CN108960042B

CN108960042B - Echinococcus proctostermias survival rate detection method based on visual saliency and SIFT characteristics

Info

Publication number: CN108960042B
Application number: CN201810476337.8A
Authority: CN
Inventors: 吕国栋; 吕小毅; 李壮; 林仁勇; 刘辉; 库尔班尼沙·阿马洪
Original assignee: First Affiliated Hospital of Xinjiang Medical University
Current assignee: First Affiliated Hospital of Xinjiang Medical University
Priority date: 2018-05-17
Filing date: 2018-05-17
Publication date: 2021-06-08
Anticipated expiration: 2038-05-17
Also published as: CN108960042A

Abstract

The invention relates to the technical field of detection methods, in particular to a method for detecting the Echinococcus proctosigmatus survival rate with visual significance and SIFT characteristics; the identification method based on the visual saliency and the SIFT features calculates the local image, only considers the suspected metacercaria area obtained by the saliency map, saves a large amount of time from SIFT feature extraction, and adopts a k-means algorithm to cluster the SIFT feature vector, so that the high-dimensional feature description of the SIFT feature vector is changed, the problems of too complex calculation, too long time consumption and the like are solved, the calculation efficiency of target search by using the SIFT features is improved, the SIFT features in the salient area are more stable, the artificial errors caused by artificial counting are changed, the problems of low working efficiency and large workload caused by long artificial counting time are solved, and the identification accuracy is ensured.

Description

Echinococcus proctostermias survival rate detection method based on visual saliency and SIFT characteristics

Technical Field

The invention relates to the technical field of detection methods, in particular to a method for detecting the Echinococcus proctosigmatus survival rate with visual saliency and SIFT characteristics.

Background

Echinococcosis is a serious parasitic disease caused by larvae of echinococcus granulosus and echinococcus multilocularis, which affects both humans and animals, and is the most common in areas where animal husbandry is ubiquitous. China belongs to one of the countries with high incidence of echinococcosis, including the autonomous region of Uygur autonomous region in Xinjiang, and parts of the China still have a high prevalence trend. At present, in the prevention and treatment process of echinococcosis, standard efficacy evaluation is urgently needed for echinococcosis drug treatment, and further improvement is needed for efficacy evaluation and reasonable utilization of echinococcosis, wherein the determination of whether echinococcus dies is crucial to the research and development of efficacy evaluation and reasonable utilization of novel echinococcosis drugs and echinococcosis drugs. The eosin dye exclusion method is a common method for judging the death of echinococcus granulosus metacercaria in vitro, has the advantages of simplicity, economy, rapidness and the like, and is commonly used in the field of in vitro drug screening and novel echinococcus hepatica drug research and development. The staining solution used in the eosin dye-repellent method is eosin staining solution. Using the dye exclusion principle, i.e. the associated damage-based cell membrane, the membrane on non-living (dead) cells, allows non-membrane-permeable dyes to enter the membrane for staining; the cell membrane of the living cell can resist the entry of dye, so that the dye-repellent phenomenon is generated, and the living cell is not colored. Survival was calculated under microscope by counting the number of viable protozoan larvae on the dead count scale after video and pictures were derived by eosin exclusion. Although the eosin dye-repellent method has the advantages of simplicity, rapidness, economy and the like, certain disadvantages exist. Such as: the manual counting has human errors and the manual counting time is longer. Therefore, it is necessary to develop an image of the intelligent recognition method of eosin dye exclusion.

Although no report is found in the current eosin dye-repellent method image intelligent recognition algorithm, some algorithms exist in the field of conventional insect body recognition, for example, Rema M. The high source and the like use a support vector machine as a classifier, research is carried out on the identification of schistosome eggs in microscopic images, and Zhang Zhenglong proposes an improved K nearest neighbor classifier to carry out identification and classification on the schistosome eggs. In recent years, many scholars at home and abroad make a lot of researches on a parasite identification technology in the microscopic medicine, but the existing method has the problems of low efficiency, large workload and human errors caused by that too many parts need manual intervention in the identification processing process, the provided image characteristics cannot reflect the image characteristics, the characteristic value ranges of various identification objects are overlapped more, and the like.

Disclosure of Invention

The invention provides a method for detecting the survival rate of echinococcus proctosomum with visual saliency and SIFT characteristics, which overcomes the defects of the prior art and can effectively solve the problems that the prior method has too many parts needing manual intervention in the identification processing process, the provided image characteristics cannot reflect the image characteristics, the characteristic value ranges of various identification objects are overlapped more, the efficiency is low, the workload is large, and human errors exist.

