KR101223598B1 - A System for Diagnosing Pancreatic Intraepithelial Neoplasia - Google Patents

Abstract

The present invention relates to a pancreatic epithelial tumor diagnosis system. The present invention identifies a lumen using a region growth method from a given tissue image, identifies a nucleus near the lumen using a color threshold technique, and provides for atypical nuclear alignment and General morphological features for measuring the lumen boundary structure and morphological features specific to pancreatic epithelial tumors such as papillary and polarity are extracted, and the extracted features and the support vector machine (SVM) Classify images by class (eg normal tissue, PanIN-1B, PanIN-2, PanIN-3).

Description

Pancreatic epithelial tumor diagnosis system {A System for Diagnosing Pancreatic Intraepithelial Neoplasia}

The present invention relates to a pancreatic epithelial tumor diagnostic system that automatically classifies intra pancreatic epithelial tumors using morphological features specific to intra pancreatic epithelial tumors.

Pancreatic cancer is the fourth leading cause of death from cancer in the United States. About 95% of exocrine systemic cancers are adenocarcinomas, most of which occur in the pancreatic duct. Most cancers of the pancreas are diagnosed too late to operate, and surgery is impossible.

Today, five-year survival rates for patients diagnosed with pancreatic adenocarcinoma are less than 4%, despite the introduction of new chemotherapy and significant advances in the understanding of molecular genetics related to surgery, pathological classification of pancreatic tumor formation, and invasive pancreatic cancer progression. to be.

Early diagnosis of non-invasive precancerous lesions (lesions), such as Pancreatic Intraepithelial Neoplasias (PanINs), is the best way to increase the survival rate of patients with pancreatic cancer. Visual analysis of patient tissue samples obtained by biopsy is the most reliable method.

Unfortunately, it is a difficult task for pathologists to analyze and interpret histopathological images under a microscope. Computer aided diagnosis (CAD) helps doctors diagnose difficult cases and reduce false positives.

Recently, many diagnostic methods for detecting and classifying cancer in various lesions such as the prostate, breast, and colon have been proposed.

This diagnostic method uses SVM with tissue features extracted through Gauss Markov Random Field, correlation function and relative entropy to diagnose prostate cancer. Method (Metehan Makinaci, “Support Vector Machine Approach for Classification of Cancerous Prostate Regions,” In Proc. Of world academy of science engineering and technology, 2005, 166-169), quantitatively identifying tissue features and classification for diagnosing breast cancer tissue. Graph theoretical techniques (Cagatay Bilgin, Cigdem Demir, Chandandeep Nagi, Bulent Yener, "Cell-Graph Mining for Breast Tissue Modeling and Classification," IEEE EMBS, 2007, 5311-5314), 2 How to classify colon tissue samples using dimensional principal component analysis (PCA) and SVM (Khalid Masood and Nasir Rajpoot, “Classification of Colon Bio psy Samples by Spatial Analysis of a Single Spectral Band from its Hyperspectral Cube, "Proceedings Medical Image Understanding and Analysis, 2007, 42-48), automatic classification and nucleation of prostate and breast tissue using low-level, high-level and specific area information Schemas (S. Naik, S. Doyle, et al., “Automated Gland and Nuclei Segmentation for Grading of Prostate and Breast Cancer Histopathology,” 5th IEEE International Symposium on In Biomedical Imaging, 2008, 284-287) are proposed. . In addition, various methods for cancer tissue classification have been studied.

The present invention is to provide a pancreatic epithelial tumor diagnostic system that separates regions of interest such as lumen and nucleus from tissue images using region growing technique and maximum entropy thresholding technique.

In addition, the present invention extracts the morphological features specific to the pancreatic epithelial tumor from the region extracted from the tissue image, and using the extracted features and SVM to classify (dividing) according to the grade (grade) of the tumor in the pancreatic epithelium To provide an intraepithelial tumor diagnostic system.

In addition, the present invention extracts the morphological features specific to the pancreatic epithelial tumor from the region extracted from the tissue image, and using the extracted features and SVM to classify (dividing) according to the grade (grade) of the tumor in the pancreatic epithelium To provide an intraepithelial tumor diagnostic system and method.

delete

In addition, the pancreatic epithelial tumor diagnostic method according to an embodiment of the present invention comprises the steps of receiving a tissue image, separating the lumen and nucleus from the received tissue image, and the morphological characteristics of the separated lumen and nucleus And classifying the tumor within the pancreatic epithelium using the extracted features and a predetermined classifier according to the progress level.

