KR102152385B1 - Apparatus and method for detecting singularity - Google Patents

Abstract

A diagnostic device is provided. A diagnosis apparatus includes: an acquisition unit for obtaining a medical image photographing an organ inside the body; and a search unit for searching for singular points protruding or depressed from a boundary line of the organ included in the medical image. The present invention can automatically diagnose protrusion-shaped lesions generated on one side of the organ such as a lung nodule accurately and quickly.

Description

Singularity diagnostic device and diagnostic method {APPARATUS AND METHOD FOR DETECTING SINGULARITY}

The present invention relates to an apparatus and method for diagnosing a singularity protruding or depressing on one surface of a body organ.

Solitary Pulmonary Nodule is a small globular lesion less than 3cm in diameter inside the lung. There is usually only one, so it is also called an isolated pulmonary nodule. When the size of the lesion exceeds 3cm in diameter, it is called a tumor or lump, not a nodule.

As it is usually asymptomatic, it is detected through chest X-rays or CT scans during a medical examination. In recent years, as CT has become more common, lesions with ground-glass opacity less than 1cm, which have not been found before, are also clinically approached in accordance with isolated pulmonary nodules.

In Korean Patent Publication No. 1887194, when a chest image of a subject is input, the contents of reading a chest image based on a deep learning model are shown.

Korean Patent Registration No. 1887194

An object of the present invention is to provide an apparatus and method for accurately and quickly automatically diagnosing protrusion-shaped lesions generated on one side of an organ such as a lung nodule.

The diagnosis apparatus of the present invention includes: an acquisition unit for obtaining a medical image photographing an organ inside a body; It may include; a search unit for searching for a singular point protruding or depressed from the boundary line of the organ included in the medical image.

The diagnostic apparatus of the present invention comprises: a first diagnostic unit for diagnosing a pleural nodule using a medical image of the chest; A second diagnostic unit for diagnosing a lung nodule using the medical image; And an integrated unit that integrates the diagnosis result of the first diagnosis unit and the diagnosis result of the second diagnosis unit.

In the diagnosis method of the present invention, a gradient is calculated for a boundary line of an organ included in a medical image, and a singular point in which the change amount or rate of change of the gradient satisfies a set value may be searched on the boundary line.

The diagnostic method of the present invention includes the steps of: when a computed tomography (CT) image of the chest is input, extracting a binary mask from which a lung boundary line is divided through image processing; Extracting a closed boundary line of a set thickness using the binary mask; Calculating a 3D gradient for the lung boundary line; Searching for a portion in which the change in the 3D gradient satisfies the set condition by using the fact that the change value of the gradient for the pleural nodule portion satisfies the set condition; Extracting the searched part in the form of a 3D cube patch; And diagnosing whether the 3D cube patch has a pleural nodule using a pleural nodule classification model derived through machine learning for only the pleural nodule.

In the diagnostic apparatus and method of the present invention, a singular point protruding or depressing on one surface (boundary line) of an organ may be identified through a method of calculating gradients on the boundary line of an organ included in a photographed medical image.

The peculiarity can be provided to the medical staff or used to automatically diagnose feces.

Only a partial fragment image (suspicious region) including a singular point in a medical image of a specific patient may be removed, and an automatic diagnosis may be performed intensively for the corresponding fragment image. The automatic diagnosis at this time may be performed through a diagnosis model generated by machine learning. Since the diagnosis object of the diagnosis model is limited to a partial image of the medical image, not the entire or the entire boundary line, the diagnosis model can diagnose the disease at high speed using a small load and a small storage space.

In particular, according to the present invention, it may be very useful for automatic diagnosis of pleural nodules, which are difficult to automatically diagnose among lung nodules. Automatic diagnosis of pulmonary nodules can be made by a method of searching for a circular shape without a lung in a medical image. The pleural nodules formed on the pleura are not circular, but have a hemispherical shape with a pleura attached thereto, making automatic detection difficult.

In the present invention, focusing on the point where the pleura corresponds to the boundary line of the lung image, a portion where the gradient change of the boundary line is abrupt can be identified or estimated as a pleural nodule.

In addition, according to the present invention, pleural nodules can be diagnosed using an AI model that intensively learns only pleural nodules that are difficult to detect and classify. Accordingly, the diagnostic performance of pleural nodules can be improved, and even the performance of the entire lung nodule detection model can be improved.

