A method and system for detecting lesions in mammography

Abstract

The application provides a breast molybdenum target lesion detection method and system, the method comprises the following steps: acquiring historical molybdenum target image data, and performing molybdenum target labeling on the historical molybdenum target image data; selecting a plurality of windows according to window values of the image, and performing normalization processing on the selected windows as the input of a training model; performing unsupervised training on the input unlabeled image data set to obtain an intermediate representation state, training the input labeled image data set, adjusting the information of the intermediate representation state, and obtaining the training model; inputting the normalized molybdenum target image data to be detected into the training model to obtain the detection result of the molybdenum target lesion. The application uses a large amount of clinical data for training to obtain a prediction model, which is used for detecting and analyzing the breast molybdenum target lesion to obtain detailed detection results of various lesions, and the molybdenum target lesion can be quickly and accurately detected, and the diagnosis is more professional and fast.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, and in particular to a breast molybdenum target lesion detection method and system. BACKGROUND

[0002] The World Health Organization proposes that 1 / 3 of cancers are completely preventable, 1 / 3 of cancers can be cured through early detection, and 1 / 3 of cancers can prolong life, reduce pain and improve quality of life through existing medical measures. Since two-cancer screening is an effective means of preventing and controlling breast cancer and cervical cancer, it is necessary to conduct early female two-cancer screening to effectively control cervical cancer and breast cancer.

[0003] Among them, breast molybdenum target is the most common method for breast cancer screening. Hospitals have the following problems through breast molybdenum target screening: there is a severe shortage of experienced breast molybdenum target diagnosis experts, and doctors in primary medical institutions are even more rare, especially primary hospitals lack professional ability for breast molybdenum target diagnosis, and often upload the molybdenum target to the superior hospital, which needs more professional doctors to diagnose and analyze. This will waste a lot of time, and patients cannot quickly get the test results; there is a severe shortage of experienced breast molybdenum target diagnosis experts, and doctors in primary medical institutions are even more rare, especially primary hospitals lack professional ability for breast molybdenum target diagnosis, and often upload the molybdenum target to the superior hospital, which needs more professional doctors to diagnose and analyze. This will waste a lot of time, and patients cannot quickly get the test results. SUMMARY

[0004] The present application provides a breast molybdenum target lesion detection method and system, which is used to solve the problem of low detection efficiency.

[0005] To achieve the above-mentioned purpose, the technical scheme adopted by the present application is as follows:

[0006] The present application provides a breast molybdenum target lesion detection method, which comprises the following steps:

[0007] Obtain historical molybdenum target image data, and perform molybdenum target labeling on the historical molybdenum target image data;

[0008] Select a plurality of windows according to the window value of the image, and normalize the selected windows as the input of the training model;

[0009] Unsupervised training is performed on the input unlabeled image data set to obtain an intermediate representation state, and the input labeled image data set is trained to adjust the information of the intermediate representation state to obtain a training model;

[0010] The molybdenum target image data to be detected is input into the training model after normalization to obtain the detection result of the molybdenum target lesion.

[0011] Further, the selection of the window is specifically:

[0012] The window value [0, 3500] is selected as a global window, the window value [500, 3500] is selected as a calcification window, and the window value [0, 3000] is selected as a mass window.

[0013] Further, the molybdenum target lesion includes benign calcification, mass, mass with calcification, suspicious calcification, lymph node enlargement, asymmetric density, asymmetric density with calcification, structure distortion, structure distortion with calcification, skin thickening, skin contraction, and nipple depression.

[0014] Further, the method further comprises classifying the benign calcification shape present in the detection result, and the specific process is:

[0015] A benign calcification shape data set is constructed, and the benign calcification shape in the data set is labeled;

[0016] The labeled image is normalized to a natural image, and each image forms a 3-channel image as an input of a training model;

[0017] A ResNet50 network with increased hole convolution is used for training to obtain a benign calcification classification prediction model;

[0018] Based on the benign calcification classification prediction model, the classification of the benign calcification shape in the to-be-detected image is predicted.

