KR20010008933A - Method for constructing and retrievalling a data base of a medical image by using a content-based indexing technique and recording medium thereof - Google Patents

Abstract

PURPOSE: A method for constructing and searching a medical image database using a content-based indexing technique and a recording media therefor are provided to effectively construct the medical image database by using a feature vector extracted based on the content of the medical image. CONSTITUTION: If a medical image data is inputted when a medical image processing system is a medical image construction mode(302), an image segmentation block segments the inputted image(304). The segmented medical image, a portion to be stored in a database, is segmented into psycho visual frequency domains(306). A feature vector to be used as a search key in the database is extracted from the medical image segmented into frequency spaces(308). An indexing feature vector is obtained by indexing the extracted feature vector(310). The obtained indexing feature vector is stored with the corresponding input medical image in the database(312).

Description

METHODS FOR CONSTRUCTING AND RETRIEVALLING A DATA BASE OF A MEDICAL IMAGE BY USING A CONTENT-BASED INDEXING TECHNIQUE AND RECORDING MEDIUM THEREOF}

The present invention relates to a technique of making a medical image signal into a database and searching for a similar medical image using the same. More particularly, the present invention relates to a method of constructing a database of medical images based on the content of each medical image. A medical image database construction method using a content-based indexing technique suitable for searching similar medical images having a similarity to a query medical image using a database of the medical images, and a method of searching the same.

In recent years, when treating a real patient in a medical facility such as a hospital or when doctors conduct research on various lesions, medical images of patients' lesions are widely used, and medical image processing systems using computers are widely used. It is put to practical use one after another.

The prior art related to the medical image processing as described above was proposed by Hachiro Mukai et al. And registered in the U.S. Patent Office as 5,019,975 under the name "Method for constructing a data base in a medical image control system" Is called).

This prior patent 1 relates to a filing system, which records character information representing each medical image in a database and based on a search result (for example, the number of searches, etc.) of each query image. A method of recording the importance of each recorded medical image is disclosed. Here, the query image refers to an image for which a user (for example, a doctor) targets a search.

That is, in the prior patent 1, when a query image (that is, query character information) is given in a state in which character information of each medical image is stored in a database as search key information, medical information corresponding to the character information of the given query image is given. The text information of the image is retrieved from the database, and the importance of the retrieved medical image is increased based on the search result. Also, when the new medical image to be stored is provided while the database of a limited capacity is filled, the importance is low. Disclosed is a method of controlling a medical image database by deleting a medical image, for example, a medical image having the least number of retrieval times and related character information (search key information) and storing new medical image and related character information. .

However, Prior Patent 1 which manages a database in the manner described above has some advantages in that it is possible to efficiently manage a database having a limited capacity and to search for important image data more efficiently. Since the medical images are variable and many text information must be stored as search key information, the fundamental problem of efficiently using the storage space has not been solved. That is, the prior patent 1 has a limit in maximizing the utilization efficiency of the storage space having a limited storage capacity.

In addition, prior patent 1 describes how to select a medical image to be deleted when a new medical image to be stored is received when the importance (that is, the number of retrieval times) of each of the medical images stored in the storage space is greater than or equal to the threshold. It contains problems that can cause serious problems such as being unknown.

Meanwhile, another related art related to medical image processing is a prior patent (hereinafter referred to as Prior Patent 2) proposed by David Gur et al. And registered with US Patent Office at 5,241,472 under the name of “Method of identifying and archiving medical images”.

The prior patent 2 relates to a method of automatically recording the contents of a medical image using text information provided therein. It is included. Here, the text file is used to form a text header in combination with the video associated with it, and the video and text header stored for each page is used as a directory address for searching. That is, the prior patent 2 discloses a technical idea of using a character header information as if a message is inserted into an image page, that is, when indexing contents of a stored medical image.

However, although Patent Document 2 described above may be useful for quickly and easily searching for a relatively simple image, the contents of the medical image are complicated and in various cases, the character header information becomes very large, thereby having a limited storage capacity. There is no limit to maximizing the utilization efficiency of space. In other words, in order to accurately represent a complex medical image, the amount of text header information is inevitably increased. In this case, the text header information for retrieval takes up an unnecessary amount of storage space. Is caused.