In addition, a comparative example in which the entire area of the lung boundary is collectively extracted as a 3D cube patch state and it is necessary to determine whether or not a pleural tubercle is formed for all 3D cube patches may be prepared. Compared to the comparative examples, the present invention selectively extracts and classifies only suspected pleural nodules using gradients calculation, thereby saving time required (only a small number of cube patches need to be determined) and memory saving (requires storing all cube patches) None) can be.

1 is a schematic diagram showing a diagnostic device of the present invention.
2 is a photograph showing the types of nodules.
3 is a photograph showing a pleural nodule.
4 is a photograph showing a patch.
5 is a schematic diagram showing the operation of the extraction unit.
6 is a block diagram showing another diagnostic device of the present invention.
7 is a flow chart showing a diagnosis method of the present invention.
8 is a diagram illustrating a computing device according to an embodiment of the present invention.

Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings so that those of ordinary skill in the art can easily implement the embodiments of the present invention. However, the present invention may be implemented in various different forms and is not limited to the embodiments described herein. In the drawings, parts irrelevant to the description are omitted in order to clearly describe the present invention, and similar reference numerals are assigned to similar parts throughout the specification.

In this specification, redundant descriptions of the same components are omitted.

In addition, in the present specification, when a component is referred to as being'connected' or'connected' to another component, it may be directly connected or connected to the other component, but other components in the middle It should be understood that may exist. On the other hand, in the present specification, when it is mentioned that a certain element is'directly connected' or'directly connected' to another element, it should be understood that no other element exists in the middle.

In addition, terms used in the present specification are only used to describe specific embodiments, and are not intended to limit the present invention.

In addition, in the present specification, a singular expression may include a plurality of expressions unless the context clearly indicates otherwise.

In addition, in the present specification, terms such as'include' or'have' are only intended to designate the existence of features, numbers, steps, actions, components, parts, or a combination thereof described in the specification, and one or more It is to be understood that the presence or addition of other features, numbers, steps, actions, components, parts, or combinations thereof, does not preclude in advance the possibility of being excluded.

In addition, in this specification, the term'and/or' includes a combination of a plurality of listed items or any of a plurality of listed items. In the present specification,'A or B'may include'A','B', or'both A and B'.

In addition, in this specification, detailed descriptions of known functions and configurations that may obscure the subject matter of the present invention will be omitted.

1 is a schematic diagram showing a diagnostic device of the present invention.

The diagnostic apparatus illustrated in FIG. 1 may include an acquisition unit 110, an extraction unit 130, a search unit 150, and a determination unit 170.

The acquisition unit 110 may acquire a medical image photographing an organ inside the body. The acquisition unit 110 may be provided with a photographing means for photographing a medical image, or a wired or wireless communication means for receiving a medical image from an external photographing means.

The internal organs of the body may refer to the internal organs of the patient. A system consisting of organs that are morphologically and functionally common and that act complementarily is called an organ system. In particular, the digestive system, respiratory system, genitourinary system, and endocrine system are combined to be called visceral system. Singularities such as nodules, tumors, and wounds may occur on one side of the organ due to various factors. The singularity may protrude or be recessed on one surface of the organ.

Medical images may be generated through computed tomography (CT), magnetic resonance imaging (MRI), X-ray, positron emission tomography (PET), or the like. The medical image is photographed by targeting an organ inside the patient's body, and may include an image representing the organ. By analyzing medical images, various singularities can be identified. For example, there may be a deep learning model that detects lung nodules by analyzing images of lungs displayed on CT images. The detection of a pulmonary nodule can have a large difference in detection difficulty and performance depending on the type of nodule.

2 is a photograph showing the types of nodules.