[0019] Further, the ResNet50 network with increased hole convolution is specifically:

[0020] A 3*3 hole convolution is respectively used for convolution on the output branch of Stage1, Stage2, and Stage3 of ResNet50;

[0021] A 1*1 convolution is performed on the output branch of Stage4 of ResNet50;

[0022] After the above two-step convolution, four convolution results are obtained, and the four convolution branches are fused through an SPP structure to obtain a final picture feature representation.

[0023] Further, the method further comprises analyzing the suspicious calcification distribution and shape present in the detection result, and the specific process is:

[0024] A suspicious calcification distribution and shape data set is constructed, and the suspicious calcification shape is labeled in the data set;

[0025] The labeled image is normalized to a natural image, and each image forms a 3-channel image as an input of a training model;

[0026] The ResNet50 network with increased hole convolution is used for training to obtain a suspicious calcification distribution and shape prediction model;

[0027] Based on the suspicious calcification distribution and shape prediction model, the suspicious calcification shape classification in the to-be-detected image is predicted.

[0028] Further, the method further comprises analyzing the mass existing in the detection result, predicting the mass shape, density and edge, and the specific process is as follows:

[0029] A mass shape, density and edge dataset is constructed, and the shape and edge of the mass in the dataset are labeled;

[0030] The labeled image is normalized into a natural image, and each image forms a 3-channel image as the input of the training model;

[0031] The ResNet50 network with increased hole convolution is used for training to obtain a mass shape and edge prediction model;

[0032] Based on the mass shape and edge prediction model, the mass shape, edge and density in the to-be-detected image are predicted.

[0033] Further, the method further comprises classifying the gland type and change mode existing in the detection result, and the specific process is as follows:

[0034] A gland type and change mode dataset is constructed, and the gland type and gland change mode of the LCC and RCC two axial images in the dataset are labeled;

[0035] The labeled image is normalized into a natural image, and each image forms a 3-channel image as the input of the training model;

[0036] The ResNet50 network with increased hole convolution is used for training to obtain a gland type and change mode prediction model;

[0037] Based on the gland type and change mode prediction model, the gland type and change mode in the to-be-detected image are predicted.

[0038] The second aspect of the present application provides a breast molybdenum target lesion detection system, the system comprises:

[0039] A molybdenum target labeling unit is used to obtain historical molybdenum target image data, and the historical molybdenum target image data is labeled by molybdenum target;

[0040] An image processing unit is used to select a plurality of windows according to the window value of the image, and the selected windows are normalized as the input of the training model;

[0041] a model training unit configured to perform unsupervised training on an input unlabeled image data set to obtain an intermediate representation state, and to train on an input labeled image data set to adjust information of the intermediate representation state to obtain a trained model;

[0042] a lesion prediction unit configured to input normalized molybdenum target image data to be detected into the trained model to obtain a detection result of molybdenum target lesions.

[0043] Further, the system further comprises:

[0044] a benign calcification analysis unit configured to classify benign calcification shapes in the detection result;

[0045] a suspicious calcification analysis unit configured to detect suspicious calcification distribution and shapes in the detection result;

[0046] a mass analysis unit configured to analyze mass shapes, densities and edges in the detection result;

[0047] a gland analysis unit configured to classify gland types and gland change modes in the detection result.

[0048] The detection system of the second aspect of the present application can implement the method of the first aspect and the method of each implementation manner of the first aspect, and achieve the same effects.

[0049] The effects provided in the summary are only the effects of the embodiments, not all the effects of the present application. One of the technical solutions has the following advantages or beneficial effects:

[0050] The present application uses a large amount of clinical data for training to obtain a prediction model for detecting and analyzing breast molybdenum target lesions, and further analyzes the detection results based on specific lesion types to obtain detailed detection results of various lesions, which can quickly and accurately detect lesions in the molybdenum target and form a professional structured report. Hospital doctors, especially primary hospital doctors, can quickly complete breast molybdenum target diagnosis based on intelligent diagnosis results, and the diagnosis is more professional and fast. BRIEF DESCRIPTION OF DRAWINGS

[0051] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, for those skilled in the art, other drawings can also be obtained without creative labor based on these drawings.

