A combined detection system for non-invasive imaging screening and minimally invasive sampling nucleic acid typing
Through a joint detection system of non-invasive imaging screening and minimally invasive sampling nucleic acid classification, combined with machine learning and deep learning technology, the diagnosis problem of early lung cancer has been solved, and rapid and accurate health screening and diagnosis have been achieved, improving diagnostic efficiency and accuracy sex.
Patent Information
- Application Number
- CN202110829210.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-07-22
- Publication Date
- 2024-11-22
- Estimated Expiration
- 2041-07-22
AI Technical Summary
The existing lung cancer diagnosis methods have limitations in early diagnosis, especially for tumors with smaller lesions. The traditional Chinese medicine "see, smell, ask, and remove" methods lack precise instruments and equipment and personalized intelligent diagnosis capabilities.
Develop a joint detection system for non-invasive imaging screening and minimally invasive sampling nucleic acid typing, combining non-invasive telecentric imaging health analysis system and minimally invasive sampling multi-channel nucleic acid amplification parallel detection system to analyze subjects through machine learning and deep learning Local surface image features, minimally invasive sampling and detection of specific genes, achieving rapid imaging screening and precise medical molecular diagnosis.
Achieve rapid imaging screening of subjects' health status and precise diagnosis of specific genes, improve the accuracy and efficiency of early lung cancer diagnosis, and combine artificial intelligence and big data technology to form a unified technical standard, suitable for personalized non-invasiveness. Quick and intelligent diagnosis.
Smart Images

Figure CN113674839B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to a combined detection system of non-invasive imaging screening and minimally invasive sampling nucleic acid typing, and relates to the technical fields of medical molecular diagnosis and traditional Chinese medicine health analysis. Background Art
[0002] According to the International Agency for Research on Cancer (IARC), there were 2.0939 million new cases of lung cancer and 1.7610 million new deaths in 2018 worldwide, with its morbidity and mortality ranking first among all tumors. my country is a major cigarette sales country, and the morbidity and mortality of lung cancer have always been higher than those in other countries in the world. Studies have found that if the tumor has not spread, the 5-year survival rate can reach 56%, but only 16% of patients can be diagnosed in the early stage; once the tumor spreads and metastasizes, the 5-year survival rate of lung cancer patients is only about 18.6%, and more than half of the patients die within 1 year after diagnosis. Therefore, the key to controlling lung cancer and improving recovery lies in early diagnosis, early detection and early treatment.
[0003] At present, the commonly used clinical lung cancer diagnosis methods include imaging diagnosis, histopathological diagnosis and tumor marker detection. Imaging examinations mainly include chest X-rays, computed tomography (CT) scans, magnetic resonance imaging (MRI) scans and PET / CT scans, which can detect tumors that are large enough (2 mm and above). However, there are certain limitations in the diagnosis of early-stage tumors, especially those with smaller lesions, and the analysis of imaging results must be closely combined with other clinical examination data, such as invasive examinations for diagnosis. Pathological testing is the gold standard for lung cancer diagnosis. It can collect lung secretions (sputum), extract chest effusions (thoracentesis), and use fine needle puncture or surgical excision (biopsy) of the lesion site to obtain suspicious cells or tissues. Sputum cytology is simple and painless, and is easy for patients to accept, but its sensitivity has been difficult to determine. Mediastinoscopy or thoracotomy biopsy needs to be performed by the surgeon in the operating room under general anesthesia, which is very traumatic to the patient. Fiberoptic bronchoscopy technology has the characteristics of low trauma and low cost, but its main drawback is the low detection rate of precancerous lesions. Immunohistochemical testing of tumor markers helps to detect abnormalities in organs. Currently, the use of non-invasive or minimally invasive samples such as sputum and peripheral blood to detect lung cancer-specific molecular markers has become a research hotspot, but it cannot be used directly to diagnose lung cancer and needs to be confirmed in combination with other examination methods.
[0004] The disease diagnosis method of TCM, which uses "looking, smelling, asking, and palpating", has a history of more than 2,000 years. The earliest record of this method can be found in the Yellow Emperor's Classic of Internal Medicine. It is the result of thousands of years of medical practice by Chinese medical practitioners. It is still widely used because of its non-invasive characteristics. It is mainly passed down orally from master to apprentice, and lacks unified technical standards. At present, these TCM technologies are rarely developed into precise instruments and equipment, let alone combined with advanced means such as artificial intelligence and big data to make them more viable. At the same time, there is no personalized non-invasive, fast and intelligent diagnosis for typical diseases such as lung cancer. Summary of the invention
[0005] In view of the above problems, the purpose of the present invention is to provide a combined detection system of non-invasive imaging screening and minimally invasive sampling nucleic acid typing, which can realize rapid imaging screening of the health status of subjects and precise medical molecular diagnosis of specific genes.
[0006] To achieve the above-mentioned object, the present invention adopts the following technical scheme: a combined detection system of non-invasive imaging screening and minimally invasive sampling nucleic acid typing, the system comprising a non-invasive telecentric imaging health analysis system and a minimally invasive sampling multi-channel nucleic acid amplification parallel detection system;
[0007] The non-invasive telecentric imaging health analysis system is configured to take local surface images of the subject, and perform machine learning and cluster analysis on the taken local surface images of the subject to obtain subjects with abnormal changes in image features;
[0008] The minimally invasive sampling multi-channel nucleic acid amplification parallel detection system is configured to perform minimally invasive sampling on subjects with abnormal changes in image features, and to detect and analyze medical molecular typing indicators or physiological and pathological change indicators on the minimally invasive sampling to obtain specific gene detection results of the typing indicators or physiological and pathological change indicators.
[0009] The combined detection system of non-invasive imaging screening and minimally invasive sampling nucleic acid typing, further, the local surface image of the subject is an image of a body part of the subject, including one or more combinations of the sclera, iris, fundus, vitreous body, retina, face, tongue, palm and human skin.
[0010] The non-invasive imaging screening and minimally invasive sampling nucleic acid typing combined detection system, further, the non-invasive telecentric imaging health analysis system includes a pupil focusing aperture angle amplifier, a focusing illumination light source, a side illumination light source, an imaging optical lens, an automatic focusing control unit, a CCD detector and a signal analysis and processing unit;
[0011] The exit aperture of the pupil focusing aperture angle amplifier coincides with the pupil position and is configured to converge the light emitted by the focusing illumination light source into the pupil at a wide angle of oblique incidence to illuminate the eyeball;
[0012] The side lighting light source adopts a dark field lighting method with oblique incidence from all sides and backlighting, and realizes wide field imaging of the local surface through the imaging optical lens;
[0013] The automatic focusing control unit is used to control the movement of the imaging optical lens to ensure clear imaging of the CCD detector;
[0014] The signal analysis processing unit is used to perform machine learning and cluster analysis on the photographed images of the subjects to obtain subjects with abnormal changes in image features.
[0015] The combined detection system of non-invasive imaging screening and minimally invasive sampling nucleic acid typing further includes the process of performing machine learning and cluster analysis on the photographed images of the subjects to obtain subjects with abnormal changes in image features, including:
[0016] Perform data cleaning and enhancement;
[0017] Perform image processing or deep convolutional neural network segmentation of regions of interest on images for data cleaning and enhancement;
[0018] Extract features of regions of interest through machine learning and deep convolutional neural networks;
[0019] Input machine learning or deep learning classifiers to judge the characteristics of the region of interest, train a big data-driven detection model, and perform screening for abnormal physiological and pathological changes in the human body and predict and warn of disease risks.