As shown in FIG. 2, the type of nodule is a circular nodule inside the lung area (well-circumscribed nodule, the leftmost picture in FIG. 2), a nodule attached to the blood vessel (juxta-vascular nodule, second from the left in FIG. 2). Photo), a nodule with a pleural tail (third picture from the left in Fig. 2), a nodule attached to the pleura (juxta-pleural nodule, a fourth picture from the left or far right picture in Fig. 2) It can be classified as

Among them, the nodules attached to the pleura (hereinafter referred to as'pleural nodules') are formed in a semispherical shape. In general, the detection performance of the deep learning model for pleural nodules is remarkably low because of the difference in shape from other spherical nodules and the similarity of voxel values between the pleural region and pleural nodules. For this reason, there is currently no deep learning model for lung cancer detection that has the function of focusing on the detection of pleural nodules. In addition, there is no algorithm capable of intensively managing only pleural nodules with higher detection difficulty than other nodules. In addition, it is necessary to consider the time required, memory usage, and cost.In general medical deep learning, it is also possible to apply a high-cost algorithm that extracts patches for all images at once to an algorithm that focuses on specific lesions such as pleural nodules. It is difficult.

Therefore, if the performance of detecting pleural nodules is improved through a low-cost pleural nodule concentration algorithm, the practicality of the pulmonary nodule detection model can also be expected to increase. The present invention can provide a method for rapidly detecting or diagnosing various peculiar points including a pleural nodule, which is difficult to detect, with a simple configuration. This method may be performed by the search unit 150.

The search unit 150 may search for a singular point protruding or depressing from a boundary line of an organ (image) included in the medical image.

Organ boundaries can represent the surface of the organ. Therefore, the singularity protruding from the surface of the organ may indicate a lesion such as a nodule, a tumor, or a lump. Singularities depressed from the surface of the organ may indicate wounds, dent lesions, and the like.

The search unit 150 needs to quickly and accurately find outliers.

3 is a photograph showing a pleural nodule.

The search unit 150 may calculate a 3D gradient targeting a boundary line of an organ (viscera) in a medical image. According to the 3D gradient operation, a slope and a gradient of a boundary line may be calculated. The search unit 150 may calculate a change amount of the 3D gradient or a rate of change of the 3D gradient once again. For example, the amount of change of the gradient or the rate of change of the gradient may be obtained by normalizing the result of the initial gradient operation and calculating the gradient once again on the result of the normalized gradient operation.

Gradient operation may use, for example, Histogram of gradients (HOG). In FIG. 3, an arrow drawn perpendicular to the boundary line of the organ (the boundary of the inner wall surface) may be a result of a gradient operation using HOG. A section in which the direction of the arrow rapidly changes can be estimated as a singular point. In this case, "abrupt" may mean that a value or condition outputted when a gradient operation is applied to an actual singular point is satisfied.

When the medical image includes data in a 3D space formed by the x-axis, y-axis, and z-axis that are orthogonal to each other, the gradient operation may be performed three-dimensionally.

The search unit 150 may search for a singularity using a gradient change amount or a gradient change rate. For example, the search unit 150 may select, as a singular point, a portion having a rate of change of a gradient that differs by more than a set value than a rate of change of the gradient of a peripheral boundary line.

The acquisition unit 110 may acquire a pleural image. In this case, the search unit 150 may calculate a three-dimensional gradient targeting the boundary line of the pleura, and use the gradient to search for a juxta-pleural nodule protruding from the boundary line of the pleura in a hemispherical shape.

The peculiar points such as the pleural nodules searched for by the search unit 150 indicate protrusions, recessed grooves, etc. on the inner wall of the organ, the surface, etc., and the actual presence of the disease may be determined through an additional determination procedure. The additional determination procedure may be performed by a medical staff or may be automatically performed through the determination unit 170.

The determination unit 170 may perform calculations for a time longer than the required operation time of the search unit 150 in order to accurately determine whether or not to change the disease. Consequently, in order to shorten the operation time of the entire diagnosis apparatus, it is necessary to shorten not only the operation time required for the search unit 150 but also the operation time required for the determination unit 170.

In order to shorten the time required for the operation of the determination unit 170, the search unit 150 may extract a suspicious region including the singularity from the medical image when the singularity is detected. In this case, the determination unit 170 may determine the presence or absence of a medical abnormality in the suspicious area using the diagnostic model acquired by machine learning. The determination unit 170 may reduce a processing load and a processing time by limiting only the suspicious region in the medical image as a determination object.

When the singular point is searched, the search unit 150 may extract a patch p corresponding to the singular point s and a peripheral boundary line o connected to the singular point s from the entire boundary line of the organ.

4 is a photograph showing a patch.

The patch p may include image data including a singular point s and a peripheral boundary line o connected to the singular point s. The patch p may be formed in a plane or three-dimensionally. The three-dimensionally formed patch p may be formed in a polyhedral shape such as a hexahedron such as a cubic.