[0052] Figure 1 is a flowchart of the method embodiment of the present application;

[0053] Figure 2 is a structural schematic diagram of an embodiment of the system of the present application. DETAILED DESCRIPTION

[0054] In order to clearly illustrate the technical features of the present application, the present application will be described in detail below with specific embodiments and in conjunction with the accompanying drawings. The following disclosure provides many different embodiments or examples for implementing the various structures of the present application. In order to simplify the disclosure of the present application, the components and settings of specific examples are described below. In addition, the present application can repeatedly refer to numerals and / or letters in different examples. Such repetition is for the purpose of simplification and clarity, and does not in itself indicate the relationship between the various embodiments and / or settings being discussed. It should be noted that the components illustrated in the drawings are not necessarily drawn to scale. The present application omits the description of well-known components and processing techniques and processes to avoid unnecessarily limiting the present application.

[0055] As shown in Figure 1 , the embodiment of the present application provides a breast molybdenum target lesion detection method, which comprises the following steps:

[0056] S1, acquiring historical molybdenum target image data, and performing molybdenum target labeling on the historical molybdenum target image data;

[0057] S2, selecting a plurality of windows according to window values for the image, and performing normalization processing on the selected windows as input of a training model;

[0058] S3, performing unsupervised training on the input unlabeled image data set to obtain an intermediate representation state, training on the input labeled image data set, and adjusting the information of the intermediate representation state to obtain a training model;

[0059] S4, inputting the normalized molybdenum target image data to be detected into the training model to obtain a detection result of the molybdenum target lesion.

[0060] The molybdenum target lesion includes 12 kinds of lesions, i.e. benign calcification, mass, mass with calcification, suspicious calcification, lymph node enlargement, asymmetric density, asymmetric density with calcification, structure distortion, structure distortion with calcification, skin thickening, skin contraction and nipple depression.

[0061] In step S1, molybdenum target image data and molybdenum target reports of the last five years are collected from multiple first-class hospitals as historical molybdenum target image data, and each molybdenum target is labeled by a director-level physician with rich experience. The labeled data is submitted to another director-level physician of a first-class hospital for review, and when both experts have no objection to the labeling result, it is saved as the final labeling result; for the results with objections, a third expert is invited to review and the majority opinion is adopted.

[0062] Step S2 is the preprocessing process of the image. The images with labels are selected according to HU values, respectively, [0, 3500] (under this window value, 12 kinds of molybdenum target lesions are clearly visible, hereinafter referred to as a global window), [500, 3500] (under this window value, the molybdenum target lesions related to calcification are clearly visible, hereinafter referred to as a calcification window), [0, 3000] (under this window value, the molybdenum target lesions related to a lump are clearly visible, hereinafter referred to as a lump window), and the molybdenum target image is normalized to the range of [0-255] to convert a natural image according to the global window, the calcification window and the lump window. Finally, three images are synthesized into a 3-channel image and normalized to serve as the input of the model.

[0063] Step S3 is network training. The normalized image (2048*2048*3) is input into the improved CenterNet2 for training. If the width and height of the input image are not 2048, the image is adjusted to have a width and height of 2048 through a bilinear difference algorithm. At the same time, a variety of data enhancement methods are used in the training process, such as rotation, horizontal flip, channel flip, multiple mosaic, multi-scale resampling and the like, to increase the number of samples and improve the robustness of the algorithm.

[0064] The improved CenterNet2 is a two-stage training. The first stage is unsupervised training (pre-training) using unlabeled data, and the second stage is normal CenterNet2 training using labeled data.

[0065] The first stage trains the model into an intermediate representation state, and the intermediate representation state represents the overall information of the lesion. The second stage fine-tunes the model using a labeled data set to train the model to a final state. The training based on the first stage can achieve good results on the basis of a small amount of labeled data set.

[0066] The process of the first stage unsupervised training is as follows: for any input image (original picture), data enhancement is performed on the image to obtain two enhanced pictures. The enhanced pictures are input into an encoder, and two encoders are shared parameters to obtain an intermediate representation. The final intermediate representation of the image is obtained through a projection head.

[0067] In one implementation manner of the method embodiment, the method further includes classifying benign calcification shapes existing in the detection result, wherein the benign calcification shapes are classified into ten kinds, respectively, as round calcification, point calcification, vascular calcification, center transparency calcification, calcified milk calcification, large rod calcification, coarse or popcorn calcification, ring calcification, dystrophic calcification and skin calcification. The specific process of the implementation manner is as follows:

[0068] A benign calcification shape data set is constructed, and the benign calcification shape in the data set is labeled; the labeled data is submitted to another chief physician of a third-grade class-A hospital for review, and when two experts have no objection to the labeling result, the labeling result is saved as the final labeling result; and for the result with objection, a third expert is invited to review, and the majority opinion is used as the criterion.