[0020] The combined detection system of non-invasive imaging screening and minimally invasive sampling nucleic acid typing further includes: performing machine learning and cluster analysis on the captured images of the subjects to obtain the subjects with abnormal changes in image features, including:
[0021] Perform data cleaning and enhancement;
[0022] Segment the region of interest on the cleaned and enhanced images using image processing or deep convolutional neural networks;
[0023] Analyze the image region of interest through machine learning algorithms and / or deep neural networks, and output the analysis results;
[0024] Construct a database consisting of images of different groups, and establish a big data-driven detection model to screen abnormal physiological and pathological changes in the human body and predict and warn of disease risks.
[0025] The combined detection system of non-invasive imaging screening and minimally invasive sampling nucleic acid typing, further, the minimally invasive sampling multi-channel nucleic acid amplification parallel detection system, comprises a fully integrated sample preparation nucleic acid analysis unit, a heating temperature control unit, a motion control unit, an optical detection unit and a signal processing analysis unit;
[0026] The fully integrated sample preparation nucleic acid analysis unit is used to perform nucleic acid reactions on samples, and includes a pretreatment area, a nucleic acid purification area, a sample distribution area, and an amplification reaction area;
[0027] The heating and temperature control unit is used to heat and control the temperature of the pretreatment area and the amplification reaction area;
[0028] The motion control unit is used to control the motion of the pretreatment area, the sample distribution area, and the amplification reaction area;
[0029] The optical detection unit is used to perform real-time fluorescence signal detection on the parallel reaction unit;
[0030] The signal processing and analysis unit is used to perform artificial intelligence, deep learning and / or machine learning analysis on the detected real-time fluorescence signal to obtain specific gene detection results of typing indicators or physiological and pathological change indicators of minimally invasive sampling.
[0031] The combined detection system of non-invasive imaging screening and minimally invasive sampling nucleic acid typing, further, the sample distribution area and the amplification reaction area each include more than two parallel reaction units, and the areas are connected by micro-channels and micro-valves to achieve orderly flow and graded isolation of fluids between the areas; the micro-valves are used to achieve flow path on-off control and prevent cross contamination; the pretreatment area is used to crush the sample and crack the nucleic acid; the nucleic acid purification area is used to purify the mixed sample from which the nucleic acid is cracked and separate the nucleic acid sample to be detected; the sample distribution area is used to mix the separated and purified nucleic acid sample with the amplification reaction reagent, and divide the mixture into multiple independent areas of equal volume for storage according to the number of detection indicators; the amplification reaction area is used to embed and fix specific gene molecular primer probes corresponding to different detection indicators at the bottom of the parallel reaction unit in the amplification reaction area, and distribute the mixture stored in the multiple independent areas of equal volume to the parallel reaction units corresponding to the amplification reaction area.
[0032] The combined detection system of non-invasive imaging screening and minimally invasive sampling nucleic acid typing further performs detection and analysis of medical molecular typing indicators or physiological and pathological change indicators on minimally invasive sampling, and the process of obtaining specific gene detection results of typing indicators or physiological and pathological change indicators includes:
[0033] Perform data preprocessing;
[0034] The preprocessed data is used for index optimization and feature extraction through deep learning or machine learning models;
[0035] Output analysis results through deep learning or machine learning classifiers;
[0036] The big data-driven detection model is trained to obtain specific gene detection of typing indicators or physiological and pathological change indicators.
[0037] The combined detection system of non-invasive imaging screening and minimally invasive sampling nucleic acid typing, further, the specific genes of accurate typing indicators or physiological and pathological change indicators, include but are not limited to one or more of CEA, CYFRA21-1, SCCA, NSE, ProGRP, CA125, EGFR, IDH1, TTF-1, SYN and CD56; wherein:
[0038]
[0039]
[0040]
[0041] The combined detection system of non-invasive imaging screening and minimally invasive sampling nucleic acid typing further includes a communication interface, a mobile terminal and / or a cloud server, and the specific gene detection results are sent to the mobile terminal and / or the cloud server in a wired or wireless manner through the communication interface.
[0042] The present invention adopts the above technical solution, which has the following advantages:
[0043] 1. The combined detection system of non-invasive imaging screening and minimally invasive sampling nucleic acid typing provided by the present invention can realize rapid imaging screening of the health status of the subject and precise medical molecular diagnosis of specific genes, meeting the needs of human health prediction and early warning and precise diagnosis of major diseases;
[0044] 2. Compared with the traditional TCM disease diagnosis method of "looking, smelling, asking and palpating", the present invention develops TCM technology into precise instruments and equipment, forms a unified technical standard, and combines advanced means such as artificial intelligence and big data, which has stronger vitality. At the same time, it includes typical diseases such as lung cancer, and can realize personalized non-invasive, fast and intelligent diagnosis;
[0045] 3. Compared with the traditional sample processing method, the present invention uses minimally invasive sampling, which is done by opening the cover once and inserting it, sealing the whole process to carry out efficient lysis, purification, and safe in-situ transfer of tissues, cells or other eluted and separated samples, effectively avoiding serious problems such as infection of subsequent operators and cross contamination throughout the process;
[0046] 4. The pretreatment area of the present invention adopts a surrounding extruded flexible storage tube to efficiently transfer the eluent in situ, which can effectively reduce the amount of eluent used and increase the initial concentration of the attached sample for subsequent use; at the same time, the subsequent stages use a centrifugal liquid transfer method to effectively reduce the dead volume and realize the multi-index combined detection and accurate diagnosis of trace samples;
[0047] 5. Compared with conventional test tubes, well plates and other carriers (usually requiring 25 μL / indicator), the sample amplification area of the present invention adopts a micro-channel and micro-chamber structure, which not only greatly reduces the consumption of sample reagents (can be <1 μL / indicator), but also can simultaneously detect dozens or even hundreds of specific gene analysis indicators at one time, greatly reducing the manual operation steps and labor intensity, and can avoid the problem of manual operation errors, making minimally invasive sampling or swab sampling detection safer, more robust and efficient;
[0048] 6. The present invention is compatible with wired / wireless communication transmission interfaces, and can be directly displayed on computers, mobile phones or other terminals. It can also directly and automatically send abnormal detection results to cloud servers or national medical and health monitoring network platforms, etc., so as to evaluate or predict the risk of major diseases in the country or various regions, and meet the actual application needs of direct reporting of human health monitoring and health and epidemic prevention in the country or various regions;
[0049] 7. The signal analysis and processing unit provided in the present invention includes a big data priori knowledge base of image feature analysis and specific gene detection of multiple precision medicine molecular typing indicators and physiological and pathological change indicators. It can conduct artificial intelligence joint analysis on the detection results of non-invasive imaging screening and minimally invasive sampling nucleic acid typing, and complement each other to reduce the probability of missed detection by a single method and improve the accuracy of detection and identification.
[0050] In summary, the present invention can be widely used in biomedical molecular diagnosis and traditional Chinese medicine health analysis. BRIEF DESCRIPTION OF THE DRAWINGS
[0051] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present invention. Throughout the accompanying drawings, the same reference numerals are used to represent the same components. In the accompanying drawings:
[0052] Figure 1 is a structural schematic diagram of a joint detection system according to an embodiment of the present invention;
[0053] Figure 2 It is a structural schematic diagram of a non-invasive imaging screening system according to an embodiment of the present invention;
[0054] Figure 3 A schematic diagram of the structure of a minimally invasive sampling nucleic acid typing system according to an embodiment of the present invention;
[0055] Figure 4 A schematic diagram of the algorithm model structure of image segmentation and feature extraction according to an embodiment of the present invention;
[0056] Figure 5 This is a schematic diagram showing an example of the sensitivity experimental results of the lung cancer typing index according to an embodiment of the present invention;
[0057] Figure 6 This is a schematic diagram showing an example of specific experimental results of lung cancer typing indicators according to an embodiment of the present invention;
[0058] Figure 7 This is a schematic diagram of the effect of non-invasive eye imaging screening for lung cancer according to an embodiment of the present invention;
[0059] Figure 8 This is a schematic diagram of the effect of minimally invasive sampling nucleic acid typing for lung cancer according to an embodiment of the present invention. DETAILED DESCRIPTION
[0060] The exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided in order to enable a more thorough understanding of the present invention and to fully convey the scope of the present invention to those skilled in the art.