In addition, the prior patent 2 using the character header information to search for similar medical images has a fundamental problem that the search efficiency for similar medical images is inevitably deteriorated due to the limitation of the expressive power of the characters.

On the other hand, another conventional technology related to medical image processing is proposed by William C. Mortimore et al., A prior patent registered as 5,740,428 in the U.S. Trademark Office under the name of "Computer based multimedia medical database management system and user interface" 3).

Such a prior patent 3 relates to a computer database for storing and effectively processing medical images from various multimedia medical image data. Each patent object is connected to each other by creating a single identifier. When the identifier is divided into video and text. That is, the encoder generates an identifier suitable for various multimedia medical image data, and the decoder analyzes the identifier and connects the medical image data.

Therefore, the prior patent 3 may extract an identifier from various medical image data and associate the extracted identifier with each medical image, thereby improving the use efficiency of the storage space to some extent, but the identifier generated by the encoder How accurately the content of medical images can be represented has a new problem. That is, since the prior patent 3 does not consider the contents of the medical image when extracting the identifier, the accuracy of the classification according to the contents of the medical image is inevitably reduced, which in turn lowers the retrieval efficiency of similar medical images for the query medical image. It causes a new problem of acting as a cause.

Accordingly, the present invention is to solve the above problems of the prior art, it is possible to extract a feature vector based on the contents of the medical image, and to effectively build a database of the medical image using the extracted feature vector The purpose is to provide a medical image database construction method using content-based indexing technique.

Another object of the present invention is a content-based indexing technique capable of quickly searching for a similar medical image corresponding to a feature vector based on the content of a query image in a database of medical images constructed by using a content-based feature vector. To provide a medical image retrieval method using a.

Another object of the present invention is to build a database for medical images using content-based feature vectors, and to record a program for retrieving similar medical images from a database constructed using content-based feature vectors extracted from query images. The present invention provides a recording medium that can be read by a computer.

According to an aspect of the present invention, there is provided a method of constructing a database capable of searching and reading medical images photographed for each part of a human body, the method comprising: input medical care based on the shape of an lesion or anatomical data A first process of dividing an image; A second process of dividing the divided medical image into a psycho visual frequency domain; A third process of extracting a content-based feature vector by analyzing the frequency-divided medical image in a frequency space; A fourth process of indexing the extracted feature vector to derive an indexing feature vector of the input medical image; And a fifth process of storing the extracted indexing feature vector together with the input medical image in the database.

According to another aspect of the present invention, there is provided a method of searching for a similar medical image corresponding to a query image in a database in which medical images photographed for each part of the human body are stored together with search key information. A first step of dividing the input query image into a psycho visual frequency region when a query image to be searched in the database is input; A second process of extracting a content-based feature vector by analyzing the frequency-divided query image in frequency space; A third process of indexing the extracted feature vector to derive an indexing feature vector of the query image; And searching for a similar medical image by comparing and searching the derived query indexing feature vector and the indexing feature vector of each medical image stored in the database, and extracting the retrieved similar medical image from the database to display on the monitor. We provide a medical image retrieval method using a four-step content-based indexing technique.

According to another aspect of the present invention, a medical image photographed for each part of the human body is constructed into a searchable and readable database, and the query image aimed at searching is constructed from the constructed database. A computer readable recording medium for recording a program for retrieving a corresponding pseudo medical image, comprising: a first step of dividing an input medical image based on a shape of an lesion or anatomical data; A second process of dividing the divided medical image into a psycho visual frequency domain; A third process of extracting a content-based feature vector by analyzing the frequency-divided medical image in a frequency space; A fourth process of indexing the extracted feature vector to derive an indexing feature vector of the input medical image; A fifth process of storing the extracted indexing feature vector in the database together with the input medical image; A sixth step of dividing the input query image into a psycho visual frequency region when a query image to be searched in the database is input; A seventh process of extracting a content-based feature vector by analyzing the frequency-divided query image in frequency space; An eighth step of indexing the extracted feature vector to derive an indexing feature vector of the query image; And searching for a similar medical image by comparing and searching the derived query indexing feature vector and the indexing feature vector of each medical image stored in the database, and extracting the retrieved similar medical image from the database to display on the monitor. Provided is a computer readable recording medium having recorded thereon a program for executing the process.