[0069] The labeled images are selected according to HU values respectively [500, 3500], the benign calcification lesion images are normalized to [0-255] range to be converted into natural images, and each image forms a 3-channel image as an input of the training model;

[0070] The normalized image (80*80*3) is input into the improved ResNet50 for training, and if the input image width and height are not 80, the image is adjusted to have a width and a height of 80 through a bilinear difference algorithm. Meanwhile, in the training process, various data enhancement methods such as rotation, horizontal flip, channel flip, etc. are used to increase the sample quantity and improve the algorithm robustness. A benign calcification classification prediction model is obtained.

[0071] Based on the benign calcification classification prediction model, the benign calcification shape classification in the to-be-detected image is predicted, and the prediction category is labeled in the original image to facilitate doctor review.

[0072] The ResNet50 network with the increased hole convolution is specifically as follows:

[0073] 3*3 hole convolution is respectively performed on the output branches of Stage1, Stage2 and Stage3 of ResNet50; 1*1 convolution is performed on the output branch of Stage4 of ResNet50; four convolution results are obtained after the above two convolution steps, the four convolution branches are fused through an SPP structure, and a final picture feature representation mode is obtained.

[0074] In one implementation manner of the method embodiment, the method further includes analyzing suspicious calcification distribution and shape existing in the detection result, and the specific process is as follows:

[0075] A suspicious calcification distribution and shape data set is constructed, and suspicious calcification shape is labeled in the data set; the labeled data is submitted to another chief physician of a third-grade class-A hospital for review, and when two experts have no objection to the labeling result, the labeling result is saved as the final labeling result; and for the result with objection, a third expert is invited to review, and the majority opinion is used as the criterion.

[0076] The labeled images are selected according to HU values in [500, 3500] respectively, the benign calcification lesion images are normalized to [0-255] range to be converted into natural images, and each image forms a 3-channel image as the input of the training model;

[0077] The normalized image (300*300*3) is trained by using a ResNet50 network with increased hole convolution, if the input image width and height are not 300, the image is adjusted to have a width and height of 300 by using a bilinear difference algorithm. Meanwhile, a plurality of data enhancement methods are used in the training process, such as rotation, horizontal flip, channel flip, etc. to increase the sample quantity and improve the algorithm robustness. A suspicious calcification distribution and shape prediction model is obtained;

[0078] Based on the suspicious calcification distribution and shape prediction model, the suspicious calcification shape classification in the to-be-detected image is predicted.

[0079] The suspicious calcification distribution includes six kinds of distribution, i.e. cluster distribution, cluster distribution, regional distribution, scattered distribution, linear distribution and segment distribution. The suspicious calcification shape includes seventeen kinds, i.e. round calcification, point calcification, vascular calcification, central transparency calcification, calcified milk calcification, large rod calcification, coarse or popcorn calcification, ring calcification, dystrophic calcification, skin calcification, coarse and uneven calcification, small and polymorphic calcification, fine line calcification, sand calcification, sand calcification and multiple morphological calcification.

[0080] In one implementation manner of the method embodiment, the method further includes analyzing the tumor in the detection result, predicting the shape, density and edge of the tumor, and the specific process is as follows:

[0081] A tumor shape, density and edge data set is constructed, the shape and edge of the tumor in the data set are labeled, the labeled data is submitted to another chief physician of a third-grade class-A hospital for auditing, when two experts have no objection to the labeling result, the labeling result is saved as the final labeling result, and the result with objection is submitted to a third expert for auditing, and the majority opinion is used as the criterion.

[0082] The labeled images are selected according to HU values in [0, 3000] respectively, the benign calcification lesion images are normalized to [0-255] range to be converted into natural images, and each image forms a 3-channel image as the input of the training model;

[0083] The normalized image (400*400*3) is trained by using a ResNet50 network with increased hole convolution, and if the width and height of the input image are not 400, the image is adjusted to have a width and height of 400 by using a bilinear difference value algorithm.