In accordance with at least one embodiment of the present invention configured as described above, the present invention uses a region growing technique and a maximum entropy thresholding technique to detect regions of interest such as lumen and nucleus in tissue images. Can be separated.

In addition, the present invention can extract the morphological features specific to the pancreatic epithelial tumor from the region extracted from the tissue image, and can be classified according to the grade of the tumor in the pancreatic epithelium using the extracted features and SVM. As described above, the present invention can further classify tumors in the pancreatic epithelium, thereby increasing classification accuracy.

1 is a diagram showing the progression of pancreatic cancer associated with the present invention.
Figure 2 is a block diagram showing the structure of the pancreatic epithelial tumor diagnostic system associated with one embodiment of the present invention.
Figure 3 illustrates the morphological differences between normal tissue and pancreatic epithelial tumors associated with the present invention.
4 is a diagram illustrating a process of detecting a convex hull of a lumen boundary in accordance with the present invention.
5 is a diagram showing the degree of papillary phase of the tube among morphological features specific to the pancreatic epithelial tumors related to the present invention.
6 shows the polarity of pancreatic epithelial tumors associated with the present invention and the polarity of each tumor as defined by the average distance between the lumen boundary and the nuclear boundary.

Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings.

The present invention detects pancreatic intratrapithelial neoplasias (PanINs) from pathological images, and is intended to facilitate pancreatic cancer diagnosis.

Pancreatic epithelial tumors are epithelial tumors that occur in the pancreas which are fine papillary or flat and non-invasive. Pancreatic epithelial tumors are columnar or cubic cells that vary in the degree of cytological or structural atypical (dysplasia) and the amount of mucus. Pancreatic epithelial tumors generally comprise tubes less than 5 mm in diameter.

In addition, pancreatic epithelial tumors can be morphologically classified into three classes.

Pancreatic epithelial tumor 1 (PanIN-1) has only mild atypical, pancreatic epithelial tumor 2 (PanIN-2) has moderate cytological and structural atypical, and pancreatic epithelial tumor 3 (PanIN-3) is clearly atypical Has formation.

Table 1 is the nomenclature for pancreatic epithelial tumors, and FIG. 1 shows the progression of pancreatic cancer.

normal
(Normal) Normal ducts and the epithelium of the ducts are cubic or low columnar epithelium. Mucous cytoplasm, nucleus colonization, and atypical symptoms do not appear. Squamous epithelium
(Squamous metaplasia) Normal cubic tubular epithelium is the process of transformation into stratified squamous or false stratified transitional epithelium without amorphism. Panin-1a These are flat epithelial lesions consisting of tall columnar cells with a nucleus and abundant nucleus mucin located at the base. These nuclei are small and round (circles, ellipses). When the nucleus is elliptical, the nucleus is located perpendicular to the membrane. Histological overlap can be significant between atypical flat tumor lesions and non-tumor flat hyperplasia lesions. That is, pancreatic epithelial tumor 1A is not significantly different from pathologically normal cells. Panin-1b Very similar to PanIN-1A, but features such as papillary phase, micropapillary phase and false stratified structure at base. PanIN-2 In the second stage (grade), the degree of papillary is severe and atypical in the nuclei. Atypical types include the loss of polarity and indiscriminate positioning of the nucleus, the clustering of the nuclei, or the increase in nucleus size, stratification, and hyperpigmentation. However, these symptoms appear in fewer than three levels.
Mitosis rarely occurs, but does not appear atypical when it appears. Necrosis and body-like structures are seen, and noticeable atypical symptoms are not usually seen. When such a phenomenon occurs, it can be seen as three steps. PanIN-3 Structurally, the lesions are papillary or micropapillary, most not flat.
Small epithelial cells invading the body shape and lumen appear budding off, and atypical necrosis can be seen in three stages.
Phase 3 has the following characteristics: The nucleus loses its polarity and is located, corrosive goblet cells (the goblet cells are composed of the mucous cytoplasm directed to the basement membrane and the nucleus directed to the lumen), and mitosis is atypical. And most nucleolus are small. Stage 3 is similar to adenocarcinoma but does not show infiltration into the basement membrane.

