MULTIARM ANALYSIS METHOD OF URINE CYTOLOGY IMAGES
Patent Information
- Application Number
- TR202612926
- Authority / Receiving Office
- TR · TR
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2026-07-31
- Publication Date
- 2026-09-21
Smart Images
Figure 00000021_0000
Abstract
Description
1 TARIFF MULTIARM ANALYSIS METHOD OF URINE CYTOLOGY IMAGES TECHNICAL AREA 5 The invention consists of at least one processing unit, at least one memory unit communicating with the said processing unit, and at least... urine by a computer-assisted cytology analysis system with a minimal user interface By processing digital image data of the cytology sample, at the cellular level... It relates to an image analysis method for generating classification outputs. 10 PREVIOUS TECHNIQUE In the field of digital pathology, computer-generated full slide images of urine cytology samples are used. Various image analysis systems are used for supported examination. These 15 In these systems, generally the full slide image is broken down into smaller image segments. identification of cell regions, nuclear and / or cytoplasmic segmentation the process involves extracting cellular characteristics and classifying cells according to the obtained characteristics. or the processes of generating a classification output at the slide level is being carried out. 20 Some of the current technical solutions select only the highest number of cells from a large number of cells. A limited number of cells with a potential for atypia or malignancy are selected and evaluated. This is done through cells. The cells are separated from the original full slide image. It is also known to be presented in cell galleries, compressed grids, or similar structures. 25 However, these approaches imply an implication of the overall distribution of the cell population and the cells themselves. This can cause the spatial relationship between them to be lost, preventing the user from selecting the relevant cell. This can make it difficult to assess the image in its true position on the original. The current systems contain 30 different types of morphological, geometric, color, and intercellular relationships. The characteristics in question are usually found within a single feature set or linear analysis pipeline. They are processed together. This situation determines which features directly influence the classification result. This makes it difficult to determine the effect and the supporting image information This can lead to interference with primary decision data. It can also be reactive or inflammatory. The mixing of altered cells with malignant cells, small and atypical cells 35 overlooked image artifacts affecting classification and analysis results 2 The inability to present the information with sufficient clarity is another technical problem encountered in the current technique. It is among them. In conclusion, all the problems mentioned above necessitate an innovation in the relevant technical field. It has made it mandatory. 5 A BRIEF DESCRIPTION OF THE INVENTION The present invention aims to eliminate the aforementioned disadvantages and to contribute to the relevant technical field. It is related to a method aimed at bringing new advantages. 10 One aim of the invention is to create high-resolution digital images of urine cytology samples. By taking into account the different image qualities of the cells found, a more reliable result can be obtained. The aim is to develop an image analysis method that enables classification. Another aim of the invention is to present the analysis results obtained at the cellular level to the original cells. the full slide image being evaluated without losing its spatial context The aim is to develop an image analysis method that provides this. Another aim of the invention is to provide data that forms the basis for primary cell classification and clinical 20 from the mixing of additional image data supporting the assessment A method of image analysis that reduces potential classification errors has been developed. to place. Another aim of the invention is to provide a large-scale, full-slide image of high-risk cell regions. an image that allows for faster identification and reduces examination time. The goal is to establish an analysis method. Another purpose of the invention is to utilize different sample preparation, staining and imaging conditions. a 30 that reduces the negative effects of resulting variability on image analysis The goal is to develop an image analysis method. All the purposes mentioned above and those that will emerge from the detailed explanation below. The present invention, for the purpose of realizing at least one processing unit, with the aforementioned processing unit 35 computer-assisted systems that include at least one communicative memory unit and at least one user interface Digital image data of the urine cytology sample by the cytology analysis system a method of generating classification outputs at the cellular level through processing 3 It is an image analysis method. Accordingly, the present invention relates to at least one urine cytology sample. the full slide image being captured by the aforementioned processing unit, the aforementioned full slide image quality control, image segment extraction, color normalization, and background. preprocessing should include at least one of the filtering operations. The detection of cell regions on the subjected full slide image and detection 5 the nuclear and cytoplasmic regions of at least some of the obtained cell regions Determination through segmentation of the nucleus / cytoplasm belonging to the mentioned cell regions at least one morphological