1 is a process of constructing a database of medical images using a feature vector extracted based on the contents of a medical image according to the present invention, and searching for similar target medical images in a database constructed using the feature vector. Block diagram of a medical image processing system comprising functional blocks

FIG. 2 is a diagram for explaining a spatial frequency region in the radial direction and the angular direction when dividing a medical image into psycho visual frequencies according to the present invention; FIG.

3 is a flowchart illustrating a process of constructing a database of medical images using a content-based indexing technique according to a preferred embodiment of the present invention;

4 is a flowchart illustrating a process of searching for a similar medical image corresponding to an arbitrary query image of interest in a database of medical images constructed by using a content-based indexing technique according to the present invention.

<Description of the code | symbol about the principal part of drawing>

102: selection signal input block 104: video segmentation block

106: frequency division block 108: feature vector extraction block

110: feature vector indexing block 112: database

114: Feature Vector Indexing Search Block 116: Similar Medical Image Array Block

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

1 is a process of constructing a database of medical images using a feature vector extracted based on the contents of a medical image according to the present invention, and searching for similar target medical images in a database constructed using the feature vector. It is a block diagram of a medical image processing system configured by functional block.

Referring to FIG. 1, a medical image processing system constructed by functional block includes a selection signal input block 102, an image segmentation block 104, a frequency division block 106, a feature vector extraction block 108, and a feature vector indexing block ( 110, database 112, feature vector indexing search block 114, and similar medical image arrangement block 116.

First, the selection signal input block 102 includes input means such as a keyboard and a mouse, for example, assuming that the medical image processing system is mounted on a personal computer. When a user manipulation signal is inputted, a medical image construction control signal for storing a corresponding selection signal (or control signal), for example, a medical image input from the outside in the database 112, being displayed on a monitor. In the medical image area selection signal corresponding to the part selected by the user (for example, a doctor, etc.) based on the shape of the lesion or the anatomical knowledge that the user (doctor) knows, Image segmentation through line L11 by generating a search control signal or the like for searching the corresponding similar medical image in database 112 Providing a lock (104) and / or feature vector indexing block 110.

Next, the medical image data provided from the outside through the line L12 for the construction of the database is provided to the image segmentation block 104 and the database 112, respectively (medical image construction mode), and also to the user request or selection. Accordingly, query image data for searching for a similar medical image in the database 112 is provided to the image segmentation block 104 (query image search mode). The image segmentation block 104 is a line L12 in the medical image construction mode. The medical image data provided through the data is provided to the frequency division block 106 as it is, or the partial medical image data selected from the medical image on the line L12 is provided to the frequency division block 106 according to the medical image region selection signal on the line L11. And a query image input from the outside in response to the search selection signal from the line L11 in the query image search mode. It provides a foundation or part of the query image data in a frequency division block 106.

That is, in the image segmentation block 104, in response to the selection signal provided through the line L11, the portion of the image (i.e., the shape of the lesion or the user (doctor)) to be stored in the database or the query search in the database is The image part selected based on the known anatomical knowledge is divided and transmitted to the frequency division block 106.

The frequency dividing block 106 divides the medical image or the query image provided from the image dividing block 104 into a psychosypic frequency domain, and then provides the extracted image to the feature vector extraction block 108. Here, psycho visual frequency division refers to two-dimensional image data in the Fourier region, as shown in FIG. 2 as an example, in the frequency space based on the radial direction of S21 and the angular direction of S22. It means to divide.

As is well known, the frequency characteristic of the human optic nerve is divided into filter banks in a band-limited region, i.e., several visual channels with constant bandwidth. The spatial frequency domain division suitable for such a human visual system is divided into octave bands in the radial direction and equal bands in the angular direction, as shown in FIG. Therefore, the present invention divides the input medical image or query image into psycho visual frequencies in consideration of the human visual characteristics.

Next, the feature vector extraction block 108 analyzes signals in the frequency space, i.e., radon space, provided from the frequency division block 106, and extracts feature vectors based on the contents of the medical image. A feature vector FV is extracted using the energy average and the energy variance as shown in Equation 1 below. That is, the feature vector is composed of energy averages and energy variances in the spatial frequency band.