The determination unit 170 may exclude the medical image and the entire boundary line of the organ, and determine whether there is an abnormality in the organ using only the patch p.

5 is a schematic diagram showing the operation of the extraction unit 130.

In order to search for a singular point, the search unit 150 needs to grasp a boundary line of an organ included in the medical image. In this case, the extraction unit 130 may be used to determine the boundary line.

The extraction unit 130 may extract only a boundary line of an organ from the medical image obtained by the acquisition unit 110 and generate a boundary image i1 including only the boundary line. In this case, the search unit 150 may search for a singular point using the boundary image i1 instead of the medical image. Since the search unit 150 only needs to analyze the boundary image i1 having a small amount of data instead of the original medical image, processing time may be shortened.

As shown in (b) of FIG. 5, the extraction unit 130 uses a medical image to mask the first value m1 on the edge of the organ and the second value on the remaining area. Can be extracted. For example, in the binary mask, the first value may be '1' and may be in a transparent state (indicated in white in the drawing). In the binary mask, the second value may be '0' and may be in an opaque state (shown in black in the drawing).

The extraction unit 130 may extract the boundary line of the set thickness using the first binary mask m1. For example, the extraction unit 130 may extract the boundary line through window processing in which the first binary mask m1 is applied to the medical image. A boundary image i1 including only the boundary line may be obtained through the corresponding window processing. The extraction unit 130 may transmit the extracted boundary line or boundary image i1 to the search unit 150. The search unit 150 may perform a gradient operation for searching for a singular point on the boundary line or boundary image i1 received from the extraction unit 130.

6 is a block diagram showing another diagnostic device of the present invention.

The diagnostic apparatus illustrated in FIG. 6 may include a pretreatment unit 10, a first diagnostic unit 100, a second diagnostic unit 200, and an integrated unit 30.

When a medical image such as a chest CT image is obtained, the preprocessing unit 10 may resize the medical image to a set size or perform image normalization.

The first diagnostic unit 100 may correspond to the diagnostic device illustrated in FIG. 1. The first diagnosis unit 100 may diagnose a pleural nodule using a medical image of the chest.

The first diagnostic unit 100 may extract a first binary mask m1 for an edge of the lung. On the medical image, the edge of the lung may be the boundary of the lung.

The first diagnostic unit 100 may extract the boundary line of the lung using the first binary mask m1.

The first diagnostic unit 100 may calculate gradients of a boundary line.

The first diagnostic unit 100 may extract a first patch of a three-dimensional polyhedron including a protrusion suspected of being a pleural nodule using the calculation result of the gradient.

The first diagnosis unit 100 may diagnose a pleural nodule using the first patch.

The second diagnosis unit 200 may diagnose a lung nodule using a medical image.

The second diagnostic unit 200 may extract a second binary mask m2 from which the inside of the lung and the outside of the lung are divided. For example, the second diagnostic unit 200 may generate a second binary mask m2 in which the inside of the lung is treated as '1' transparent and the outside of the lung is treated as opaque '0' through image processing of the medical image. I can.

The second diagnostic unit 200 may extract the entire inner region of the lung using the second binary mask m2. As an example, the second diagnosis unit 200 may generate an area image i2 including only image data for an inner area of the lung through window processing in which a second binary mask m2 is applied to the medical image.

The second diagnostic unit 200 may divide the inner region of the lung into second patches of a 3D polyhedron having a set size using the region image i2.

The second diagnostic unit 200 may diagnose a lung nodule using the second patch.

As illustrated in FIG. 2, the second diagnosis unit 200 may search and diagnose various circular nodules with high accuracy. The second diagnostic unit 200 may be replaced with an existing pulmonary nodule diagnostic equipment. However, the second diagnosis unit 200 has a problem in that it is difficult to diagnose a pleural nodule formed by being attached to the pleura. According to the present invention, the pleural nodule can be diagnosed by the first diagnostic unit 100.

The integrated unit 30 may generate a final diagnosis result in which the diagnosis result of the pleural nodule and the diagnosis result of the pulmonary nodule are integrated. The final diagnosis result may correspond to the diagnosis result for all pulmonary nodules-related feces including pleural nodules, and can be of great help to medical personnel.