feature containing the ratio is processed by the aforementioned processing unit. removal in the first analysis arm, relating to at least one boundary of the mentioned cellular regions The aforementioned process 10 requires at least one geometric image feature containing edge and / or corner information. extraction in the second branch of analysis conducted by the unit, obtained from the first branch of analysis The morphological features obtained and the geometric image obtained from the second analysis arm. a combined feature representation is created by combining the features of the aforementioned processing unit. the creation of the aforementioned combined feature representation by the aforementioned processing unit 15 associated clinical classification categories for each cell region identified through processing. generating a cell-level classification score, the aforementioned cell-level classification with location information of the relevant cell regions on the full slide view of their scores by relating them, a spatial heat map is created and the aforementioned spatial the heat map should be presented as a visual output in the user interface, including cytoplasmic overlap, color intensity, intercellular distance, cell distribution frequency, and presence of artifacts 20 At least one of its features is related to the aforementioned combined feature representation and at the cell level. The aforementioned processing unit is not included in the creation of the classification score. The determination of the aforementioned cell level in the third arm of analysis conducted by classification score, spatial heat map and the most important data obtained from the third arm of analysis 25 as distinct output components of a small supporting analysis output This includes presenting it in the aforementioned user interface. Thus, different types of Cellular image features are being evaluated in a controlled manner, and supporting analysis is being performed. This prevents the data from negatively affecting the primary classification result and the cell the interpretation of results at this level within the context of the original image is provided. 30 The characteristic of a possible construction of the invention is that the aforementioned complete slide image can be obtained. Each image segment obtained has coordinates indicating its position on the complete slide image. It is the process of dividing an image into multiple parts that can be associated with information. Thus The processing unit's calculation for a large, full-size slide image is lower than 35. processing with the load and the position of each image segment within the original image This ensures identification. 4 Another possible configuration of the invention involves the aforementioned image components and Full slide view of the cell regions identified in the aforementioned image segments. associating them with coordinate information that indicates their location on the mentioned surface and The coordinate relationship is preserved throughout the image processing stages. Thus, cell 5 The spatial analysis results of the regions are independent of the image fragments. This prevents the loss of context and ensures that the analysis results are presented in their original full slide view. It ensures that it is associated with the correct location. Another possible configuration feature of the invention is the identification of the aforementioned cell regions. This involves identifying candidate cell regions on the full slide image. the first image analysis model providing and identifying candidate cell regions using a second image analysis model that enables verification in two stages This is achieved by rapidly scanning large image areas and identifying cells. The precise verification of candidate regions is ensured together, avoiding false positives and 15 This helps to reduce false negative diagnoses. Another possible configuration of the invention involves the aforementioned first image analysis. the model is a YOLO-based object detection model and the aforementioned second image analysis The model is a Faster R-CNN based object detection model. Thus, the cell candidate is 20 More precise verification of selected candidates through rapid identification of regions. A balanced structure is achieved between them in terms of processing speed and detection accuracy. Another possible configuration of the invention involves the aforementioned segmentation process. Determining the nuclear and cytoplasmic boundaries of the identified cell regions 25 This means that cells are not only seen as a visual field, but also biologically... analysis of the nucleus and cytoplasm along with their significant subregions They are determined separately. Another characteristic of a possible configuration of the invention is that in the first branch of analysis mentioned, 30 Nucleus / cytoplasm ratio and the size and shape of the cell, nucleus and / or cytoplasm. It is the identification of at least one of the characteristics as a morphological feature. Thus, cellular Formal and dimensional characteristics of structures that are significant for malignancy assessment It is represented in the form of numerical data. 35 Another characteristic of a possible configuration of the invention is that the cell is located in the second analytical arm mentioned. At least edge detection and corner detection procedures at the nuclear and / or cytoplasmic boundary. by implementing at least one edge and / or corner-based image feature This involves removing the structural and geometric characteristics of cellular boundaries, thus preserving their morphological properties. It is defined as a group of image features distinct from the features and between the cells. The level of distinctiveness is being increased. Another possible configuration of the invention is characterized by the results obtained from the first branch of analysis. Morphological features and edge and / or corner-based data obtained from the second arm of analysis. By combining image features, the aforementioned composite feature representation is a composite feature. It is created as a vector. Thus, it is obtained from different analytical sources. A common data structure that can be processed by machines, containing complementary cellular features. 