[0061] It should be understood that the terms used herein are only for the purpose of describing specific example embodiments and are not intended to be limiting. Unless the context clearly indicates otherwise, the singular forms "one", "an" and "said" as used herein may also be meant to include plural forms. The terms "include", "comprise", "contain", and "have" are inclusive, and therefore specify the existence of stated features, steps, operations, elements and / or parts, but do not exclude the existence or addition of one or more other features, steps, operations, elements, parts, and / or combinations thereof. The method steps, processes, and operations described herein are not interpreted as necessarily requiring them to be performed in the specific order described or illustrated, unless the execution order is clearly indicated. It should also be understood that additional or alternative steps may be used.
[0062] For ease of description, spatially relative terms may be used herein to describe the relationship of one element or feature relative to another element or feature as shown in the figures, such as "inside", "outside", "inner side", "outer side", "below", "above", etc. Such spatially relative terms are intended to include different orientations of the device in use or operation in addition to the orientation depicted in the figures.
[0063] In the early lesions of lung cancer, there are often some changes in the related cell and gene levels, and some special symptoms will appear in the "look, smell, ask, and palpate" of traditional Chinese medicine. At present, there is no good technical means to obtain the correspondence between these special symptoms and microscopic changes in cells and genes. Therefore, it is very necessary to develop a joint technology that combines the "look, smell, ask, and palpate" of traditional Chinese medicine with Western medicine imaging, cell / gene level in vitro diagnosis, etc., and further combine AI with big data analysis, so that the test results of traditional Chinese medicine and Western medicine can be applied to each other, so as to better realize the early diagnosis of lung cancer, especially suitable for screening and accurate diagnosis of high-risk groups for lung cancer.
[0064] Embodiment 1
[0065] like Figure 1 As shown, the combined detection system of non-invasive imaging screening and minimally invasive sampling nucleic acid typing provided in this embodiment includes non-invasive telecentric imaging screening No-Invasive Imaging, minimally invasive sampling nucleic acid precise typing detection Minimal Invasive Typing, artificial intelligence and big data analysis AI-Bigdata Analysis, interactive display terminal Handset and cloud medical resource network server MEDNET.
[0066] The combined detection system of non-invasive imaging screening and minimally invasive sampling nucleic acid typing provided in this embodiment includes a non-invasive telecentric imaging health analysis system and a minimally invasive sampling multi-channel nucleic acid amplification parallel detection system.
[0067] The non-invasive telecentric imaging health analysis system is configured to capture images of the subject's eyes, face, tongue, hands, and other local surfaces of the human body, and perform machine learning and cluster analysis on the captured images of the subject to obtain subjects with abnormal changes in image features;
[0068] The minimally invasive sampling multi-channel nucleic acid amplification parallel detection system is configured to perform minimally invasive sampling on subjects with abnormal changes in image features, including obtaining rice-grain-sized (single-side size 2-3mm) biopsy tissue samples using organoscopes or punctures, or scraping some epidermal or debris samples from the lesions using swabs, chisels, etc., and performing specific gene detection and cluster analysis of medical molecular typing indicators or multiple physiological and pathological change indicators on the minimally invasive samples to obtain accurate typing indicators or specific gene detection results of multiple physiological and pathological change indicators.
[0069] In some preferred embodiments of the present invention, Figure 2As shown, the non-invasive telecentric imaging health analysis system of this embodiment includes a pupil focusing aperture angle amplifier KC, a focusing illumination light source S1, a four-directional or annular illumination light source S2, an imaging optical lens LEN, an automatic focusing control unit KZ, a CCD detector DE, a signal analysis and processing unit PR, a communication interface TX, a display terminal Handset and a cloud big data server MEDNET.
[0070] The exit aperture of the pupil focusing aperture amplifier KC coincides with the pupil position of the eye EYE, and through refraction or reflection, the light emitted by the focused illumination light source S1 is focused at a wide viewing angle of more than 120 degrees and is obliquely incident and converged into the pupil to illuminate the eyeball.
[0071] The four-directional or annular lighting source S2 adopts a dark field lighting method with oblique incidence from all sides and backlighting, and realizes shadowless wide-field imaging with a large depth of field of the sclera, iris, face, tongue, palm and other local surfaces of the human body through the imaging optical lens LEN. The imaging methods of the optical lens LEN include but are not limited to telecentric imaging, zoom imaging, variable magnification imaging or confocal scanning imaging.
[0072] The automatic focusing control unit KZ is used to control the forward and backward movement of the imaging optical lens LEN to ensure clear imaging of the CCD detector DE, and to store, analyze and process the collected images and display the results through the signal analysis and processing unit PR. It can also send the collected images, analysis and processing results, etc. to the mobile terminal Handset and / or the cloud server MEDNET through the communication interface TX in a wired or wireless manner.
[0073] In summary, this embodiment adopts a method of combining a focused illumination light source S1 and a pupil focusing aperture angle amplifier KC to perform pupil focusing wide-angle incident astigmatism dark field illumination, and combines it with an imaging optical lens LEN for imaging, thereby expanding the imaging depth of field range and achieving clear imaging of the fundus, vitreous body, retina, etc. with a large solid angle of more than 120 degrees and a wide depth of field range.
[0074] In some preferred embodiments of the present invention, the non-invasive telecentric imaging health analysis system of this embodiment adopts one or more combinations of images of the sclera, iris, fundus, vitreous body, retina, etc. of the eye or images of the face, tongue, palm and other local surfaces of the human body. Among them, the process of performing machine learning and cluster analysis on the captured images of the subjects to obtain subjects with abnormal changes in image features includes:
[0075] S1. Data cleaning and enhancement
[0076] Specifically, data cleaning and enhancement include: first, using fast Fourier transform to clear blurred data; then performing normalization processing on the image including histogram normalization and zero mean centering; and subsequently using Gaussian blur processing and image mixing enhancement to enhance the features of the image.
[0077] S2. Improve the U-net model to segment the region of interest, specifically:
[0078] S21. Introduce the attention mechanism in the U-net model. The attention mechanism can help the network resist the interference of noise and extract features that are more conducive to distinction, thereby improving the segmentation performance of the model; use the "He Initialization" method to initialize the network weights. This method is suitable for the initialization method of the ReLU activation function and can prevent the output of the deep network from disappearing or exploding; perform batch normalization (BatchNormalization) in the convolution layer and the upsampling convolution layer. The data is standardized in each step of the network operation to reduce the impact of the distribution of input data on the network and speed up the network operation.