[0084] Based on the prediction model of the shape and edge of the tumor, the shape, edge and density of the tumor in the to-be-detected image are predicted, and the prediction categories are marked in the original image to facilitate doctor review.

[0085] The shape of the tumor includes three types of circular, oval and irregular. The edge of the tumor includes five types of clear, shield, fuzzy, microlobulated and irregular.

[0086] In one implementation mode of the embodiment of the application, the method further includes classifying the types and change modes of the detected glands, and the specific process is as follows:

[0087] The gland type and change mode dataset is constructed, and the types and change modes of the glands in the LCC and RCC two-axis position images in the dataset are labeled.

[0088] The labeled images are selected according to HU values respectively, [0, 3500], the benign calcification lesion images are normalized to [0-255] range to be converted into natural images, each image forms a 3-channel image, and the image is used as the input of the training model.

[0089] The normalized image (512*512*3) is trained by using a ResNet50 network with increased hole convolution, and if the width and height of the input image are not 512, the image is adjusted to have a width and height of 512 by using a bilinear difference value algorithm.

[0090] Based on the prediction model of the shape and edge of the tumor, the shape, edge and density of the tumor in the to-be-detected image are predicted, and the prediction categories are marked in the original image to facilitate doctor review.

[0091] As Figure 2As shown, an embodiment of the present invention further provides a breast mammography lesion detection system, which includes a mammography labeling unit 1, an image processing unit 2, a model training unit 3 and a lesion prediction unit 4.

[0092] The molybdenum target annotation unit 1 is used to obtain historical molybdenum target image data and perform molybdenum target annotation on the historical molybdenum target image data; the image processing unit 2 is used to select several windows from the image according to the window value, and normalize the selected windows as the input of the training model; the model training unit 3 is used to perform unsupervised training on the input unlabeled image data set to obtain an intermediate representation state, train the input labeled image data set, adjust the information of the intermediate representation state, and obtain the training model; the lesion prediction unit 4 is used to normalize the molybdenum target image data to be detected and input it into the training model to obtain the detection result of the molybdenum target lesion.

[0093] The system also includes a benign calcification analysis unit, a suspicious calcification analysis unit, a mass analysis unit, and a gland analysis unit. The benign calcification analysis unit is used to classify the shape of benign calcifications in the test results; the suspicious calcification analysis unit is used to detect the distribution and shape of suspicious calcifications in the test results; the mass analysis unit is used to analyze the shape, density, and edge of the mass in the test results; and the gland analysis unit is used to classify the gland type and gland change pattern in the test results.

[0094] Although the above describes the specific embodiments of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without any creative work are still within the scope of protection of the present invention.

Claims

1. A method for detecting mammary gland lesions by mammography, characterized in that: The method comprises the following steps: Acquiring historical mammography image data, and performing mammography labeling on the historical mammography image data; Several windows were selected from the image according to the window value, and the selected windows were normalized and used as the input of the training model. Specifically, the labeled image was selected according to the HU value as [0, 3500], [500, 3500], and [0, 3000] as the global window, calcification window, and mass window, respectively. The mammography image was normalized to the range of [0-255] according to the global window, calcification window, and mass window to convert it into a natural image. Finally, the three images were synthesized into a 3-channel image and normalized before being used as the input of the model. Performing unsupervised training on an input unlabeled image dataset to obtain an intermediate representation state, training an input labeled image dataset, adjusting information of the intermediate representation state, and obtaining a training model; Normalizing the mammography image data to be detected and inputting it into the training model to obtain the detection results of the mammography lesions; The specific principles of model training are as follows: The model training is network training. Normalized images are input into the improved CenterNet2 for training. If the input image width and height are not 2048, the image is resized to 2048 in width and height using a bilinear interpolation algorithm. During the training process, rotation, horizontal flipping, channel flipping, multi-mosaicing, and multi-scale resampling are used to increase the number of samples and improve the robustness of the algorithm. The improved CenterNet2 is trained in two stages: the first stage uses unsupervised training with unlabeled data, and the second stage uses normal CenterNet2 training with labeled data. The first stage trains the model into an intermediate representation state that represents the overall information of the lesion; the second stage uses the labeled dataset to fine-tune the model and train it to the final state. The first stage of unsupervised training is as follows: for any input image, data augmentation is performed on it to obtain two enhanced images; the enhanced images are input into the encoder, and the two encoders share parameters to obtain an intermediate representation, and the final intermediate representation of the image is obtained through the projiection head; The mammographic lesions include benign calcification, mass, mass with calcification, suspicious calcification, lymphadenopathy, asymmetric density, asymmetric density with calcification, architectural distortion, architectural distortion with calcification, skin thickening, skin contraction, and nipple retraction. The method further includes classifying the shapes of benign calcifications present in the detection results, and the specific process is as follows: Construct a benign calcification shape dataset and annotate the benign calcification shapes in the dataset; Normalize the labeled images into natural images, and each image forms a 3-channel image as the input of the training model; The ResNet50 network with added hole convolution was used for training to obtain a benign calcification classification prediction model; Based on the benign calcification classification prediction model, predicting the shape classification of benign calcification in the image to be detected; The ResNet50 network with added hole convolution is specifically: Convolution is performed on the output branches of Stage 1, Stage 2, and Stage 3 of ResNet50 using 3*3 dilated convolutions respectively; Perform 1*1 convolution on the output branch of Stage 4 of ResNet50; After the above two steps of convolution, four convolution results are obtained. The four convolution branches are fused through the SPP structure to obtain the final image feature representation.