10 It ensures that it is represented within it. Another possible configuration of the invention involves the aforementioned combined feature vector. each by being processed by at least one machine learning and / or deep learning model 15 is the cell-level classification score representing the risk of malignancy for that cell region. This involves the production of the morphological and geometric characteristics of each cell region together. By evaluating the situation, a quantitative and comparable risk output is generated. Another possible configuration of the invention is characterized by the aforementioned cell-level classification. A negative score, atypical urothelial cells, suspicion of high-grade urothelial carcinoma, and 20 It is associated with at least one of the high-grade urothelial carcinoma categories. Thus, numerical classification outputs are used in clinical evaluation of cellular processes. The information is presented in a way that is consistent with the categories. Another possible configuration of the invention features 25 in the aforementioned third branch of analysis. cytoplasmic overlap, color intensity, intercellular distance, cell distribution frequency, and The presence of an artifact requires the identification of at least one of its characteristics. Thus, primary classification is established. In addition to the output, it also relates to cellular proximity, distribution, staining, and image quality. Supporting data is being obtained. Another possible configuration of the invention is characterized by the information obtained from the third branch of analysis. the aforementioned combined feature representation and cell-level classification of analysis outputs as a supporting analysis output without being included in the calculation of the score The goal is to create a primary classification based on the characteristics and clinical features that form the basis of this classification. Additional features supporting interpretation are separated from each other and the supporting data is 35 This prevents the creation of noise or bias as a result of the classification process. 6 Another possible configuration of the invention involves the aforementioned spatial heat map. Coordinate information of the relevant cell regions from cell-level classification scores. using the original full slide image to project back onto it and visualize the scores. It is created by coding it as such. Thus, it is created at the cell level. 5 The classification results are shown with their actual positions on the original sample image. They are evaluated together. Another possible structural feature of the invention is that it is more integrated with other cellular regions. Cellular regions with high cell-level classification scores exhibit the aforementioned spatial thermal properties. This means that the 10 high-risk areas are visually distinguishable on the map. This allows the user to more quickly identify areas where cells are concentrated and provides a complete slide. This helps reduce the triage time required for image examination. Another possible configuration feature of the invention is that the aforementioned user interface is fully functional. slide view, spatial heat map, cell-level classification scores, cell locations, 15 classification confidence values and supporting analysis obtained from the third arm of analysis. This involves presenting at least two of the outputs. Thus, the user is given only a final... instead of the classification result, what constitutes the result and helps interpret the result. Multiple related image and data outputs are provided. Another possible configuration of the invention features the aforementioned full slide image in front. from sample preparation and / or staining conditions during processing color normalization is applied in order to reduce the resulting color differences. This is the retention of information. Thus, it prevents problems that occur between different samples or imaging processes. The negative impact of possible color variability on cell detection and classification 25 is being reduced. Another possible configuration of the invention is characterized by the aforementioned cell-level classification. their scores, spatial heat map, and supporting data from the third arm of analysis 30 analysis outputs as distinct output components in a clinical report This involves presenting primary classification outputs, spatial representation data, and The supporting analysis results are presented to the user in a structured and interpretable report format. It is being transmitted. BRIEF DESCRIPTION OF THE FIGURE 35 Figure 1 shows a representative view of the system. 7 DETAILED DESCRIPTION OF THE INVENTION This detailed explanation of the invention does not merely aim to provide a better understanding of the subject matter. This is explained with examples that will not create a limiting effect. 