The technical scheme of the invention is realized by the following measures: a method for detecting the survival rate of Echinococcus proctosigmatus with visual saliency and SIFT characteristics comprises the following steps: firstly, carrying out eosin dye exclusion treatment or trypan blue dyeing treatment on 20 to 100 Echinococcus crude metacercaria to be detected, and then photographing to obtain an image of the Echinococcus crude metacercaria to be detected; secondly, extracting a color and brightness insect body image saliency map from the image processed by the echinococcus prolercaria to be detected; thirdly, carrying out linear weighting on the insect body image saliency map of the color and the brightness to generate a total saliency map; fourthly, extracting a salient region of the total salient map, finding a central point of the suspected polypide in the salient region, cutting all suspected living polypide slices, marking the suspected target regions in the salient region, extracting the SIFT features of the living polypide slices through an SIFT algorithm, and generating the SIFT feature vector of the corresponding suspicious region; and fifthly, comparing and identifying the sift characteristic vector of the suspicious region with the clustered sift characteristic vector in the image data graph of the echinococcus protocephalis live larvae, if the result of comparison and identification is a live larvae, marking as a live larvae target, if the result of comparison and identification is not a live larvae, canceling the marking, and finally reducing the marking result to the image for processing the echinococcus protocephalis to be detected and counting to obtain the survival rate of the echinococcus protocephalis.

The following is further optimization or/and improvement of the technical scheme of the invention:

the image data map of the echinococcus metacercaria living body is obtained according to the following steps: step one, taking 20 to 100 echinococcus protocoicaria, and carrying out eosin dye exclusion treatment or trypan blue dye treatment on the echinococcus protocoicaria, and taking a picture after treatment to obtain an echinococcus protocoicaria treatment image; secondly, selecting live parasite images in 50 to 70 echinococcus metacercaria processing images and background images corresponding to the live parasite images to establish a database; thirdly, extracting SIFT characteristic vectors of live insect body images and background images in a database through an SIFT algorithm; and fourthly, clustering the sift characteristic vectors through a k-means algorithm, and then putting the clustered sift characteristic vectors into an svm classifier to obtain an image data map of the echinococcus prototheca larvae live larvae.

The echinococcus metacercaria survival rate detection method based on visual significance and SIFT features identifies the metacercaria image to be detected, and the live metacercaria basically falls in the marked image area; the identification method based on the visual saliency and the SIFT features is characterized in that image local calculation is carried out, only a suspected metacercaria area obtained from a saliency map is considered, a large amount of time is saved from SIFT feature extraction, and in terms of obtaining of SIFT feature vectors of sample images, the k-means algorithm is adopted for clustering, so that high-dimensional feature description of the SIFT feature vectors is changed, the problems of too complex calculation, too long time consumption and the like are solved, the calculation efficiency of target search by using the SIFT features is improved, the SIFT features in the salient area are more stable, human errors caused by manual counting are changed, the problems of low working efficiency and large workload caused by long manual counting time are solved, and the identification accuracy is ensured.

Drawings

FIG. 1 is an image of Echinococcus procymphatis.

FIG. 2 is a significant diagram of the original ITTI obtained from the image of Echinococcus echinococcus metacercaria by the prior art.

FIG. 3 is an improved ITTI significance map obtained from images of Echinococcus proctosigmatus processed by the present invention.

FIG. 4 is a diagram showing the identification result of the image of Echinococcus proctosigmatus after the detection of the present invention.

FIG. 5 is a flow chart of the method for detecting Echinococcus proctosigmatus survival rate with visual saliency and SIFT characteristics of the present invention.

Detailed Description

The present invention is not limited by the following examples, and specific embodiments may be determined according to the technical solutions and practical situations of the present invention.