2. The method for detecting mammary gland lesions according to claim 1, wherein: The method further includes analyzing the distribution and shape of suspicious calcifications in the detection results, and the specific process is as follows: Construct a suspicious calcification distribution and shape dataset, and annotate the suspicious calcification shapes in the dataset; Normalize the labeled images into natural images, and each image forms a 3-channel image as the input of the training model; The ResNet50 network with added hole convolution was used for training to obtain a model for predicting the distribution and shape of suspicious calcifications. Based on the suspicious calcification distribution and shape prediction model, the shape classification of the suspicious calcification in the image to be detected is predicted.

3. The method for detecting mammary gland lesions according to claim 1, wherein: The method further includes analyzing the mass present in the detection result and predicting the mass shape, density and edge, and the specific process is as follows: Construct a dataset of tumor shape, density, and edge, and annotate the shape and edge of the tumors in the dataset; Normalize the labeled images into natural images, and each image forms a 3-channel image as the input of the training model; The ResNet50 network with added hole convolution was used for training to obtain a prediction model of the mass shape and edge; Based on the prediction model of the tumor shape and edge, the shape, edge and density of the tumor in the image to be detected are predicted.

4. The method for detecting mammary gland lesions according to claim 1, wherein: The method further includes classifying the gland types and changes in the test results, specifically by: A dataset of glandular types and changes was constructed, and the glandular types and changes of the two axial images of LCC and RCC in the dataset were annotated. Normalize the labeled images into natural images, and each image forms a 3-channel image as the input of the training model; A ResNet50 network with added atrous convolution was used for training to obtain a prediction model for glandular type and change pattern. Based on the prediction model of the gland type and the change mode, the gland type and the change mode in the image to be detected are predicted.

5. A mammary gland mammography lesion detection system, characterized in that: The system is used to implement the method according to claim 1, and the system includes: A molybdenum target marking unit is used to obtain historical molybdenum target image data and perform molybdenum target marking on the historical molybdenum target image data; An image processing unit is used to select a number of windows from the image according to the window value, and normalize the selected windows as input for the training model; A model training unit is used to perform unsupervised training on an input unlabeled image dataset to obtain an intermediate representation state, and to train an input labeled image dataset, adjust information of the intermediate representation state, and obtain a training model; The lesion prediction unit is used to normalize the mammography image data to be detected and then input it into the training model to obtain the detection result of the mammography lesion.

6. The mammary gland mammography lesion detection system according to claim 5, wherein: The system further comprises: a benign calcification analysis unit, configured to classify the shape of benign calcifications in the detection result; a suspicious calcification analysis unit, configured to detect the distribution and shape of suspicious calcifications in the detection results; a mass analysis unit, configured to analyze the mass shape, density, and edge in the detection result; The gland analysis unit is used to classify the gland types and gland change patterns in the detection results.

Images