[0079] S22. Design the improved U-net model, the specific structure is:
[0080] The improved network structure is a "U-shaped" one, including a contraction path and an expansion path. The attention module CBAM is introduced at the intersection of the contraction path and the expansion path. The contraction path is equivalent to an encoder, which contains multiple convolution units. Each convolution unit contains a convolution layer with a convolution kernel of 3×3, a batch normalization layer, and a ReLU activation function connected in sequence. The two convolution units are a 2×2 maximum pooling layer with a stride of 2 to achieve feature downsampling. This structure constitutes a convolution module; four stacked convolution modules constitute the contraction path. The depth of the convolution layer in each convolution module is consistent, but the depth of the convolution layer of different convolution modules is different; as the network depth increases, the number of channels of the convolution block also increases; the number of input channels of the network is 3, and the number of channels then increases to 64, 128, 256, and 512; at the same time, the size of the feature map output by the convolution layer becomes smaller and smaller, and finally becomes H×W×512.
[0081] The attention module CBAM is divided into channel attention and module spatial attention module. The input features are decomposed into spatial features and channel features. The features first enter the channel attention module, which includes a global average pooling layer and a maximum pooling layer, followed by a multi-layer perceptron with 64 hidden layer neurons, a ReLU function as the activation function, and 512 output layer neurons. Then the computing unit adds the output of the perceptron at the element level, passes through the Sigmoid function, obtains the attention weight of the original input feature according to the output of the sigmoid function, and then scales the original input feature according to the weight to obtain the channel attention feature. Then, the spatial attention module is followed, and the features are average pooled and maximum pooled on the channel in this module, and then the two output features are spliced, and then a convolution layer with a convolution kernel of 7×7 is passed, and the Sigmoid function is used as the activation function to obtain the spatial attention weight. Finally, the input feature is scaled according to the weight to obtain the final feature after the attention module.
[0082] The expansion path includes multiple upsampling convolution modules, each of which includes three upsampling convolution layers and one pooling layer. The convolution kernel size, number of channels and step size of the upsampling convolution module are consistent with the convolution module corresponding to the contraction path. In the upsampling convolution module, the high-resolution information from the contraction path and the low-resolution information from the upsampling convolution unit of the expansion path are converged through crossing connections to achieve accurate positioning of the segmented area. Finally, a convolution layer with a convolution kernel of 1×1 and a step size of 1 is connected to the output of the expansion path, and the sigmoid function is used as the activation function.
[0083] S23. The specific process of segmenting the region of interest through the improved U-net model is:
[0084] After the image is input, it first enters the contraction path described in S22, passes through the above-mentioned multiple convolution units, realizes feature downsampling, gradually compresses the size and number of channels of the feature, and finally obtains a feature of size H×W×512;
[0085] Then the features are input into the CBAM module described in S22 to obtain channel attention features and spatial attention features; finally, the attention features are input into the expansion path described in S22, and after passing through the above-mentioned multiple upsampling convolution modules, the size and number of channels of the feature map are gradually restored, and finally a sclera segmentation map with the same size as the input image is output. The improved U-net network structure and image segmentation results are shown in Figure 2. Figure 4 (a) shown.
[0086] S3. Extract deep features of the region of interest through the improved Resnet model
[0087] Specifically, the deep feature extraction network model consists of three parts: residual network module, attention module and multi-layer perceptron module.
[0088] Residual network module: The residual network module consists of multiple residual learning units. Each learning unit has two convolutional layers of size 3×3 and the same number of output channels. The stride of the first convolutional layer is 2, and the stride of the second convolutional layer is 1. The ReLU function is used as the activation function after each convolutional layer. The convolutional layer with a stride of 2 implements feature downsampling, making the feature input size become 1 / 2 of the original (H×W→H / 2×W / 2).
[0089] Attention module: It uses CBAM to highlight important information in features from the two dimensions of channel and space, so that the network can learn important information from the two dimensions of content and position in the channel and space respectively, and improve the ability of feature presentation.
[0090] Multilayer Perceptron Module: This module consists of fully connected layers. The input is a feature size of 1×512 after the average layer. The first hidden layer has 512 neurons and the activation function is ReLU. The second hidden layer has 128 neurons and the activation function is ReLU. The final output layer has two neurons and the activation function is Sigmoid function.
[0091] like Figure 4 As shown in (b), the input of the network is a sclera image with a dimension of 3×512×512. "3" represents the three color channels of RGB, and "512×512" is the pixel size of the image. The first layer of the network is a convolution layer with a convolution kernel size of 7×7, a stride of 2, and an output channel number of 64. The activation function after the convolution layer uses the ReLU function, followed by a maximum pooling layer with a convolution kernel of 3×3 and a stride of 2. The network is then connected to four residual network modules in sequence. The number of channels of the two convolution layers in the four residual network modules are 64, 128, 256, and 512, respectively, to achieve the purpose of feature dimensionality increase.
[0092] It should be noted that since the maximum pooling layer with a stride of 2 has been passed, the stride of the first convolutional layer of the residual network module is 1. After CBAM, the output feature size is adjusted to 1×512 by the average pooling layer. During the training process, this part of the output features will enter the multi-layer perceptron module. The output of the training network is a 1×2 vector. The network is trained and optimized according to the y label in the data set. After the training is completed, the multi-layer perceptron module is removed, and the output of the average pooling layer is used as the deep feature extracted by the model.
[0093] S4, input multiple instance learning (MIL) classifier for judgment, such as Figure 4 The structure of the multi-instance learning classifier (MIL) shown in (c) is:
[0094] Part ① in the figure is the feature input module, which uses the feature extraction method of S3 to extract the low-dimensional deep features of each sclera image in the dataset, also called sample features. Each sample contains eight sclera images, so it corresponds to eight corresponding low-dimensional deep features, which are packaged into a package and input into the attention layer.
[0095] Part ② in the figure is a symmetric function p that fuses multiple example features, combining eight independent example features into a bag feature. In ②, the example features enter the three-layer attention layer to obtain attention weights, and then the weights are assigned to the identity mapping of the example features, and feature embedding is achieved at the same time.
[0096] Part ③ in the figure is a multi-layer perceptron. The activation function of the first hidden layer is Leaky ReLU. The second hidden layer is connected to the third Dropout layer, in which a certain proportion of neurons are discarded to prevent overfitting of the model. Finally, after the Sigmoid function, the "0 / 1" label is output to achieve classification: for example, "0" can represent that the subject corresponding to the sample is not sick, and "1" can represent that the subject corresponding to the sample is sick.
[0097] S5. Through training, a big data-driven detection model is obtained to realize the screening of abnormal physiological and pathological changes in the human body and the prediction and early warning of the risk of disease, that is, a detection model that can be used for the screening of abnormal physiological and pathological changes in the human body and the prediction and early warning of the risk of disease.
[0098] In some preferred embodiments of the present invention, non-invasive telecentric imaging screening can also use non-image feature extraction methods, such as determining whether it is lung cancer based on sclera images, such as Figure 7 As shown, data preprocessing and region of interest segmentation are first performed, and then the analysis results are directly analyzed and output through a multiple instance learning (MIL) classifier; similarly, a database consisting of images of different groups can be constructed, and a big data-driven detection model can be established to screen abnormal physiological and pathological changes in the human body and predict and warn of disease risks.
[0099] Specifically, the scleral image preprocessing and segmentation of the region of interest are as follows: the collected scleral images are preprocessed and then input into the classification model, the scleral area is marked with a bounding box on each image, and then cropped and resized to form a single image of size 512×384. For each patient, 5 scleral images can be used to capture the left eye, and the other images can capture the right eye. In order to simplify the learning of the model, the 5 images of the right eye are horizontally flipped.