According to the present invention, a diagnostic method for calculating a gradient targeting a boundary line of an organ included in a medical image and searching for a singular point in which a change amount or rate of change of the gradient satisfies a set value may be provided. The corresponding diagnosis method may extract a suspicious area in which a singularity is detected among the boundary lines, and diagnose whether a nodule has occurred based on machine learning for only the suspicious area. 7 is a detailed illustration of the diagnostic method of the present invention.

7 is a flow chart showing a diagnosis method of the present invention.

First, a chest medical image such as computed tomography (CT) may be resized or normalized to a set size (S510). This operation may be performed by the preprocessing unit 10.

When the pre-processed medical image of the chest is input, a first binary mask m1 for dividing the lung boundary line may be extracted through image processing. The lung boundary line o having a set thickness may be extracted using the first binary mask m1 (S520). As an operation performed by the extraction unit 130, a boundary image i1 remaining only the lung boundary line o in the medical image may be generated.

A 3D gradient can be calculated for the lung boundary line o. As an operation performed by the search unit 150, a histogram of gradients (HOG) may be used (S530). The search unit 150 may search for a portion in which the change in the 3D gradient satisfies the set condition by using the fact that the change value of the gradient for the pleural nodule portion satisfies the set condition (S540). The search unit 150 may extract the searched part (suspect area) in the form of a 3D cube patch (S550).

Using a pleural nodule classification model derived through machine learning targeting only pleural nodules, it is possible to diagnose whether a 3D cube patch has a pleural nodule (S560). It may be an operation performed by the determination unit 170. The determination unit 170 may diagnose whether or not a pleural nodule is detected using only the patch received from the search unit 150 without targeting the entire boundary line.

According to the diagnostic method of the present invention, the detection performance of a pleural nodule, which is difficult to detect, can be improved. In addition, it is possible to detect pulmonary nodules with low cost and high efficiency through a cost-effective detection algorithm related to the time required for diagnosis of pleural nodules and memory usage.

8 is a diagram illustrating a computing device according to an embodiment of the present invention. The computing device TN100 of FIG. 8 may be a device (eg, a diagnostic device or a first diagnostic unit) described herein.

In the embodiment of FIG. 8, the computing device TN100 may include at least one processor TN110, a transmission/reception device TN120, and a memory TN130. In addition, the computing device TN100 may further include a storage device TN140, an input interface device TN150, an output interface device TN160, and the like. Components included in the computing device TN100 may be connected by a bus TN170 to communicate with each other.

The processor TN110 may execute a program command stored in at least one of the memory TN130 and the storage device TN140. The processor TN110 may mean a central processing unit (CPU), a graphics processing unit (GPU), or a dedicated processor on which methods according to an embodiment of the present invention are performed. The processor TN110 may be configured to implement procedures, functions, and methods described in connection with an embodiment of the present invention. The processor TN110 may control each component of the computing device TN100.

Each of the memory TN130 and the storage device TN140 may store various information related to an operation of the processor TN110. Each of the memory TN130 and the storage device TN140 may be configured with at least one of a volatile storage medium and a nonvolatile storage medium. For example, the memory TN130 may be composed of at least one of a read only memory (ROM) and a random access memory (RAM).

The transmission/reception device TN120 may transmit or receive a wired signal or a wireless signal. The transmission/reception device TN120 may be connected to a network to perform communication.

Meanwhile, the embodiment of the present invention is not implemented only through the apparatus and/or method described so far, but may be implemented through a program that realizes a function corresponding to the configuration of the embodiment of the present invention or a recording medium in which the program is recorded. In addition, such an implementation can be easily implemented by a person skilled in the art from the description of the above-described embodiments.

Although the embodiments of the present invention have been described in detail above, the scope of the present invention is not limited thereto, and various modifications and improvements of the skilled person using the basic concept of the present invention defined in the following claims are also present. It is within the scope of the invention.