5 The invention's image analysis method consists of at least one processing unit (110), the aforementioned processing unit (110) containing at least one memory unit (120) and at least one user interface (130) that communicates with (110). Cytology is performed by a computer-assisted cytology analysis system (100). The analysis system (100) processes the digital image data of the urine cytology sample and analyzes the cells 10 classification outputs at the level of spatial imaging outputs and supporting analysis It generates outputs. Processing unit (110), one or more central processing units, graphics processing units, artificial intelligence accelerator, image processing processor, or a processing infrastructure in which these work together 15 This can be done in the following way. The memory unit (120) is controlled by the processing unit (110). computer program relating to the image processing and classification steps performed commands, digital images to be analyzed, coordinate information of the images, trained personnel It stores model parameters, interim analysis results, and final analysis outputs. The user interface (130) allows viewing, reviewing and analysis results. It ensures that it is reported. In the method described in the invention, a complete slide image of a urine cytology sample is produced for cytology. It is taken into the analysis system (100). The full slide image mentioned belongs to the sample. 25 represents the microscopic image field in a high-resolution and digitized form The full slide image is displayed in a pyramidal structure where different magnification levels are stored. This can be in the form of an image structure or at a single resolution level. It can be presented. The processing unit (110) displays the full slide image directly. from a device, a remote data source, a server or memory unit (120) can obtain. 30 The full slide view is processed by the processing unit (110) before cell analysis is performed. It undergoes pre-processing. Pre-processing includes quality control, image component extraction, and color correction. It includes at least one of the following processes: normalization and background filtering. Quality The analyzability of the full slide image is evaluated through the control process, and cell 35 Image distortions that may negatively affect detection are identified. Quality control. 8 This includes image clarity, image contrast, image brightness, and image analysis. It can be assessed whether or not it contains usable cellular content. Processing the entire slideshow directly as a single image requires high processing power and Because memory capacity may be required, the processing unit (110) can display the full slide image more than once. It divides the image into segments. Each segment represents a specific part of the complete slide image. It represents a section. Image segments will either overlap or not overlap with each other. This can be done in this way. Image segment dimensions depend on the image resolution and the method used. according to the image analysis model and the computational capacity of the processing unit (110) It can be determined. 10 To ensure that the spatial relationship between the image segments and the complete slide image is not lost, The coordinates of each image element specify its position on the complete slide image. It is associated with the information. The aforementioned coordinate information is the complete slide of the image segment. the starting position on the image, the size of the image segment, and the full slide 15 It can include at least one piece of information indicating the relevant resolution level of the image. Cell regions detected in image segments also correspond to the coordinates of the relevant image segment. It is associated with the information. Thus, the cell regions within the image segment A relationship is established between their positions and their positions on the full slide image. Coordinate relationship, image preprocessing, cell detection, cell validation, segmentation, feature It is preserved throughout the extraction and classification processes. In this way, the image Cell-level analysis results obtained from the samples are presented as original full slides at the end of the analysis. It can be re-associated with the relevant cell locations on the image. The aforementioned The structure is created by extracting individual cell images from a full slide view of the cells. 25 This reduces the loss of spatial context resulting from its evaluation in this manner. Color normalization, different sample preparations, and are carried out as part of the pre-processing. It helps to reduce color differences caused by dyeing conditions. Thus, the same or similar cellular structures have different color distributions in different images. 30 The effect of color display on cell detection and classification is reduced. normalization, full slide image or full slide prior to cell detection It can be applied to image fragments obtained from the image. Background filtering removes 35 empty spaces that do not contain cellular structure or contribute to the analysis. Image regions are identified and then excluded from subsequent image processing stages. Background filtering results in only cells, cell clusters, and biological cells being left behind. 