Example 1, the method for detecting Echinococcus proctosigmatus survival rate with visual significance and SIFT characteristics comprises the following steps: firstly, carrying out eosin dye exclusion treatment or trypan blue dyeing treatment on 20 to 100 Echinococcus crude metacercaria to be detected, and then photographing to obtain an image of the Echinococcus crude metacercaria to be detected; secondly, extracting a color and brightness insect body image saliency map from the image processed by the echinococcus prolercaria to be detected; thirdly, carrying out linear weighting on the insect body image saliency map of the color and the brightness to generate a total saliency map; fourthly, extracting a salient region of the total salient map, finding a central point of the suspected polypide in the salient region, cutting all suspected living polypide slices, marking the suspected target regions in the salient region, extracting the SIFT features of the living polypide slices through an SIFT algorithm, and generating the SIFT feature vector of the corresponding suspicious region; and fifthly, comparing and identifying the sift characteristic vector of the suspicious region with the clustered sift characteristic vector in the image data graph of the echinococcus protocephalis live larvae, if the result of comparison and identification is a live larvae, marking as a live larvae target, if the result of comparison and identification is not a live larvae, canceling the marking, and finally reducing the marking result to the image for processing the echinococcus protocephalis to be detected and counting to obtain the survival rate of the echinococcus protocephalis. The operation of the method for detecting the Echinococcus proctosigmatus survival rate with the visual significance and SIFT characteristics can be carried out on computer software Matlab R2016 a.

Example 2, as an optimization of the above example, the image data map of Echinococcus proctosomum live larva is obtained by the following steps: step one, taking 20 to 100 echinococcus protocoicaria, and carrying out eosin dye exclusion treatment or trypan blue dye treatment on the echinococcus protocoicaria, and taking a picture after treatment to obtain an echinococcus protocoicaria treatment image; secondly, selecting live parasite images in 50 to 70 echinococcus metacercaria processing images and background images corresponding to the live parasite images to establish a database; thirdly, extracting SIFT characteristic vectors of live insect body images and background images in a database through an SIFT algorithm; and fourthly, clustering the sift characteristic vectors through a k-means algorithm, and then putting the clustered sift characteristic vectors into an svm classifier to obtain an image data map of the echinococcus prototheca larvae live larvae. Eosin exclusion treatment, trypan blue staining treatment, SIFT algorithm and k-means algorithm are all known and used.

Theory and algorithm (Echinococcus proctosterns are abbreviated as proctosterns in theory and algorithm)

Visual saliency

The human visual system has the capability of image understanding, recognition and processing, and the computer is used for simulating the visual system to establish a visual attention model, which is a research hotspot in the field of image processing. Human visual observation is selective, does not analyze and think about everything that is seen by the line of sight, and the brain is concerned about only those things that he is concerned about. In short, the brain processes only obvious and specific things, i.e., salient things. The computer-simulated visual attention model focuses on rapidly locating salient regions. In a broad sense, an image contains a variety of information that can be perceived by humans, such as color, texture, brightness, and the like. However, in general, not all information is of interest, and only some partial region of information is of interest. Therefore, locating the region of interest and extracting it from the image becomes a very necessary step in image processing; currently, there are 4 major models for the study of visual significance: a spectrum residual model, a Hu-Rajan-Chia model, a Stentiford model and an ITTI visual attention model; the Itti model simulates human perception capability according to the characteristics between the target and the background and the difference between the contrast, and extracts the region of interest.

Improved ITTI model

According to the traditional method, because the medical microscopic metacercaria image is analyzed and treated by an eosin dye rejection method, the color of the dyed image is relatively clear, and the ITTI visual model mainly utilizes color characteristics, direction characteristics and brightness characteristics, the ITTI visual model is relatively suitable for selection; aiming at the problems that the accuracy of extracting the image saliency region by using the traditional ITTI visual model is not high and the complete region of interest is difficult to extract, the invention improves the saliency feature combination mode in the traditional ITTI visual model according to the fact that the sensitivity of human eyes to different saliency features is different, basically does not have too large effect on the direction of original metacercaria in the original metacercaria image, ignores the direction saliency map, and linearly weights the color and brightness saliency map to generate a total saliency map S; after improvement, the obtained total significant map S can better perform coarse positioning on the metacercaria.

Algorithm

The SIFT algorithm searches extreme points in a scale space, extracts position, scale and rotation invariants, keeps invariants to rotation, scale scaling and brightness change, and also can keep stability to affine transformation and noise to a certain extent; the method mainly comprises the following four steps: detecting extreme points in a scale space, filtering and accurately positioning characteristic points, distributing the directions of the characteristic points and describing key points; through the three steps, complete characteristic points are obtained, and each characteristic point comprises three information of a coordinate position, a scale and a direction. Through the above steps, for each key point, three pieces of information are possessed: location, scale, and orientation. Next, a descriptor is built for each keypoint, which is described by a set of vectors so that it does not change with various changes, such as illumination changes, view angle changes, and so on. This descriptor contains not only the keypoints but also the pixels around the keypoints that contribute to them, and should be highly unique in order to increase the probability of correct matching of feature points. In a specific operation, the keypoint and surrounding points are generally compared, each extremum point is taken as a central point, 16 adjacent sub-regions are selected, and the regions are sorted according to coordinate bits, so that a 128-dimensional feature descriptor for generating the extremum point can be constructed because each region contains 8 directions. And obtaining the whole scale-invariant feature descriptors of the target image. In order to further improve the stability of the feature descriptors, normalization processing is required.