PanIN-1A and 1B are also found in normal people, but higher grade pancreatic epithelial tumors are most commonly found with invasive cancer.

Therefore, in the present invention, the main aim is to diagnose a higher level pancreatic epithelial tumor. The difference between Tier 2 and Tier 3 is as follows:

The size of the third stage nuclei is larger than that of the second stage nuclei, and in the third stage, epithelial cells sometimes infiltrate the lumen and often lose their polarity. Also, in the third stage, body-shaped structures are found.

The criteria for distinguishing steps 2 and 3 are similar to each other, making it difficult to distinguish them.

In order to classify these, the present invention defines morphological features specific to pancreatic epithelial tumors.

Figure 2 is a block diagram showing the structure of the pancreatic epithelial tumor diagnostic system related to an embodiment of the present invention.

Referring to FIG. 2, the pancreatic epithelial tumor diagnosis system (hereinafter, referred to as a system) includes a tissue image input unit 110, a preprocessor 120, a feature extractor 130, a diagnosis unit 140, and a controller 150. .

The tissue image input unit 110 obtains a tissue image from the outside and transmits the tissue image to the preprocessor 120. For example, the tissue image input unit 110 acquires a tissue image by using a medical imaging apparatus such as computerized tomography (CT). The tissue image input unit 110 is connected to the medical imaging apparatus through an interface or a wired or wireless network.

In addition, the tissue image input unit 110 may access the tissue image from the memory and output the tissue image to the preprocessor 120. Here, the memory may be a portable removable memory or a memory provided in the system.

The preprocessor 120 detects a collection of tubes and separates lumens and nuclei from the tubes. The preprocessor 120 includes a lumen separator 121 for detecting (separating) the lumen from a tissue image and a nuclear separator 122 for detecting (separating) a nucleus located near the detected lumen.

The lumen separating unit 121 converts an RGB color tissue image stained with hematoxylin eosin to a gray image to separate the lumen through the gray scaling unit 121a. The lumen separating unit 121 generates a binary image by applying maximum entropy thresholding to the converted gray image using the threshold processing unit 121b.

The lumen boundary detector 121c of the lumen separating unit 121 detects the tube and lumen boundary in the binary image by using a region growing technique. In other words, the lumen boundary detection unit 121c identifies a set of boundary points of the lumen located at the center of the tube using a region growth method based on intensity.

The nuclear separator 122 includes a color threshold processor 122a and a nuclear boundary detector 122b. The color threshold processor 122a extracts a value of the RGB range specific to the nucleus from the tissue image. The nuclear boundary detector 122b detects the boundary of the nucleus by separating pixels corresponding to the nucleus based on the extracted value.

The feature extractor 130 extracts morphological features and general morphological features specific to the pancreatic epithelial tumor from the separated (detected) lumen, cytoplasm and nucleus. Here, the specialized morphological features include the papillary ratio of the tube and the polarity distance of the nucleus. General morphological features are then extracted to measure the atypical and lumen boundary structures of the nuclear array. The specialized morphological features and / or general morphological features are used to train the classifier.

The diagnosis unit 140 uses the extracted morphological features output from the feature extractor 130 and four grades (steps) for each tissue image by using a predetermined support vector machine (SVM). (Eg normal, PanIN-1B, PanIN-2, PanIN-3). The predetermined classifier may be a support vector machine (SVM). The SVM is learned about morphological features specific to tumors in the pancreatic epithelium. In other words, the diagnosis unit 140 may classify the tumor in the pancreatic epithelium according to the grade.

The controller 150 controls each of the above components to control the overall operation of the diagnostic system. Programs for operating the controller 150 and data input / output are stored in a memory (not shown). The memory may store tissue images.

In addition to the above components, the system may include a user input unit (not shown) for generating input data for controlling the operation of the system according to the manipulation of a user (pathologist or doctor) and a display for displaying various data generated according to system operation. It may further include.

Figure 3 is a diagram showing the morphological differences between normal tissue and pancreatic epithelial tumors associated with the present invention.

The present invention defines new morphological features specific to tumors in the pancreatic epithelium.

The main difference between normal tissue and pancreatic epithelial tumors is whether the papillary papillary is observed. The biggest difference between PanIN-2 and PanIN-3 is the presence of nuclear polarity.