9 image identified as containing material or other cytological structure requiring analysis The parts can be transferred to the next stages. Thus, the processing unit (110) is not unnecessary. This prevents the processing of image areas and reduces the computational load. Pre-processed full slide image or 5 derived from full slide image Cellular regions are detected on the image segments. A cellular region has at least one the cell, the nuclear and cytoplasmic structures of a cell, or multiple adjacent cells It refers to the image area containing the cell. Cell detection is performed by the processing unit (110). This is carried out using at least one trained image analysis model. In an application, cell detection is performed in two stages. In the first stage, complete High-sensitivity, fast scanning of slide images or image segments. This is being carried out and candidate regions that are likely to contain cells are being identified. First The aim of this stage is to scan large image areas in a short time and to identify small, atypical or minor images. The aim is to reduce the likelihood of missing distinct cellular regions. 15 In the second stage, a more precise image of the cell candidate regions identified in the first stage is obtained. It is examined with an analysis model. The second image analysis model examines the candidate cell regions. to verify whether it contains a real cellular structure and, in the first stage, as a cell Elimination of regions that do not contain biological cell structure, even though they are marked. 20 This provides the second stage, which also allows for more precise analysis of small and atypical cells. This enables the identification and improvement of confidence values associated with cell regions. In one application, the first image analysis model is a YOLO-based object detection model, the second The image analysis model, on the other hand, is a Faster R-CNN based object detection model. This is being carried out. However, the scope of the invention is limited to the aforementioned model types. not, but rather rapid candidate site identification in the first stage and precise verification in the second stage. Different image analysis models capable of performing this task can also be used. First and second Image analysis models are executed by the processing unit (110) and the models are Weights and other model parameters are stored in the memory unit (120). 30 Nuclear and cytoplasmic regions were found in at least some of the detected cell regions. This is determined through segmentation. The segmentation process involves separating each cell region. the image element with one of the classes of nucleus, cytoplasm or extracellular region It may include the association of the nuclear boundary with the cytoplasm. Segmentation results in the identification of a 35-degree gap between the nuclear boundary and the cytoplasm. The boundaries are separating and the cell is not just in the form of a bounding box, but the cell's They are represented along with their biologically significant underlying structures. The segmentation process uses a trained segmentation model, an image density-based approach. using the method, the boundary demarcation method, or a combination thereof This can be achieved. The resulting nuclear and cytoplasmic masks are then used to analyze the respective cells. This is associated with the coordinate information of the region and used in subsequent feature extraction processes. 5 It is used. Following detection and segmentation procedures, cell regions were analyzed across multiple arms. are being processed. The aforementioned analytical arms are from the same cell region or belonging to the same cell region. It creates different types of image features on segmentation data. Analysis 10 The separation of its arms, different types of features in a single data structure at an early stage. to prevent mixing and to ensure that each feature group is suitable for its own analytical structure. It ensures that it is processed in this way. In the first arm of analysis, morphological characteristics of cell regions are extracted. Morphological 15 Features are the formal and measurable characteristics of a cell, its nucleus, and cytoplasm. It includes. In the first arm of analysis, at least the nucleus / cytoplasm ratio is determined. The nucleus / cytoplasm ratio is the nuclear region determined as a result of segmentation. It represents the relationship between the size of the body and the size of the cytoplasm. In addition to the nucleus / cytoplasm ratio, the cell also measures the size and shape of the nucleus and cytoplasm. At least one of its characteristics can be determined. Dimensional characteristics of the relevant cellular structure. It represents the image area it occupies or the size relationships of cellular structures relative to each other. It is able to do so. The shape characteristics are the geometric boundary of the cell, nucleus, or cytoplasm. It can represent information regarding its structure. The morphological 25 obtained in the first analysis arm. morphological features are a group of morphological characteristics or features associated with each cell region. It is created in the form of a feature vector. In the second arm of analysis, edge and corner-based image features for the same cell regions. is being removed. In this analysis arm, the boundaries of the cell, nucleus and / or cytoplasm are marked with edge 30 Edge detection, corner detection, or both are applied. Edge detection, image It enables the identification of boundary regions where the intensity or color information changes. Corner detection involves identifying points in the boundary geometry where a change in direction is pronounced. It ensures that it is done. 35 The features extracted in the second arm of analysis are derived from the features generated in the morphological analysis arm. It is stored in a separate feature space. Thus, the nucleus / cytoplasm ratio and cellular data are represented. 