FV = [{F _mean (n, m)}, {F _dev (p, q)}]

In Equation 1, n and m represent the order of the radial direction and the angular direction of the psycho visual frequency division for energy average, respectively. Thus, F _mean (n, m) represents the energy mean in n, m spatial frequency bands. In addition, p, q represents the order of the radial direction and the angular direction of the psycho visual frequency division for energy dispersion, respectively. Thus, F _dev (p, q) represents the energy variance in the p, q spatial frequency band.

Here, the energy average F _mean (n, m), which is one of the elements constituting the feature vector, may be calculated as in Equation 2 below.

In Equation 2, P (ω _γ , ω _θ ) denotes energy in the spatial frequency domain, W _r denotes a weight in the radial direction, W _a denotes a weight in the angular direction, and fd denotes a frequency domain.

In addition, the energy dispersion F _dev (p, q), which is one of the elements constituting the feature vector, may be calculated as in Equation 3 below.

That is, the feature vector extraction block 108 obtains the energy average and the energy dispersion in the spatial frequency band using Equations 2 and 3, respectively, and based on the obtained energy average and the energy dispersion, the psycho visual frequency-division medical The feature vector of the image or the query image is extracted, and the extracted feature vector is provided to the feature vector indexing block 110.

Next, the feature vector indexing block 110 performs indexing on the feature vector provided from the feature vector extraction block 108, where the indexing feature vectors obtained are provided to the database 112 via line L13, or Or to feature vector indexing search block 114 via line L14. That is, the feature vector indexing block 110 extracts the indexing feature vector extracted from the input medical image when the medical image building selection signal is provided from the selection signal input block 102 through the line L11 (that is, in the medical image building mode). When the query image search selection signal is provided from the selection signal input block 102 via the line L11 to be stored with the corresponding medical image, and is transmitted through the line L13 (ie, the query image search mode). The indexing feature vector extracted from the query medical image is transferred to the feature vector indexing search block 114 through the line L14.

Therefore, when the current mode is the medical image construction mode, the database 112 stores the medical image data provided through the line L11 and the associated indexing feature vector provided through the line L13 in an arbitrary space.

That is, according to the present invention, the database 112 stores the medical images to be databased and indexing feature vectors corresponding to the respective medical images, wherein the indexing feature vectors corresponding to the medical images are stored in the query image. It is used as search index (or search key) information when looking for a similar medical image.

In this case, the medical images stored in the database 112 according to the present invention search for an indexing feature vector (that is, a feature vector extracted by considering human visual characteristics) extracted based on an energy average and an energy dispersion in a spatial frequency band. Because it stores as key information, it is possible to accurately express the characteristics of each of the stored medical images, and unlike the aforementioned patents, which require storing a lot of text information as search key information, it is extracted based on energy average and energy dispersion. Since only indexing feature vectors are stored, the occupancy of the memory space occupied by the search key information can be greatly reduced, thereby achieving high efficiency of using storage space in a database.

On the other hand, the feature vector indexing search block 114 performs a database based on the indexing feature vector extracted from the query image when the indexing feature vector of the query image is provided through the line L14, that is, when the current mode is the query image search mode. Search for indexing feature vectors of the medical images stored at 112, that is, search indexing feature vectors of the medical images using the indexing feature vector of the query image as a search key, thereby sequentially searching similar medical images having similarities with the query image, The retrieved similar medical images are sequentially retrieved (or read) from the database 112 and provided to the next similar medical image array block 116. For example, assuming that 20 similar medical images were retrieved for a query image, the feature vector indexing search block 114 reads sequentially from the most similar medical image (ie, the medical image with the most similar indexing feature vector). The similar medical images retrieved are retrieved.

At this time, according to the present invention, since each medical image has an indexing feature vector extracted based on an energy average and an energy dispersion in a spatial frequency band as its search key, the content-based feature of each medical image stored in the database 112. High-definition representation of the image, and similar medical image of the query image by searching the indexing feature vector of each stored medical image using the indexing feature vector of the query image extracted based on the energy mean and energy dispersion in the spatial frequency band Because of this, the search for similar medical images corresponding to the query images can be precisely realized.