[0100] The multiple instance learning model is as follows: for each patient, different regions of both eyes are shown in different images. Through multiple instance learning (MIL), analysis based on all sclera images of each patient can be achieved; the preprocessed sclera images are input into the multiple instance learning (MIL) model, which uses Renet-18 as the kernel; Figure 7 As shown, all images of the patient are fed into ReNET-18, and the sequence of feature vectors is extracted by RESNET-18; in order to utilize the information from all images of each patient, a fused feature vector is calculated, which collects useful information from each feature vector as a reliable representative. The fused feature vector is aggregated from the feature vector using average pooling, and finally the fused feature vector is fused and fed into a multi-layer perceptron (MLP) to obtain a prediction score. The combined use of the above models, driven by big data, constitutes a detection model that can be used for screening abnormal conditions of physiological and pathological changes in the human body and predicting and warning of the risk of disease.
[0101] In some preferred embodiments of the present invention, Figure 3 As shown, the minimally invasive sampling multi-channel nucleic acid amplification parallel detection system includes a fully integrated sample preparation nucleic acid analysis unit UTAS, a heating temperature control unit WK, a motion control unit MK, an optical detection unit OD, a signal analysis and processing unit PR, a communication interface TX, a mobile terminal Handset, and a cloud server MEDNET.
[0102] The fully integrated sample preparation nucleic acid analysis unit UTAS includes a pretreatment area YC, a nucleic acid purification area TC, a sample distribution area FP and an amplification reaction area DC.
[0103] The pretreatment zone YC includes a flexible preservation tube with an injection port on the top. The pretreatment zone uses an in-situ transfer of high-efficiency eluent by squeezing the flexible preservation tube around it, which can effectively reduce the amount of eluent used and increase the initial concentration of the attached sample for subsequent use. Centrifugal liquid transfer is used between subsequent stages to effectively reduce dead volume and achieve multi-indicator combined detection and accurate diagnosis of trace samples.
[0104] The sample distribution area FP and the amplification reaction area DC both include more than two parallel reaction units A1∽A3, U1∽U3, C1∽C3, etc. with the same volume and independent of each other. The areas are connected by micro-pipes and micro-valves to achieve orderly flow and graded isolation of fluids between areas; preferably, the micro-valves adopt but are not limited to one or more of the expansion methods LH such as liquid metal heating or light irradiation or water injection, memory alloy heating or electric expansion or mechanical extrusion YH, siphon CH, etc. to achieve flow path on-off control and prevent cross contamination.
[0105] In some implementations, the heating temperature control unit WK may use one or a combination of heating films, heating rods, semiconductor heating, laser irradiation, infrared radiation or chemical heating to perform program heating and temperature control on the pretreatment area YC and the amplification reaction area DC, such as periodic cycle temperature change, isothermal or constant temperature, etc., to meet the temperature control conditions required for the experiment;
[0106] In other implementations, the motion control unit MK can use a control method that combines one or more multi-channel motors, servos or other motion mechanisms to perform motion control on the pretreatment area YC, sample distribution area FP, amplification reaction area DC, etc. to meet the needs of automatic control during the experiment.
[0107] In some further implementations, the signal processing and analysis unit PR is used to detect and analyze the medical molecular typing index or physiological and pathological change index of the minimally invasive sampling, and obtain the specific gene detection results of the typing index or the physiological and pathological change index, including:
[0108] First, perform data desensitization, cleaning and other preprocessing;
[0109] Then, artificial intelligence, deep learning or traditional machine learning models are used to optimize specific gene indicators and extract features;
[0110] For example, for nucleic acid amplification methods, for each sample, several specific gene nucleic acid fragments can be detected. Each nucleic acid fragment is an indicator. It can detect the presence or absence of this indicator, or the absolute number of this indicator. Taking cancer as an example, cancer is caused by gene mutation, that is, some fragments of nucleic acid have changed, but some of these changes are highly pathogenic and some are not. The optimization process is to select those indicators that are closely related to the disease to be detected. Feature extraction uses traditional machine learning such as SVM or deep learning methods such as MLP, which are not limited here.
[0111] Finally, the analysis results are output through artificial intelligence, deep learning or traditional machine learning classifiers, and this method is used to train a big data-driven detection model, effectively improving the accuracy of the combined detection and analysis results of multiple precision medicine molecular typing.
[0112] In some preferred embodiments of the present invention, the signal analysis and processing unit PR is provided with a big data prior knowledge base of image feature analysis and specific gene detection including a variety of precision medicine molecular typing indicators and physiological and pathological change indicators, which can perform artificial intelligence joint analysis on the detection results of non-invasive telecentric imaging screening and minimally invasive sampling nucleic acid precise typing, complement each other, reduce the probability of missed detection by a single method, and improve the accuracy of detection and identification.
[0113] Some preferred embodiments of the present invention also include wired / wireless communication functions, which can display the analysis results locally or remotely by optical and acoustic means or transmit them to the mobile phones of relevant personnel for text message display, as well as transmit them to the national human health monitoring network platform, etc., so as to evaluate or predict the risk of disease in the whole country or various regions, etc.
[0114] Embodiment 2
[0115] Minimally invasive sampling nucleic acid precise typing uses a minimally invasive sampling multi-channel nucleic acid amplification parallel detection system, uses bronchoscope or puncture to obtain rice-sized (single-side size 2-3mm) biopsy tissue samples or scrapes some epidermis or debris samples from the lesion, and then directly puts the minimally invasive sample into a fully integrated sample preparation nucleic acid analysis unit. After a series of automated biochemical reaction processing, the precise medical molecular typing indicators or multiple physiological and pathological change indicators are finally obtained. The specific process includes:
[0116] S1. Add the original sample tissue obtained by minimally invasive sampling directly into the fully integrated sample preparation and nucleic acid analysis unit UTAS from the injection port;
[0117] S2. In the pretreatment area YC, the original sample in the flexible storage tube is broken by one or a combination of mechanical, ultrasonic, heating or chemical reagent methods to lyse the nucleic acid;
[0118] S3. In the nucleic acid purification zone TC, the mixed sample from which the nucleic acid is lysed is purified by a method such as centrifugation to obtain the supernatant, filter paper capture and elution, or dielectric electrophoresis separation to separate the nucleic acid sample to be detected;
[0119] S4. In the sample distribution area FP, the separated and purified nucleic acid sample is mixed with the amplification reaction reagent, and evenly distributed into n×m+2 units, such as A1∽A3, U1∽U3, etc., according to the number n of indicators or types of molecular typing detection or physiological and pathological changes, the number m of repetitions required for each indicator, and positive and negative quality control references;
[0120] S5. In the amplification reaction area DC, the specific gene molecular primer probes corresponding to the above-mentioned multiple molecular typing indicators or specific detection indicators of physiological and pathological changes, positive and negative quality control references, etc. are first embedded and fixed at the bottom of the parallel reaction units with the same volume and independent of each other in the amplification reaction area, such as C1 to C3, etc.; then, the separated and purified nucleic acid sample and the amplification reaction reagent mixture in the sample distribution area FP are evenly distributed to the parallel reaction units with the same volume and independent of each other in the amplification reaction area, such as C1 to C3, etc., by centrifugation, positive or negative air pressure or extrusion;
[0121] S6. According to the temperature control condition requirements of nucleic acid amplification, the heating temperature control unit WK heats and programs the temperature of the parallel reaction units C1~C3 with the same volume and independent of each other in the amplification reaction area DC; the optical detection unit OD performs real-time fluorescence signal detection on the parallel reaction units C1~C3; finally, the signal analysis and processing unit PR performs artificial intelligence analysis and result display on the real-time fluorescence signal, completes the precise medical molecular typing index or the specific gene detection and artificial intelligence analysis of the minimally invasive sampled tissue, and obtains the precise typing index or the specific gene detection and identification results of the specific gene detection ...