Here, C is the number of pixels included in the convex hull region with respect to the lumen boundary, and L is the number of pixels included in the lumen region.

Also, the number of papillary phases is the number of contacts between the lumen boundary and its convex hull.

4 is a diagram illustrating a process of detecting convex hull of the lumen boundary related to the present invention, and FIG. 5 is a diagram illustrating the papillary degree of the tube among morphological features specific to the pancreatic epithelial tumor related to the present invention.

When the system receives a tissue image as shown in FIG. 4A as an input, the system detects a lumen boundary from the tissue image (b). If the lumen boundary is detected, the system detects a convex hull surrounding the outer shell of the detected lumen boundary (c). The feature extractor 130 may calculate a papillary degree of the tube based on the detected lumen boundary and convex hull. The papillary degree corresponds to the ratio of the convex hull region (region in red line) of FIG. 5 to the region (yellow region) of the convex hull region except the lumen region.

In addition, the present invention defines a morphological feature called polarity degree in order to evaluate the degree of polarity loss of the nucleus.

In order to measure the polarity of the nucleus, the boundaries of the nuclei must be detected. The nuclei are separated using a color threshold technique to extract the boundaries of the nuclei. The nuclei appear purple on color tissue images stained with hematoxylin and eosin. Therefore, only nuclei can be extracted using the color information.

In the case of lower stage pancreatic epithelial tumors, the extracted nuclei are interconnected due to the polarity of the nuclei.

However, the nuclei in higher-level pancreatic epithelial tumors lose their polarity and are not connected to each other, making it difficult to extract the boundaries of the nuclei. Thus, the present invention fills the gaps between nuclei through morphological operations such as erosion, dilation, opening, closing. The boundaries of the nuclei are detected by using the region growth method.

6 is a diagram showing the polarities (d, e, f) of the pancreatic epithelial tumors (a, b, c) and the respective tumors defined as the average distance between the lumen boundary and the nuclear boundary in accordance with the present invention.

This gap is wide when the polarity of the nucleus is present, and the gap is very narrow and irregular in the case of upper stage pancreatic epithelial tumors. As shown in Figure 6, in the normal tissue, such as (a), the polarity of the nucleus exists because the gap between the lumen boundary and the nuclear boundary is wide, and in the upper stage pancreatic epithelial tumor such as (c) the polarity of the nucleus Lost, the gap between the lumen boundary and the nuclear boundary is very narrow and irregular.

In order to quantify the polarity of the nucleus, the average distance (D) from the center point (c) of the tube to the lumen boundary and the average distance from the nucleus boundary are calculated and their differences are expressed as the polarity of the nucleus.

와

는 핵과 내강 각각의 경계점 집합이다. N은 핵들의 경계를 구성하는 점의 개수이고, m은 내강의 경계를 구성하는 점들의 수이다.here,

Wow

Is the set of boundary points for each of the nucleus and the lumen. N is the number of points constituting the boundary of the nucleus, m is the number of points constituting the boundary of the lumen.

Experiments were performed with the diagnostic method proposed in the present invention.

The pancreatic biopsy sample was provided with 13 tissue images of 7 patients, and the tissue images obtained were taken by magnification of the tissue stained 40 times using the hematoxylin eosin method. In addition, three images of the received tissue images are normal tissue images, and 10 images are 10, 18, or 9 PanIN-1B images, PanIN-2 images, and PanIN-3 images, respectively.

In addition, in the present invention, a plug-in of ImageJ (http://rsbweb.nih.gov/ij/) was developed and used to extract morphological features specific to pancreatic epithelial tumors from tissue images. ImageJ is a unified platform that supports various image processing and analysis techniques. In the present invention, a plug-in for extracting morphological features specific to pancreatic epithelial tumors from tissue images is developed based on the ImageJ platform. However, the present invention is not limited thereto and a plug-in may be developed based on various platforms.

In the present invention, an SVM model is generated by extracting existing morphological features in addition to the morphological features specific to the pancreatic epithelial tumor.

In order to show the importance of the new morphological features defined in the present invention, the features were compared for each tissue type.

Table 2 shows the average values of the morphological features for each step. Table 3 shows the numerical values of the general features.