11 biological morphological features such as size, and the distribution of edges and corners at the cell boundary. The related geometric image properties are obtained independently of each other. The aforementioned edge and corner-based features reflect structural changes at the cell boundary and It allows for the representation of irregularities. Morphological characteristics obtained from the first analysis arm and those obtained from the second analysis arm Edge and / or corner-based image properties are processed by the processing unit (110). They are combined. As a result of the merging, a composite feature is obtained for each cell region. The representation is being created. In an application, combined feature representation is added to the first branch of analysis. the morphological feature vector of the first arm and the image feature vector of the second analysis arm are 10 It is a composite feature vector formed by combining several features. Combined feature representation displays a cell based not just on a single visual attribute, but also on its morphological characteristics. and allows for the combined evaluation of their geometric properties. However, Because the features were extracted separately in the first and second branches of analysis, combining them is 15. The process is performed after feature extraction. This allows for the extraction of different feature types. The analysis arm from which the data was obtained is preserved. Combined feature representation requires at least one machine to be executed by the processing unit (110). It is learned and / or processed by a deep learning model. The model in question is 20 the relationship between combined feature representation and cellular classification categories It may be pre-trained in a way that will determine the model parameters in the memory unit. (120) is stored and used by the processing unit (110) during classification. As a result of processing the combined feature representation, 25 for each detected cell region. A cell-level classification score is generated. The cell-level classification score is used to determine the relevant... the risk of malignancy of the cell region or its relationship with a clinical classification category It represents the cell-level classification score, a numerical value belonging to multiple classes. a score group or cell that allows association with a specific clinical category It can be in the form of a classification output. 30 In one application, the cell-level classification score was negative, atypical urothelial cells, high. from the categories of suspected grade 1 urothelial carcinoma and high-grade urothelial carcinoma It is associated with at least one of these. The categories mentioned are clinical aspects of urine cytology images. 35 in a way that is consistent with the classification structure used in its interpretation. This can be generated. The classification output shows which category the relevant cell region belongs to. 12 that it is associated and the confidence value of this association in the classification model It may include. Cell-level classification scores are based on coordinate information for cell regions. It is related. For this purpose, image segment extraction was performed and image 5 The coordinate relationship is preserved throughout the processing stages. Each cell level The classification score is the score obtained on the original full slide image of the relevant cell region. It is matched with its location. As a result of the aforementioned matching, the cell-level classification scores are from the original full slide 10. The image is projected back onto it. During the back projection, the relevant cell The region's location and cell-level classification score are used together, and the full slide... A spatial heat map is created on the image. The spatial heat map, Visually represent the distribution of classification scores on a full slideshow. It is doing. 15 In the heat map, different cell-level classification scores are represented by different visual codes. It can be done. It has a higher classification score compared to other cell regions. Cell regions are visually distinguishable in the user interface (130) This is shown. Thus, the user carries a higher risk on the full slide image. It is possible to visualize areas where the identified cell regions are concentrated. A spatial heat map is a visual representation of cell-level classification scores. It is used as output. The creation of the aforementioned heat map is done at the cellular level. This is done after the classification score is calculated. Therefore, heat 25 the map is not an input that generates the cell-level classification score, but rather a generated one. classification scores with their positions on the original full slide image It is used as a presentation output in which it is presented. The method described in the invention also involves a third branch of analysis. The third branch of analysis is 30 The main classification that combines the features obtained from the first and second branches of analysis. It operates separately from the main line. Data obtained in the third branch of analysis, combined involved in feature display and cell-level classification score generation. This branch of analysis does not directly alter the classification output, but... Supporting analysis that helps the user evaluate the cells and sample 35 It generates outputs. 13 The third arm of analysis included cytoplasmic overlap, color intensity, and intercellular distance. At least one of the following characteristics is determined: cell distribution frequency and the presence of artifacts. One, several, or all of the analyses mentioned are on the same full slide image. It can be implemented. Cytoplasmic overlap analysis shows the overlapping of cytoplasmic regions belonging to neighboring cell areas. It determines their spatial relationship with each other. Within the scope of this analysis, segmentation whether the cytoplasmic regions determined as a result overlap or not whether the cellular structures form a clustered appearance on top of each other It can be evaluated. 