[0122] In some preferred embodiments of the present invention, precise typing indicators or specific gene detection of multiple physiological and pathological change indicators only require the addition of a rice-grain-sized biopsy tissue sample or other trace original sample at one time, and more than two, up to 100, molecular typing indicators or specific gene detection of physiological and pathological change indicators can be performed in parallel. The sample reagent consumption for each detection indicator can be reduced from tens of microliters to 100 nanoliters as needed.
[0123] In some preferred embodiments of the present invention, specific genes of precise typing indicators or physiological and pathological change indicators include but are not limited to part or all of CEA, CYFRA21-1, SCCA, NSE, ProGRP, CA125, EGFR, IDH1, TTF-1, SYN and CD56 of lung cancer, and a set of corresponding primer designs are as follows:
[0124]
[0125]
[0126]
[0127] The sensitivity of all the above indicators can reach 1.0×10 2 copies / μL, some indicators such as ProGRP and NSE can reach 1.0×10 1 The following is a sensitivity verification experiment result using ProGRP as an example. Figure 5 As shown, from left to right, they represent different template concentrations: 1.0×10 6 ,1.0×10 5 ,1.0×10 4 ,1.0×10 3 ,1.0×10 2 and 1.0×101 The above 11 indicators were tested for sample cross-specificity. The experimental results are as follows: Figure 6 As shown, from Figure 6 It can be seen that there are significant specific differences in the expression of these 11 indicators in different histological types [benign (Benign), non-small cell lung cancer (NSCLC-non-small cell lung cancer) (including adenocarcinoma LUAD-lung adenocarcinoma and squamous cell carcinoma LUSC-lung squamous cell carcinoma), small cell lung cancer (SCLC-small cell lung cancer), metastatic lung cancer and primary lung cancer (including NSCLC and SCLC)].
[0128] In some preferred embodiments of the present invention, the non-invasive telecentric imaging health analysis system of the present invention is used in combination with the minimally invasive sampling multi-channel nucleic acid amplification parallel detection system, such as Figure 7 , Figure 8 As shown. Figure 7 The following is an example of a set of experimental results of eye health analysis of lung cancer. Figure 7 In the figure, part A shows the scleral images of each subject; parts B to E show the average ROC curves with 95% CI and the best points of different scleral image input strategies, including separate eyes, only two images in the central direction, and all images except the central direction. Different line shapes represent different input strategies; part F shows the comparison results of different scleral image analysis strategies, including separate eyes, only two images in the central direction, and all images except the central direction, where ± refers to the 95% confidence interval. Figure 8 The following is an example of a set of experimental results for accurate typing of lung cancer using minimally invasive sampling, multi-channel nucleic acid amplification and parallel detection. Figure 8 The results of an experiment involving 95 subjects (including 20 patients with benign lung lesions and 75 patients with malignant lung lesions) were presented in Figure 8 In A, the expression of CYFRA21-1 in lung squamous cell carcinoma (LUSC) was higher than that in benign lung lesions (Benign), lung adenocarcinoma (LUAD) and small cell lung cancer (SCLC). Figure 8 In B, EGFR expression is related to metastasis, and its expression in lung metastatic cancer (Metastatic) is higher than that in benign lung lesions (Benign) and primary lung cancer (Primary). Figure 8 In C, the expression of CD56 in small cell lung cancer (SCLC) is higher than that in benign lung lesions (Benign), lung adenocarcinoma (LUAD) and lung squamous cell carcinoma (LUSC). Figure 8In D, the expression of IDH1 in small cell lung cancer (SCLC) was higher than that in benign lung lesions (Benign) and non-small cell lung cancer (NSCLC).
[0129] After determining the subtypes that each biomarker could distinguish, all 95 clinical samples were binary classified according to the cutoff values of the 11 biomarkers in turn. In short, for each biomarker, samples with high mRNA expression were considered as specific subtypes, and samples with low mRNA expression were considered as other subtypes. Figure 8 E shows the detailed classification performance of 11 biomarkers, with evaluation indicators including sensitivity and specificity. CYFRA21-1 and SCCA can be combined to measure LUSC. EGFR can effectively identify metastatic lung cancer with high sensitivity and specificity. For SCLC, CEA and ProGRP showed better specificity than the other five markers. But in general, statistical analysis based on a single biomarker cannot clearly distinguish different subtypes.
[0130] In order to comprehensively consider multiple biomarkers, the performance of machine learning classification models in identifying different subtypes of lung cancer was evaluated under five binary classification settings: benign and malignant, adenocarcinoma and non-adenocarcinoma, squamous cell carcinoma and non-squamous carcinoma, small cell lung cancer and non-small cell lung cancer, lung metastasis and non-lung metastasis. By summarizing the sensitivity and specificity at different threshold levels, the receiver operating characteristic (ROC) curve was plotted, and the AUC (area under the ROC curve) was calculated. The performance of the five binary classification models is shown in Figure 2. Figure 8 FK. The test set is a 10-repetition triple cross validation procedure, which generates 30 independent experiments for each classification task, and the results are the average of all test sets. The ROC curve and 95% confidence interval are shown in Figure 8 FJ, where the best point is the threshold point with the shortest distance to the upper left corner of the ROC curve. In addition, the sensitivity, specificity and AUC scores of the best point are shown as Figure 8 K, the mean and 95% confidence interval are given respectively. In general, benign lesions and small cell lung cancer are easier to identify, while adenocarcinoma, squamous cell carcinoma and lung metastasis have slightly lower sensitivity and specificity. However, all binary classifications achieved an AUC score of 0.9 or higher. The results show that the use of mRNA biomarkers for minimally invasive and rapid diagnosis of lung cancer is effective. In addition, we Figure 8 The final results of the feature selection module are provided in L. Although some biomarkers were not significant for specific binary classifications, the combination of multiple biomarkers showed clear and meaningful improvements in the classification of lung cancer compared to a single biomarker.