Algorithm overview for identifying metacercaria based on visual significance and SIFT

According to the invention, through an improved ITTI model and a SIFT algorithm, firstly, a significant map of an image of the coenuruses obtained by an eosin dye exclusion method is extracted, the central points of the suspected live coenuruses are obtained, and all the images of the suspected live coenuruses are cut at the positions of the central points according to the size of the live coenuruses to be detected. Through an SIFT algorithm, SIFT features are extracted from a known live metacercaria sample image to obtain an SIFT feature vector, the vector is clustered by using a k-means clustering method to generate a k-dimensional feature vector, and the feature vector is subjected to SVM training to obtain an SVM classifier; and extracting SIFT characteristics from the suspected live metacercaria images, putting the SIFT characteristics into a trained SVM classifier, carrying out 'yes' and 'no' classification, returning classification results to the original image, and marking to obtain a final detection result.

The verification test of the above embodiment of the present invention (Echinococcus echinococcus metacercaria is abbreviated as metacercaria in the verification test)

To verify the effectiveness of the algorithm presented herein, experiments were performed under Matlab R2016a using micrographs of metacercaria treated by the eosin dye exclusion method, live metacercaria sample pictures being different target and background images obtained from different metacercaria pictures. Firstly, establishing a 60-frame live metacercaria image library and a 60-frame background image library to generate an SVM classifier, and then identifying metacercaria images to be detected by the echinococcus metacercaria survival rate detection method with visual significance and SIFT characteristics; extracting an image saliency map and marking suspected live metacercaria; extracting SIFT characteristics of the suspected live metacercaria, putting the extracted SIFT characteristics into a trained SVM classifier for classification, returning the classification result to the original metacercaria image and marking the image; FIG. 1 is an image of Echinococcus procymphatis; FIG. 2 is a primary ITTI saliency map obtained by current method from Echinococcus echinococcus metacercaria image; FIG. 3 is an improved ITTI saliency map obtained from Echinococcus echinococcus metacercaria images treated by the present invention; FIG. 4 is a diagram showing the identification result of an image of Echinococcus proctosigmatus after detection by the present invention; FIG. 5 is a flow chart of the method for detecting Echinococcus proctosigmatus survival rate with visual saliency and SIFT characteristics of the present invention. As can be seen after the comparison between the attached drawings 2 and 3, compared with the original ITTI saliency map, the improved ITTI saliency map omits the direction of the image, but does not affect the recognition effect, but makes the connected domain of the living insect body clearer and more effectively helps to roughly recognize the living insect body.

Test 1, taking 23 complete original coenuruses in a microscopic picture of the original coenuruses to be detected, which are treated by an eosin dye exclusion method or trypan blue dye treatment, wherein one of the original coenuruses is a dead original coenurus, identifying the image of the original coenuruses to be detected by the method for detecting the survival rate of the original coenuruses of the echinococcus taenioides with the visual significance and SIFT characteristics, wherein the identification result is shown in table 1, the improved ITTI significant map in the table 1 is a total significant map generated by linearly weighting the significant maps of the color and brightness of the image of the echinococcus taenis with the visual significance and SIFT characteristics in the third step of the method for detecting the survival rate of the original coenuruses of the echinococcus taenioides with the visual significance and SIFT characteristics, and the original ITTI significant map in the table 1 is a total significant map generated by linearly weighting the significant maps of the color, direction and brightness of the image by the existing method; as can be seen from Table 1, the results show that 21 live metacercaria are identified in total, the identification rate reaches 95.4%, and the experimental requirements are met; the identification rate of the Echinococcus proctosigmatus metacercaria survival rate to be detected by the Echinococcus proctosigmatus survival rate detection method with the visual saliency and SIFT characteristics reaches 95.4 percent, which is higher than that of the original ITTI saliency map, and compared with the original ITTI saliency map, the improved ITTI saliency map can show the characteristics of echinococcosis body images better, and can accurately mark all suspicious targets; thus, the improved ITTI saliency maps of the present invention are superior to the results obtained from the original ITTI saliency maps.

