into a training set and a verification set, training the prediction model in the training set, and verifying the accuracy of the model in the test set.

The characteristic extraction method in the step (3) is as follows: and finding out characteristic wave bands by comparing spectral data of different tissues in an experiment, and searching for available characteristic parameters in the characteristic wave bands.

And (3) dividing the characteristic parameters into three types of spectral peak area, slope parameter and optical parameter.

And (4) adopting a neural network model for modeling in the step (3).

The invention has the following beneficial effects:

the method can be used as a means for screening the cervical precancerous lesions and can also be used as an auxiliary tool. The method is completely based on the prediction result of the spectral feature model, does not need to rely on the subjective judgment of doctors, does not need high-level technical operators, and provides a more appropriate screening means for the areas with deficient medical resources. The screening process of the method does not need biopsy, dyeing and the like, optimizes the examination experience of the patient, can improve the acceptance of the patient on cervical cancer screening, and is favorable for improving the cervical cancer screening rate. The cervical tissue classification model established based on clinical trial data has the advantages that the prediction sensitivity of the existing data is higher than 90% and the specificity is higher than 85%, and compared with other methods applied clinically, the cervical tissue classification model has certain advantages.

Drawings

FIG. 1 is a flow chart of a method for screening for pre-cervical lesions.

Fig. 2 is a block diagram of a spectrum acquisition apparatus.

Detailed Description

In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention will be described in further detail below with reference to the accompanying drawings.

Fig. 1 is a flow chart of a cervical cancer pre-lesion screening method, which mainly comprises the steps of data acquisition, data screening and preprocessing, characteristic parameter calculation, modeling and prediction.

The light source of the spectrum collection system irradiates the cervical tissue through the optical fiber, the receiving optical fiber of the spectrum optical fiber collects the light returned from the tissue and transmits the light to the optical fiber spectrometer, and the optical fiber probe moves around the cervix for a circle to complete the collection of the spectrum.

The diameter of the front end of the double-optical-fiber probe is 6.5mm, and the diameter of the optical fiber is 400 mu m, so that sufficient diffuse reflection light is detected. At the detecting end of the probe, the distance between the two optical fiber sources and the probe is 500 mu m, the detecting depth is generally half of the distance between the sources and the probe, namely 250 mu m, the average thickness of the epithelial tissues of the cervix is about 350 mu m, the distance between the sources and the probe can reduce the influence of the stroma layer as much as possible, and the most useful information is ensured to come from the epithelial tissues with atypical hyperplasia.

The spectrum screening mainly considers the influence of environmental factors and human factors, and at present, the environment light pollution, the probe distance, the probe jitter and the amplitude are mainly screened. And screening the spectrum through the slope of a specific waveband and the peak value of the spectrum.

Experiments show that the spectrums of normal tissues and pathological tissues have obvious difference in certain wave bands, and slope and area under the peak are selected as characteristic parameters in the wave bands. After verification, the slope and the area under the peak are taken as characteristics and can be used as markers for distinguishing normal tissues from pathological tissues. The slope area characteristic selected in the method is S_500-520、S_524-532、S_540-560、S_565-570、S_575-590、S_750-850、A_400-800、A_500-550、A_550-600The subscripts of which represent the wavelength ranges, S the slope, A the area, e.g. S_500-520I.e. the slope of the wavelength 500-520nm band.

The spectrum sampling time interval is adjustable, and the preset value is 150 ms. The spectral data acquisition mainly comprises the following steps:

(1) the device is sterilized before use, and the spectrum of the current detection environment in darkness is prestored and used as the spectrum background.

(2) The front end of the probe is slightly close to the outer edge of the cervical orifice translation band, so that the front end of the probe slightly contacts the translation band and spirally and inwards slides across the whole translation band in the clockwise direction. If the examination is carried out by combining the colposcope, the suspected lesion area can be collected by mainly circling the picture for several times.

(3) And after the equipment is used, storing the acquired data and sterilizing the equipment.

Due to human operation and environmental influences, some unqualified spectrum data exists in the spectrum data. These invalid data need to be screened before data processing can take place. The factors causing unqualified spectra mainly include environmental light pollution, overlarge distance between the probe and the tissue, probe jitter and low spectrum peak value, and the spectrum peak value and the slopes of a plurality of specific wave bands are adopted for screening aiming at the unqualified spectra. After spectrum screening, the qualified spectrum is subjected to average smoothing and normalization. And storing the spectrum data and the original spectrum data after the pretreatment in a database.

For the interference caused by the ambient light, the spectrum takes peaks at 545nm and over 612nm, the spectrum intensities at the two points are usually far greater than the normal spectrum intensity at the periphery, and whether the spectrum is stored or not is determined by comparing the spectrum intensity at 612nm with the spectrum intensity at 700 nm. Under the fixed light source intensity, the peak value of the spectrum of the cervical tissue is generally more than 1500counts (spectrum intensity), and the spectrum lower than 1500counts mostly belongs to the spectrum with poor contact. The frequency spectrum of the jitter interference and the frequency spectrum of the probe far away from the tissue are obviously distinguished from the qualified frequency spectrum at the wave band slope of 470-574nm and the wave band slope of 693-731 nm. Through statistics, the spectrum is filtered when the slope of the 470-574nm band is less than 0.9, and similarly, the spectrum is filtered when the slope of the 693-731nm band of the normalized spectrum is less than-0.053. And filtering out invalid spectrum by combining the above modes.

Characteristic parameter mu'_s、μ_a、S_500-520、S_524-532、S_540-560、S_565-570、S_575-590、S_750-850、A_400-800、A_500-550、A_550-600The eleven dimensions are respectively the reduced scattering coefficient, the absorption coefficient, the spectral slope with the wavelength of 500-520nm, the spectral slope with the wavelength of 524-532nm, the spectral slope with the wavelength of 540-560nm, the spectral slope with the wavelength of 565-570nm, the spectral slope with the wavelength of 575-590nm, the spectral slope with the wavelength of 750-850nm, the area under the spectral peak with the wavelength of 400-800nm, the area under the spectral peak with the wavelength of 500-550nm and the area under the spectral peak with the wavelength of 550-600 nm. The characteristic parameters are stored in a database.

And (3) taking eleven-dimensional characteristic parameters as input, utilizing BP (back propagation) neural network modeling, selecting the model with the highest accuracy as a final classification prediction model from the multiple models, and storing the model in a database. When new data needs to be predicted, the model is extracted from the database, and a prediction result is output.

Fig. 2 is a diagram of a spectrum acquisition apparatus, which mainly includes a fiber probe, a light source, a fiber spectrometer, and a computer. Light generated by the light source is transmitted to the surface of the tissue through the incident optical fiber, is transmitted to the optical fiber spectrometer through the receiving optical fiber after being subjected to diffuse reflection of the tissue, and finally enters the computer for storage.