In addition, the similar medical image arrangement block 116 configures each of the medical images provided with the indexing feature vectors in a similar order in a graphic menu screen according to the similarity order, and the graphic menu screen data configured here is displayed on the display side, not shown. Delivered. Accordingly, the medical image menu screen data including a plurality of searched similar medical images similar to the query image selected (or input) by the user will be displayed on the monitor (not shown).

On the other hand, in the above-described embodiment, the input medical image is stored as it is stored in the database together with the corresponding indexing feature vectors. However, the present invention is not limited thereto, and JPEG, which is well known in the art, It is also applicable to the case where the compressed and compressed medical image data is stored in a data base using a standardized compression algorithm such as MPEG. In this case, an encoder and a decoder for encoding the medical image on the input and output sides of the medical image are used. It will be provided with.

Next, a process of constructing a database of medical images using content-based feature vectors in the medical image processing system having the above-described configuration will be described.

FIG. 3 is a flowchart illustrating a process of constructing a database of medical images by using a content-based indexing technique according to a preferred embodiment of the present invention.

Referring to FIG. 3, when the medical image data is input while the medical image processing system is in the medical image building mode (step 302), the image segmentation block 104 of FIG. 1 stores the input medical image as a database. The desired portion is divided (step 304). The division of the medical image may be arbitrarily performed based on the shape of the lesion or anatomical knowledge known to the user (doctor) according to the user's selection.

Next, in step 306, the divided medical image (i.e., the portion of the medical image to be stored in the database) is divided into psycho-visual frequency regions, more specifically, as shown in FIG. The frequency space is divided based on the radial direction of S21 and the angular direction of S22 in the Fourier region with respect to the image data.

Further, in step 308, a feature vector is extracted for use as a search key in the database for the medical image divided into the frequency space, and more specifically, based on the energy average and energy dispersion in the spatial frequency band, Extract the feature vector of the image. To this end, in the feature vector extraction block 108 of FIG. 1, the energy averages F _mean (n, m) and energy variance F _dev (p, q) of the input image are obtained by using Equations 2 and 3 described above. As shown in Equation 1, the calculated feature vector FV of the medical image is calculated using the calculated energy mean F _mean (n, m) and energy variance F _dev (p, q). Extract.

Next, in step 310, the feature vector extracted from the input medical image is indexed to obtain an indexing feature vector, and the indexing feature vector of the obtained medical image is a predetermined region of the database 112 together with the corresponding input medical image. Is stored in step 312.

That is, in the method for constructing a medical image database according to the present invention, a feature vector is extracted based on an energy average and an energy dispersion in a frequency band for each input medical image, and the indexed feature vector obtained by indexing the extracted feature vector is applied. The search key (or search index) information for searching a medical image is stored in a database.

Therefore, the medical images stored in the database according to the present invention store the indexing feature vectors extracted based on the energy average and energy dispersion in the spatial frequency band, that is, the feature vectors extracted in consideration of the human visual characteristics, as the search keys. Therefore, unlike the above-mentioned patents, in which a precise representation of the characteristics of each of the stored medical images is possible, and a lot of textual information has to be stored as search key information for each of the medical images constructed in the database, energy average and energy Since only indexing feature vectors extracted based on the variance are stored, the occupancy of the memory space occupied by the search key information can be greatly reduced, that is, the excessive use of the memory space by the search information of the medical image can be suppressed. It can greatly improve the efficiency of use.

Next, a process of searching for a similar medical image using an indexing feature vector of a query image in a database constructed using a content-based feature vector in a medical image processing system having the above-described configuration will be described. do.

FIG. 4 is a flowchart illustrating a process of searching for a similar medical image corresponding to an arbitrary query image of interest in a database of medical images constructed by using a content-based indexing technique according to the present invention.

Referring to FIG. 4, as described above, a medical image database is constructed according to the present invention, that is, a medical image and an indexing feature vector (search key information) corresponding to each medical image are stored (step 402). ), When the user selects a query image search mode, that is, a query image is input (step 404), the image segmentation block 104 of FIG. 1 divides a portion to be searched from the database for the input query image. The division of the query image may be arbitrarily performed based on the shape of the lesion or anatomical knowledge known to the user (doctor) according to the user's selection.