Referring to Tables 2 and 3 above, the papillary degree of the pancreatic epithelial tumor can be seen to have a greater value than the normal tissue due to the papillary phase of the tube. Therefore, the papillary degree can be used to efficiently classify normal tissues.

Panin-1b PanIN-2 PanIN-3 The number of papilla 25.25 27.7 20.78 21.11 Papilla ratio 7.0 26.3 44.17 50.33 Polarity distance 1.0 15.27 15.38 7.03

Panin-1b PanIN-2 PanIN-3 Total area of lumen 50902.25 191079.60 130599.44 161445.10 Stddev of lumen 28.59 19.91 36.77 36.37 Lumen perimeter 1240.61 2454.25 3430.56 5273.90 Circularity of lumen 0.45 0.36 0.16 0.09 Total area of nuclei 4189.25 46131.40 79932.33 81998.67 Nuclei perimeter 26.85 42.81 38.51 31.17 Circularity of nuclei 0.60 0.78 0.80 0.84

However, the papillary degrees of PanIN-2 and PanIN-3 are similar. Polarity features play an important role in classifying these two steps. In PanIN-3, the polarity of the nucleus is reduced, resulting in smaller polarity values than in other stages.

Table 4 shows the classification results for the general morphological features and the features defined in the present invention.

Task Accuracy Classification of Tissue, PanIN-1B, PanIN-2, PanIN3 Using General Morphological Features 75% Classification of Normal Tissue, PanIN-1B, PanIN-2, PanIN3 Using Papilla ratio, Polarity distance and General Morphological Features 91% Classification of Normal Tissue, PanINs Using Papilla ratio, Polarity distance and General Morphological Features 95% Classification of PanIN-2 vs. PanIN-3 Using Papilla ratio, Polarity distance and General Morphological Features 90%

Average accuracy is calculated by ten random bootstrap resampling.

The accuracy when using the features defined in the present invention showed higher accuracy than when using the general morphological features. In particular, in the case of two and three levels of classification that are difficult to diagnose, the method in the present invention shows a high accuracy of 90%. As a result, the method proposed in the present invention can clearly classify each stage of normal tissue and tumor in the pancreatic epithelium.

110: tissue image input unit 120: preprocessor
130: feature extraction unit 140: diagnostic unit
150:

Claims (9)

A tissue image input unit for providing a tissue image,
A pre-processing unit for separating the lumen and the nucleus by pre-processing the tissue image;
A feature extraction unit for extracting morphological features from the separated lumen and nucleus,
Diagnosis unit for classifying the tumor in the pancreatic epithelial epithelium using the extracted morphological features and classifier,
Pancreatic epithelial tumor diagnosis system, characterized in that it comprises a control unit for controlling each of the above components. The method of claim 1, wherein the preprocessing unit,
And lumen separation unit for separating the lumen from the tissue image,
Pancreatic epithelial tumor diagnostic system, characterized in that it comprises a nuclear separation unit for separating the nucleus from the tissue image. The method of claim 2, wherein the lumen separating portion,
A gray scaling unit converting the tissue image into a gray image;
A threshold processor for generating a binary image by performing a maximum entropy threshold on the converted gray image;
Pancreatic epithelial tumor diagnosis system, characterized in that it comprises a lumen boundary detection unit for detecting the lumen boundary by applying a region growth method to the generated binary image. The method of claim 2, wherein the nuclear separation unit,
A color threshold processor for extracting a color range specific to the nucleus from the tissue image;
Pancreatic epithelial tumor diagnosis system, characterized in that it comprises a nuclear boundary detection unit for detecting the nuclear boundary from the tissue image based on the extracted value. The method of claim 1, wherein the morphological features,
Pancreatic epithelial tumor diagnostic system, characterized in that it comprises a papillary degree of the tube and the polarity of the nucleus. The apparatus according to claim 1,
Pancreatic epithelial tumor diagnostic system, characterized in that SVM (Support Vector Machine). The method of claim 6, wherein the SVM,
Pancreatic epithelial tumor diagnostic system, characterized in that it is learned about the morphological features. The method of claim 1, wherein the diagnostic unit,
The pancreatic epithelial tumor diagnosis system, characterized in that it is classified into any one of normal tissue, PanIN (pancreatic intraepithelial neoplasia) -1B, PanIN-2, PanIN-3. delete

Images