[0131] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein, and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention. Sequence Listing <110> Tsinghua University <120> A combined detection system for non-invasive imaging screening and minimally invasive sampling nucleic acid typing <160> 66 <170> SIPOSequenceListing 1.0 <210> 1 <211> 16 <212> DNA <213> Artificial Sequence <400> 1 caccactgcc actcac 16 <210> 2 <211> 17 <212> DNA <213> Artificial Sequence <400> 2 tctcgaccac tatgcgg 17 <210> 3 <211> 44 <212> DNA <213> Artificial Sequence <400> 3 ctggggcaga ttgtgaaggt gtgtgaatcc acgccgttca atgt 44 <210> 4 <211> 44 <212> DNA <213> Artificial Sequence <400> 4 aagagtggat ggcaaccagt gtgtctgggg tagcttgttg agtt 44 <210> 5 <211> 13 <212> DNA <213> Artificial Sequence <400> 5 acctcctccc tct 13 <210> 6 <211> 16 <212> DNA <213> Artificial Sequence <400> 6 ttataggatg taatag 16 <210> 7 <211> 17 <212> DNA <213> Artificial Sequence <400> 7 agtgacatgc agccaat 17 <210> 8 <211> 18 <212> DNA <213> Artificial Sequence <400> 8 gctttcatgc agctgtga 18 <210> 9 <211> 44 <212> DNA <213> Artificial Sequence <400> 9 agcgacctcc cggtattctg tgtgtcgagc agaaccggaa ggat 44 <210> 10 <211> 44 <212> DNA <213> Artificial Sequence <400> 10 cacacggagc agctccagat gagtgtgtgc tcaatctcaa gacc 44 <210> 11 <211> 18 <212> DNA <213> Artificial Sequence <400> 11 gctggtgaac ggcttcag 18 <210> 12 <211> 17 <212> DNA <213> Artificial Sequence <400> 12 aggttactga cgcggcg 17 <210> 13 <211> 20 <212> DNA <213> Artificial Sequence <400> 13 tcaagaaatt tccagaccag 20 <210> 14 <211> 16 <212> DNA <213> Artificial Sequence <400> 14 ttattctccc tgccct 16 <210> 15 <211> 50 <212> DNA <213> Artificial Sequence <400> 15 ttgactttcc accgagttaa tctgtgtgtc tactgatttt gcaaatgctc 50 <210> 16 <211> 48 <212> DNA <213> Artificial Sequence <400> 16 aacctatttc ctgatgggac tattgtgtga tagattgcgt tcacaaga 48 <210> 17 <211> 16 <212> DNA <213> Artificial Sequence <400> 17 tctttcgact tttctg 16 <210> 18 <211> 16 <212> DNA <213> Artificial Sequence <400> 18 gcaatgatga cactgg 16 <210> 19 <211> 17 <212> DNA <213> Artificial Sequence <400> 19 tctctggctg tgtaagg 17 <210> 20 <211> 17 <212> DNA <213> Artificial Sequence <400> 20 agacctctgc acagtcg 17 <210> twenty one <211> 41 <212> DNA <213> Artificial Sequence <400> twenty one cacgttgaag gccgcagtgt gtggaactgc ccctgtatcg c 41 <210> twenty two <211> 42 <212> DNA <213> Artificial Sequence <400> twenty two tggctctcat gctggcaaca agtgtgtggc atcccgaaag ct 42 <210> twenty three <211> 18 <212> DNA <213> Artificial Sequence <400> twenty three aggatgaggt cgttcccg 18 <210> twenty four <211> 19 <212> DNA <213> Artificial Sequence <400> twenty four tgcaggagtt catcctccc 19 <210> 25 <211> 17 <212> DNA <213> Artificial Sequence <400> 25 gctgaccaat gtacccg 17 <210> 26 <211> 16 <212> DNA <213> Artificial Sequence <400> 26 acgaagtgct gattgc 16 <210> 27 <211> 41 <212> DNA <213> Artificial Sequence <400> 27 ctcagctgct gctgctctgt gtgtggggca cttaatgggg a 41 <210> 28 <211> 43 <212> DNA <213> Artificial Sequence <400> 28 acatcaggtg ggaagagctg tgtgggctgg tggtttctgt tct 43 <210> 29 <211> 20 <212> DNA <213> Artificial Sequence <400> 29 gaaacagaag aactcccctg 20 <210> 30 <211> 20 <212> DNA <213> Artificial Sequence <400> 30 gctgggtctc agaagcaaag 20 <210> 31 <211> 18 <212> DNA <213> Artificial Sequence <400> 31 tctggtgaca cacagatg 18 <210> 32 <211> 16 <212> DNA <213> Artificial Sequence <400> 32 atgtccagac cgaagt 16 <210> 33 <211> 44 <212> DNA <213> Artificial Sequence <400> 33 ggccggtctc tcatggatgg tgtgtgcaca acaagcccag agtc 44 <210> 34 <211> 43 <212> DNA <213> Artificial Sequence <400> 34 gacaccacac agaagccggt gtgtccctac attggtcact gcg 43 <210> 35 <211> twenty two <212> DNA <213> Artificial Sequence <400> 35 gctcaaattg aggtgaactg gt 22 <210> 36 <211> 20 <212> DNA <213> Artificial Sequence <400> 36 ccatgcatca tcacaaacac 20 <210> 37 <211> 16 <212> DNA <213> Artificial Sequence <400> 37 cacgtgcaag cacctg 16 <210> 38 <211> 18 <212> DNA <213> Artificial Sequence <400> 38 cccttcgcac ttctcact 18 <210> 39 <211> 44 <212> DNA <213> Artificial Sequence <400> 39 ggcaccaaag cttttgccct gtgtgatgct ctacaacccc acca 44 <210> 40 <211> 42 <212> DNA <213> Artificial Sequence <400> 40 gtgacaacgg ctcgtgctgt gtggccgtct tcctccatct ca 42 <210> 41 <211> 18 <212> DNA <213> Artificial Sequence <400> 41 acatccatct gacgtggt 18 <210> 42 <211> 14 <212> DNA <213> Artificial Sequence <400> 42 gcctgtgggc gaca 14 <210> 43 <211> 16 <212> DNA <213> Artificial Sequence <400> 43 taggctcaga cgacat 16 <210> 44 <211> 16 <212> DNA <213> Artificial Sequence <400> 44 ctttctggta tgcggt 16 <210> 45 <211> 46 <212> DNA <213> Artificial Sequence <400> 45 tgcacgtcac catagttttt gtgtgcaagc tatgaaatca gaggga 46 <210> 46 <211> 40 <212> DNA <213> Artificial Sequence <400> 46 gatgaccagc gtgctggttt tgtgtgacgg gttacagtcc 40 <210> 47 <211> 17 <212> DNA <213> Artificial Sequence <400> 47 tacaggccca gatagcc 17 <210> 48 <211> 18 <212> DNA <213> Artificial Sequence <400> 48 acagtagaag caggctgc 18 <210> 49 <211> 16 <212> DNA <213> Artificial Sequence <400> 49 agtctgaaac cgaaac 16 <210> 50 <211> 19 <212> DNA <213> Artificial Sequence <400> 50 ggttagaatt tctgcactt 19 <210> 51 <211> 47 <212> DNA <213> Artificial Sequence <400> 51 aggattttct tggttggaat cctgtgtgca aggctgtcga agaagtg 47 <210> 52 <211> 46 <212> DNA <213> Artificial Sequence <400> 52 agttgcctat caattgttcg ggaatgtgtg ccaatttctg gttctt 46 <210> 53 <211> 18 <212> DNA <213> Artificial Sequence <400> 53 acctctttta cctgggga 18 <210> 54 <211> twenty two <212> DNA <213> Artificial Sequence <400> 54 gcatatcttg ggaagtagaa gc 22 <210> 55 <211> 15 <212> DNA <213> Artificial Sequence <400> 55 cactacccct tcggt 15 <210> 56 <211> 17 <212> DNA <213> Artificial Sequence <400> 56 catcttctcg tgcttca 17 <210> 57 <211> 44 <212> DNA <213> Artificial Sequence <400> 57 cacctccaaa ggtgagtcct gtgtggtgac ctccatcccc aact 44 <210> 58 <211> 43 <212> DNA <213> Artificial Sequence <400> 58 tcgtctcagg gaccttgctg tgtgaagggc cacaaagagt gtc 43 <210> 59 <211> 17 <212> DNA <213> Artificial Sequence <400> 59 gctgcgtgag ttgttgt 17 <210> 60 <211> 19 <212> DNA <213> Artificial Sequence <400> 60 gtacgagagg gaacaggag 19 <210> 61 <211> 17 <212> DNA <213> Artificial Sequence <400> 61 gtggcaggat gccaaag 17 <210> 62 <211> 18 <212> DNA <213> Artificial Sequence <400> 62 ttggcgcatt ctaacatg 18 <210> 63 <211> 44 <212> DNA <213> Artificial Sequence <400> 63 tggtgagggt ggaggaggaa tgtgttcccc caatggagaa aagc 44 <210> 64 <211> 42 <212> DNA <213> Artificial Sequence <400> 64 catcgacgac gccgttacgt gtgtgttgac ggtggcctct ga 42 <210> 65 <211> 18 <212> DNA <213> Artificial Sequence <400> 65 cactgagacg ctgctggt 18 <210> 66 <211> 17 <212> DNA <213> Artificial Sequence <400> 66 caggcgagtgcagtga 17
Claims
1. A combined detection system of non-invasive imaging screening and minimally invasive sampling nucleic acid typing, It is characterized in that The system includes a non-invasive telecentric imaging health analysis system and a minimally invasive sampling multi-channel nucleic acid amplification parallel detection system; The non-invasive telecentric imaging health analysis system is configured to capture local surface images of subjects, and perform machine learning and cluster analysis on the captured local surface images of subjects to obtain subjects with abnormal changes in image features. The non-invasive telecentric imaging health analysis system includes a pupil focusing aperture angle amplifier, a focusing illumination light source, a side illumination light source, an imaging optical lens, an automatic focusing control unit, a CCD detector, and a signal analysis and processing unit; the exit diaphragm of the pupil focusing aperture angle amplifier coincides with the pupil position, and is configured to converge the light emitted by the focusing illumination light source into the pupil at a wide angle of oblique incidence to illuminate the eyeball; the side The surface illumination light source adopts a dark field illumination method with oblique incidence from all sides and backlighting, and realizes wide-field imaging of the local surface through the imaging