Next, in step 406, the divided query image (i.e., the part of the query image to be searched for similar medical images in the database) is divided into psycho visual frequency regions, more specifically, as shown in FIG. 2 as an example. Likewise, the frequency space is divided with respect to the two-dimensional image data based on the radial direction of S21 and the angular direction of S22 in the Fourier region.

In operation 408, the content-based feature vector is extracted from the query image divided into the frequency space, that is, the feature vector of the query image is extracted based on the energy average and energy dispersion in the spatial frequency band. The feature vector extraction block 108 of Figure 1 For this purpose, the energy average of the query image by using the aforementioned equation (2) and 3 (F _mean (n, m)) and the energy distribution (F _dev (p, q)) for The content-based feature vector (FV) of the query image is calculated by using the calculated energy mean (F _mean (n, m)) and the energy variance (F _dev (p, q)) as shown in Equation 1 above. ).

Next, in step 410, a feature vector extracted from the query image is indexed to obtain an indexing feature vector. The indexed feature vector of the query image thus obtained is cut into the feature vector indexing search block 114 through the line L14.

Meanwhile, in step 412, indexing feature vectors of the medical images stored in the database 112 are searched based on the indexing feature vector of the query image, that is, the indexing feature vector of the medical images using the indexing feature vector of the query image as a search key. By searching the data, the similar medical images having similarity with the query image are sequentially searched, and the retrieved similar medical images are sequentially retrieved from the database 112. For example, assuming that 20 similar medical images are retrieved for one query image, the similar medical images retrieved are retrieved by sequentially reading from the most similar medical image (that is, the medical image with the most similar indexing feature vector). do.

Next, in step 414, each of the medical images from which the indexing feature vectors are drawn in a similar order is arranged in a similarity order to form a graphic menu screen, and the graphic menu screen data configured here is transferred to the display side, not shown. do. Therefore, on the monitor not shown, medical image menu screen data including a plurality of searched similar medical images related to the query image selected by the user is displayed.

Therefore, the medical images stored in the database according to the present invention have an indexing feature vector extracted based on the energy average and the energy dispersion in the spatial frequency band as its search key, thereby providing high precision for the content-based features of the stored medical images. It is possible to express a similar medical image based on the indexing feature vector of the query image and the indexing feature vector of each stored medical image based on the energy mean and energy dispersion in the spatial frequency band. The search for similar medical images corresponding to the query images can be precisely realized.

On the other hand, according to the method for constructing a medical image database of the present invention, a database for medical images is constructed using content-based feature vectors as search key information, and data is constructed using content-based feature vectors extracted from a query image. A program or algorithm for retrieving similar medical images from a base can be recorded and used on a portable and portable recording medium (for example, CD), and the recording medium can be read using a device such as a computer.

As described above, according to the present invention, unlike the aforementioned patents, which require storing a large amount of text information as search key information for each medical image constructed in a database, the energy average and energy in the spatial frequency band of each medical image. Since indexing feature vectors extracted based on variance are stored in the database together with each medical image as the search key information, not only accurate representation of the content-based features of each medical image is possible, but also in the database occupied by the search key information. It is possible to greatly reduce the occupancy of the memory space, that is, to prevent excessive consumption of the memory space by the search information of the medical image, it is possible to significantly improve the utilization efficiency of the storage space.

In addition, according to the medical image retrieval method of the present invention, the similar medical image for the query image based on the indexing feature vector of the query image extracted based on the energy average and energy dispersion in the spatial frequency band and the indexing feature vector of each stored medical image. Since the image is searched, the similar medical image search corresponding to the query image can be accurately realized.