Experiment 2, taking a micropicture a of the original metacercaria to be detected, a micropicture b of the original metacercaria to be detected and a micropicture c of the original metacercaria to be detected, which are processed by an eosin dye exclusion method or trypan blue dye, identifying the image of the original metacercaria to be detected by the echinococcus protothecalis survival rate detection method with visual significance and SIFT characteristics, wherein the experimental results of the influence of the number of live bodies in the micropicture of the original metacercaria to be detected on the experimental results are shown in Table 2, and as can be seen from the table 2, the image of the original metacercaria to be detected is identified by the echinococcus protothecium survival rate detection method with visual significance and SIFT characteristics, the identification rate is high and is over 90 percent, the live bodies in the image of the metacercaria under a microscope can be basically identified, thereby providing powerful evidence for drug effect detection and greatly reducing the workload of detection personnel, the problems of low efficiency, large workload and easy error caused by the existing manual counting are avoided.

In conclusion, the echinococcus metacercaria survival rate detection method with visual significance and SIFT characteristics identifies the metacercaria images to be detected, and the live metacercaria basically falls in the marked image area; the identification method based on the visual saliency and the SIFT features is characterized in that image local calculation is carried out, only a suspected metacercaria area obtained from a saliency map is considered, a large amount of time is saved from SIFT feature extraction, and in terms of obtaining of SIFT feature vectors of sample images, the k-means algorithm is adopted for clustering, so that high-dimensional feature description of the SIFT feature vectors is changed, the problems of too complex calculation, too long time consumption and the like are solved, the calculation efficiency of target search by using the SIFT features is improved, the SIFT features in the salient area are more stable, human errors caused by manual counting are changed, the problems of low working efficiency and large workload caused by long manual counting time are solved, and the identification accuracy is ensured.

The technical characteristics form an embodiment of the invention, which has strong adaptability and implementation effect, and unnecessary technical characteristics can be increased or decreased according to actual needs to meet the requirements of different situations.

Claims

1. A method for detecting the survival rate of Echinococcus proctosigmatus with visual saliency and SIFT characteristics is characterized by comprising the following steps: firstly, carrying out eosin dye exclusion treatment or trypan blue dyeing treatment on 20 to 100 Echinococcus crude metacercaria to be detected, and then photographing to obtain an image of the Echinococcus crude metacercaria to be detected; secondly, extracting a color and brightness insect body image saliency map from the image processed by the echinococcus prolercaria to be detected; thirdly, carrying out linear weighting on the insect body image saliency map of the color and the brightness to generate a total saliency map; fourthly, extracting a salient region of the total salient map, finding a central point of the suspected polypide in the salient region, cutting all suspected living polypide slices, marking the suspected target regions in the salient region, extracting the SIFT features of the living polypide slices through an SIFT algorithm, and generating the SIFT feature vector of the corresponding suspicious region; and fifthly, comparing and identifying the sift characteristic vector of the suspicious region with the clustered sift characteristic vector in the image data graph of the echinococcus protocephalis live larvae, if the result of comparison and identification is a live larvae, marking as a live larvae target, if the result of comparison and identification is not a live larvae, canceling the marking, and finally reducing the marking result to the image for processing the echinococcus protocephalis to be detected and counting to obtain the survival rate of the echinococcus protocephalis.

2. The method for detecting Echinococcus proctosigmatus survival rate with visual significance and SIFT features according to claim 1, wherein the image data map of Echinococcus proctosigmatus live larvae is obtained by the following steps: step one, taking 20 to 100 echinococcus protocoicaria, and carrying out eosin dye exclusion treatment or trypan blue dye treatment on the echinococcus protocoicaria, and taking a picture after treatment to obtain an echinococcus protocoicaria treatment image; secondly, selecting live parasite images in 50 to 70 echinococcus metacercaria processing images and background images corresponding to the live parasite images to establish a database; thirdly, extracting SIFT characteristic vectors of live insect body images and background images in a database through an SIFT algorithm; and fourthly, clustering the sift characteristic vectors through a k-means algorithm, and then putting the clustered sift characteristic vectors into an svm classifier to obtain an image data map of the echinococcus prototheca larvae live larvae.