optical lens; the automatic focusing control unit is used to control the movement of the imaging optical lens to ensure clear imaging of the CCD detector; the signal analysis and processing unit is used to perform machine learning and cluster analysis on the photographed subject images to obtain subjects with abnormal changes in image features, wherein the subject local surface image is an image of the subject's body part, and the subject's body part image includes one or more combinations of the sclera, iris, fundus, vitreous body, retina, face, tongue, palm and human skin of the eye; The minimally invasive sampling multi-channel nucleic acid amplification parallel detection system is configured to perform minimally invasive sampling on subjects with abnormal changes in image features, and to perform detection and analysis of medical molecular typing indicators or physiological and pathological change indicators on the minimally invasive sampling to obtain specific gene detection results of the typing indicators or physiological and pathological change indicators; wherein: Specific genes of typing indicators or physiological and pathological change indicators, including one or more of CEA, CYFRA21-1, SCCA, NSE, ProGRP, CA125, EGFR, IDH1, TTF-1, SYN and CD56; The primers corresponding to CEA are shown in SEQ ID NO: 1 to SEQ ID NO: 6; The primers corresponding to CYFRA21-1 are shown in SEQ ID NO: 7 to SEQ ID NO: 12; The primers corresponding to SCCA are shown in SEQ ID NOs. 13 to 18; The primers corresponding to NSE are shown in SEQ ID NOs. 19 to 24; The primers corresponding to ProGRP are shown in SEQ ID NOs. 25 to 30; The primers corresponding to CA125 are shown in SEQ ID NOs. 31 to 36; The primers corresponding to EGFR are shown as the molecules from sequence 37 to sequence 42; The primers corresponding to IDH1 are shown in SEQ ID NOs. 43 to 48; The primers corresponding to TTF-1 are the molecules shown in sequences 49 to 54; The primers corresponding to SYN are shown in SEQ ID NOs: 55 to 60; The primers corresponding to CD56 are shown as molecules from sequence 61 to sequence 66.
2. The combined detection system of non-invasive imaging screening and minimally invasive sampling nucleic acid typing according to claim 1, It is characterized in that The process of performing machine learning and cluster analysis on the captured images of subjects to obtain subjects with abnormal changes in image features includes: Perform data cleaning and enhancement; Perform image processing or deep convolutional neural network segmentation of regions of interest on images for data cleaning and enhancement; Extract features of regions of interest through machine learning and deep convolutional neural networks; Input machine learning or deep learning classifiers to judge the characteristics of the region of interest, train a big data-driven detection model, and perform screening for abnormal physiological and pathological changes in the human body and predict and warn of disease risks.
3. The combined detection system of non-invasive imaging screening and minimally invasive sampling nucleic acid typing according to claim 1, It is characterized in that The captured images of the subjects are subjected to machine learning and cluster analysis to obtain subjects with abnormal changes in image features. The process includes: Perform data cleaning and enhancement; Segment the region of interest on the cleaned and enhanced images using image processing or deep convolutional neural networks; Analyze the image region of interest through machine learning algorithms and / or deep neural networks, and output the analysis results; Construct a database consisting of images of different groups, and establish a big data-driven detection model to screen abnormal physiological and pathological changes in the human body and predict and warn of disease risks.
4. The combined detection system of non-invasive imaging screening and minimally invasive sampling nucleic acid typing according to any one of claims 1 to 3, It is characterized in that The minimally invasive sampling multi-channel nucleic acid amplification parallel detection system includes a fully integrated sample preparation nucleic acid analysis unit, a heating temperature control unit, a motion control unit, an optical detection unit, and a signal processing and analysis unit; The fully integrated sample preparation nucleic acid analysis unit is used to perform nucleic acid reactions on samples, and includes a pretreatment area, a nucleic acid purification area, a sample distribution area, and an amplification reaction area; The heating and temperature control unit is used to heat and control the temperature of the pretreatment area and the amplification reaction area; The motion control unit is used to control the motion of the pretreatment area, the sample distribution area, and the amplification reaction area; The optical detection unit is used to perform real-time fluorescence signal detection on the parallel reaction unit; The signal processing and analysis unit is used to perform artificial intelligence, deep learning and / or machine learning analysis on the detected real-time fluorescence signal to obtain specific gene detection results of typing indicators or physiological and pathological change indicators of minimally invasive sampling.
5. The combined detection system of non-invasive imaging screening and minimally invasive sampling nucleic acid typing according to claim 4, It is characterized in that The sample distribution area and the amplification reaction area each include more than two parallel reaction units, and the areas are connected by micro-pipes and micro-valves to achieve orderly flow and graded isolation of fluids between the areas; the micro-valves realize flow path on-off control and prevent cross contamination; the pretreatment area is used to crush the sample and crack the nucleic acid; the nucleic acid purification area is used to purify the mixed sample of the cracked nucleic acid and separate the nucleic acid sample to be detected; the sample distribution area is used to mix the separated and purified nucleic acid sample with the amplification reaction reagent, and divide the mixture into multiple independent areas of equal volume for storage according to the number of detection indicators; The amplification reaction zone is used to embed and fix specific gene molecular primer probes corresponding to different detection indicators at the bottom of the parallel reaction unit in the amplification reaction zone, and distribute the mixtures stored in the multiple independent areas with the same volume to the parallel reaction units in the amplification reaction zone.
6. The combined detection system of non-invasive imaging screening and minimally invasive sampling nucleic acid typing according to claim 5, It is characterized in that The process of detecting and analyzing medical molecular typing indicators or physiological and pathological change indicators for minimally invasive sampling and obtaining specific gene detection results of typing indicators or physiological and pathological change indicators includes: Perform data preprocessing; The preprocessed data is used for index optimization and feature extraction through artificial intelligence, deep learning or machine learning models; Output analysis results through artificial intelligence, deep learning or machine learning classifiers; The big data-driven detection model is trained to obtain specific gene detection of typing indicators or physiological and pathological change indicators.
7. The combined detection system of non-invasive imaging screening and minimally invasive sampling nucleic acid typing according to any one of claims 1 to 3, It is characterized in that It also includes a communication interface, a mobile terminal and / or a cloud server, and the specific gene detection result is sent to the mobile terminal and / or the cloud server in a wired or wireless manner through the communication interface.