10... pretreatment unit 30... integrated unit
100...1st diagnostic unit 110...acquisition section
130...extracting section 150...finding section
170... judgment part 200... second diagnostic unit

Claims (12)

An acquisition unit that acquires a medical image photographing organs inside the body;
A search unit for searching for a singular point corresponding to a protrusion continuously protruding from a boundary line of the organ included in the medical image or a groove continuously depressed from the boundary line;
Diagnostic device comprising a.
The method of claim 1,
An extraction unit for extracting only the boundary line of the organ from the medical image obtained by the acquisition unit and generating a boundary image including only the boundary line is provided,
The diagnostic apparatus for the search unit to search for the singular point using the boundary image instead of the medical image.
The method of claim 1,
An extraction unit is provided for extracting a binary mask in which a first value is masked on an edge of the organ and a second value is masked in the remaining area using the medical image,
The extraction unit extracts the boundary line of a set thickness using the binary mask, and transmits the extracted boundary line to the search unit.
The method of claim 1,
The search unit calculates 3D gradients for the boundary line,
The search unit is a diagnostic device for selecting, as the singularity, a portion having a change rate of a gradient that differs by more than a set value than a change rate of a gradient of a peripheral boundary line.
The method of claim 1,
The acquisition unit acquires a pleural image,
The search unit calculates a three-dimensional gradient targeting the boundary line of the pleura, and uses the gradient to search for a juxta-pleural nodule protruding from the boundary line of the pleura in a hemispherical shape.
The method of claim 1,
When the search unit detects the singularity, extracts a suspicious region including the singularity from the medical image,
A diagnostic device provided with a determination unit for determining the presence or absence of a medical abnormality in the suspicious region using a diagnostic model obtained by machine learning.
The method of claim 1,
When the singular point is detected, the search unit extracts a patch corresponding to a partial boundary line including the singular point from the entire boundary line of the organ,
A diagnosis apparatus including a determination unit configured to exclude the medical image and the entire boundary line of the organ, and to determine whether or not the organ is abnormal using only the patch.
A first diagnostic unit for diagnosing a pleural nodule using a medical image of the chest;
A second diagnostic unit for diagnosing a lung nodule using the medical image;
And an integrated unit that integrates the diagnosis result of the first diagnosis unit and the diagnosis result of the second diagnosis unit,
The first diagnostic unit,
A diagnostic apparatus for diagnosing the pleural nodule by searching for a singular point corresponding to a protrusion continuously protruding from the boundary line of the organ included in the medical image or a groove continuously depressed from the boundary line.
The method of claim 8,
The first diagnostic unit,
Extracting a binary mask for the edge of the lung,
Extracting the boundary line of the lung using the binary mask,
Calculate the gradients of the boundary line,
Extracting a first patch of a three-dimensional polyhedron containing a protrusion suspected as the pleural nodule using the calculation result of the gradient,
Diagnosing the pleural nodule using the first patch,

The second diagnostic unit,
Extracting a binary mask that separates the inside of the lung and the outside of the lung,
Extracting the entire inner region of the lung using the binary mask,
Dividing the inner region of the lung into a second patch of a three-dimensional polyhedron of a set size,
Diagnosing the lung nodule using the second patch,

The integrated unit is a diagnostic device for generating a final diagnosis result in which the diagnosis result of the pleural nodule and the diagnosis result of the pulmonary nodule are integrated.
A diagnostic method performed by a diagnostic device for searching for a singular point corresponding to a protrusion continuously protruding from a boundary line of an organ included in a medical image or a groove continuously depressed from the boundary line,
A diagnostic method in which a gradient is calculated for a boundary line of an organ included in a medical image, and a singular point in which the change amount or rate of change of the gradient satisfies a set value on the boundary line.
The method of claim 10,
A diagnostic method of extracting a suspicious region in which the singularity is searched among the boundary lines, and diagnosing a nodule based on machine learning for only the suspicious region.
A diagnostic method performed by a diagnostic device for searching for a singular point corresponding to a protrusion continuously protruding from a boundary line of an organ included in a medical image or a groove continuously depressed from the boundary line,
When a medical image of the chest is input, extracting a binary mask from which a lung boundary line is divided through image processing;
Extracting a closed boundary line of a set thickness using the binary mask;
Calculating a 3D gradient for the lung boundary line;
Searching for a portion in which the change in the 3D gradient satisfies the set condition by using the fact that the change value of the gradient for the pleural nodule portion satisfies the set condition;
Extracting the searched part in the form of a 3D cube patch;
Diagnosing whether the 3D cube patch has a pleural nodule using a pleural nodule classification model derived through machine learning targeting only the pleural nodule;
Diagnosis method comprising a.

Images