10 Color intensity analysis of the cell, nucleus and / or cytoplasm region. It evaluates the intensity or color information. As a result of color intensity analysis... The obtained values provide the user with supporting information regarding the staining properties of cellular structures. It provides information. The aforementioned color intensity analysis is performed in the pre-processing stage at 15. Unlike color normalization, which is performed on a cell or cell region... This is carried out as an analysis output. Intercellular distance analysis is the process of identifying cellular regions on a full slide image. They evaluate their positions relative to each other. In the analysis mentioned, the cell regions are 20 The spatial distances between them are determined, and whether the cells are close to each other or not is assessed. a supporting output regarding whether they are widely distributed or not is being created. Cell distribution frequency analysis, full slide image or full slide 25 of the detected cells It evaluates the distribution of the image within specific image areas. Thus whether cells are concentrated in certain areas or whether certain cell types are visible in the image The frequency of its occurrence can be presented as a supporting analysis output. Artifact presence analysis, cell detection in full slide images or image fragments, 30 an image that may affect the interpretation of segmentation or classification results It identifies their structures. The identified artifact information is assigned to the relevant image region or cell. This is linked to the analysis results for the region and provided to the user as supporting information. It is presented. 35 Cytoplasmic overlap, color intensity, and intercellular relationships obtained in the third arm of analysis. Data regarding distance, cell distribution frequency, and artifact presence, at the cellular level. 14 It is not used in the creation of the classification score. The aforementioned data are for the first and second grades. It is kept separate from the composite feature representation obtained from the second arms of analysis. Thus, the primary classification result explains the classification or clinical review. The secondary analysis data produced to support this are separated from each other. The analysis results included a cell-level classification score, a spatial heat map, and a third analysis. At least one supporting analysis output obtained from the arm is in the user interface (130) The user interface (130) is presented as distinct output components. full slide view, spatial heat map, cell-level classification scores, cell their positions, classification confidence values, and supporting analysis for the third branch of analysis 10 They can view the outputs. The user interface (130) includes the full slideshow and the heat map together or separately. It can be viewed. The heat map is spatially aligned with a full slideshow. Thanks to this, the user is associated with a high cell-level classification score of 15. It can determine the location of regions on the sample. Cell locations and cell Level classification scores are correlated with the image fields of the relevant cells. It is presented. Supporting analysis outputs from the third arm of analysis, primary classification 20 The score can be displayed in a separate output area. This allows the user to see the classification. The cell-level result generated by the model, independently of this result. cytoplasmic overlap, color intensity, intercellular distance, cell distribution frequency, and It can distinguish between data indicating the presence of artifacts. Cytology analysis system (100) produces cell-level classification scores, spatial heat map and The supporting analysis results from the third arm of analysis are presented in a clinical report. It can present. The data mentioned in the clinical report are distinct outcome components. It is located as follows. The clinical report can be created in digital format and stored in the memory unit (120). It can be hidden or viewed via the user interface (130). 30 The invention concerns a multi-arm analytical structure, specifically the first arm of analysis from which morphological features are extracted. The second arm of analysis, in which edge and corner-based image features are extracted, is the primary cell. It is used for classification. The third analytical arm is primary cell. It generates supporting analytical data independent of classification. Thanks to this structure, 35 Data that forms the basis of cell classification and data that provides additional information to the user. They are processed and presented without being mixed together. The method described in the invention is also only considered to be high-risk or most atypical. It is not limited to displaying a limited number of cells. On the full slide view The identified cell regions can be subjected to the relevant analysis stages, and the cells Classification results can be stored along with coordinate information. This allows for 5 The user sees the analysis results within the original image context of the cells. can evaluate. Figure 1 shows a representative flowchart of the method described in the invention. Accordingly, the full slide The image is being acquired by the system, pre-processed, and cell regions are being identified. 