Claims (11)

In the method of constructing a database that can retrieve and read medical images taken for each part of the human body, A first process of segmenting the input medical image based on the shape of the lesion or the anatomical data; A second process of dividing the divided medical image into a psycho visual frequency domain; A third process of extracting a content-based feature vector by analyzing the frequency-divided medical image in a frequency space; A fourth process of indexing the extracted feature vector to derive an indexing feature vector of the input medical image; And And a fifth process of storing the extracted indexing feature vector in the database along with the input medical image. The method of claim 1, wherein the third process is: Calculating an energy average (F _mean (n, m)) in the spatial frequency band of the frequency-divided medical image; Calculating an energy variance F _dev (p, q) in the spatial frequency band of the frequency-divided medical image; And And extracting a feature vector of the input medical image based on the calculated energy average and the energy variance. The method of claim 2, wherein the energy average F _mean (n, m) is calculated by the following equation. (Where n and m represent the order of the radial and angular directions of the psycho visual frequency division for the energy average, P (ω _γ , ω _θ ) represents the energy of the spatial frequency domain, and W _r represents the radial direction Denotes the weight of, W _a denotes the weight in the angular direction, fd denotes the frequency domain) The method of claim 3, wherein the energy variance F _dev (p, q) is calculated by the following equation. (P and q in the above equations represent the order of the radial and angular directions of the psycho visual frequency division for energy dispersion, respectively) A method of searching for a similar medical image corresponding to a query image in a database in which medical images photographed for each part of the human body are stored together with search key information thereof, A first process of dividing the input query image into a psycho visual frequency region when a query image to be searched in the database is input; A second process of extracting a content-based feature vector by analyzing the frequency-divided query image in frequency space; A third process of indexing the extracted feature vector to derive an indexing feature vector of the query image; And A fourth medical image for retrieving a similar medical image by comparing and searching the derived query indexing feature vector and the indexing feature vector of each medical image stored in the database, and extracting the retrieved similar medical image from the database and displaying the same on the monitor Medical Image Retrieval Method using Content-based Indexing Technique. The method of claim 5, wherein the medical image search method comprises sequentially arranging a plurality of similar medical images extracted when a plurality of similar medical images corresponding to the query image are searched and retrieved in order of similarity of indexing feature vectors. A medical image retrieval method using a content-based indexing technique, characterized in that it further comprises a screen configuration. The method of claim 5 or 6, wherein the second process is: Calculating an energy average (F _mean (n, m)) in a spatial frequency band of the frequency-divided query image; Calculating an energy variance (F _dev (p, q)) in a spatial frequency band of the frequency-divided query image; And And extracting a feature vector of the query image based on the calculated energy average and the energy variance. The method of claim 7, wherein the energy average (F _mean (n, m)) is calculated by the following equation. (Where n and m represent the order of the radial and angular directions of the psycho visual frequency division for the energy average, P (ω _γ , ω _θ ) represents the energy of the spatial frequency domain, and W _r represents the radial direction Denotes the weight of, W _a denotes the weight in the angular direction, fd denotes the frequency domain) The method of claim 8, wherein the energy variance F _dev (p, q) is calculated by the following equation. (P and q in the above equations represent the order of the radial and angular directions of the psycho visual frequency division for energy dispersion, respectively) A program for retrieving and reading a medical image of each part of the human body into a searchable and readable database, and recording a program for retrieving a similar medical image corresponding to a query image to be searched from the constructed database. A readable recording medium, A first process of segmenting the input medical image based on the shape of the lesion or the anatomical data; A second process of dividing the divided medical image into a psycho visual frequency domain; A third process of extracting a content-based feature vector by analyzing the frequency-divided medical image in a frequency space; A fourth process of indexing the extracted feature vector to derive an indexing feature vector of the input medical image; A fifth process of storing the extracted indexing feature vector in the database together with the input medical image; A sixth step of dividing the input query image into a psycho visual frequency region when a query image to be searched in the database is input; A seventh process of extracting a content-based feature vector by analyzing the frequency-divided query image in frequency space; An eighth step of indexing the extracted feature vector to derive an indexing feature vector of the query image; And A ninth medical image is searched by comparing and searching the derived query indexing feature vector and the indexing feature vector of each medical image stored in the database, and the retrieved similar medical image is retrieved from the database and displayed on a monitor. A computer-readable recording medium having recorded thereon a program for executing a process. The method of claim 10, wherein the program recorded on the recording medium sequentially indexes the plurality of similar medical images extracted when the plurality of similar medical images corresponding to the query image are searched and retrieved, according to the similarity order of the feature vectors. And arranging to form a menu screen.