10 Nuclear and cytoplasmic segmentation is performed in the identified cell regions. After this, morphological features were analyzed in the first arm, based on edge and corner bases. Image characteristics are extracted in the second analysis arm. The first and second analysis arms... The outputs are combined to create a cell-level classification score. The generated cell Level classification scores, cell coordinates on the full slide view 15 It is presented in the form of a spatial heat map using this data. In the third branch of analysis, however... Supporting analyses not included in the creation of the classification score This is being carried out through cell-level classification score, heat map, and supporting analysis. The outputs are presented via the user interface (130) and / or clinical report. The scope of protection of the invention is specified in the claims attached hereto, and these details are strictly adhered to. The explanation cannot be limited to those given for illustrative purposes. Because a technically skilled person... the person, without deviating from the main theme of the invention, in light of what has been described above, similar It is clear that these structures can emerge. 16 REFERENCE NUMBERS GIVEN IN THE FIGURE 100 Cytology analysis systems 110 Operation units 120 Memory units 5 130 User interface
Claims
17 REQUESTS 1. At least one processing unit (110), at least one memory communicating with the aforementioned processing unit (110). Computer-assisted cytology analysis unit (120) and at least one user interface (130) Digital image data of urine cytology sample by system (100) 5 to generate classification outputs at the cellular level through processing It is an image analysis method, and its characteristic feature is; ∑ a complete slide image of at least one urine cytology sample from the aforementioned procedure to be received by unit (110), ∑ Quality control of the aforementioned full slide image, image part extraction, color 10 It will include at least one of the following processes: normalization and background filtering. pre-processing in this way, ∑ Detection of cell regions on the pre-processed full slide image and the nuclei of at least some of the identified cell regions, Determination of cytoplasmic regions through segmentation, 15 ∑ at least one containing the nucleus / cytoplasm ratio of the mentioned cell regions The morphological feature is the first processed by the mentioned processing unit (110). removal from the analysis arm, ∑ Edge and / or corner relating to at least one boundary of the mentioned cell regions At least one geometric image feature containing information about the aforementioned processing unit 20 (110) removal in the second branch of analysis conducted by ∑ Morphological characteristics obtained from the first analysis arm and from the second analysis arm the aforementioned processing unit (110) for the obtained geometric image features by combining them to create a unified feature representation, ∑ The aforementioned combined feature representation is represented by the aforementioned processing unit (110) 25 clinical classification for each identified cell region through processing. Generating a cell-level classification score associated with categories, ∑ The aforementioned cell-level classification scores fully represent the respective cell regions. spatial information is determined by associating it with positional information on the slide image. the creation of a heat map and the aforementioned spatial heat map to user 30 (130) presented as visual output on the interface, ∑ cytoplasmic overlap, color intensity, intercellular distance, cell distribution At least one of the characteristics of frequency and artifact presence in the aforementioned combination feature display and cell-level classification score generation without including the third 35 performed by the aforementioned processing unit (110). determination in the analysis arm and 18 ∑ the aforementioned cell-level classification score, spatial heat map, and at least one supporting analysis output from the third arm of analysis in the user interface referred to as distinct output components (130) presentation It includes the following steps. 5 2. It is an image analysis method according to Claim 1, and its characteristic is; the aforementioned full slide. of the image, each resulting image fragment on the complete slide image Multiple images that will be associated with coordinate information indicating their location. It is the separation into parts. 10 3. It is an image analysis method according to Claim 2, and its characteristic is that the aforementioned image components and a full slide view of the cell regions identified in the aforementioned image segments. associating them with coordinate information that indicates their location on the mentioned surface and The coordinate relationship is preserved throughout the image processing stages. 15 4. It is an image analysis method according to any of the previous requirements, and its characteristic is; Identifying the aforementioned cell regions involves identifying the cells on the full slide image. The first image analysis model that enables the identification of candidate regions and the identified cells. Using a second image analysis model that enables the verification of candidate regions, two 20 It is implemented in stages.
5. It is an image analysis method according to Claim 4, and its characteristic is; the first image analysis mentioned above. the model is a YOLO-based object detection model and the aforementioned second image analysis The model is a Faster R-CNN based object detection model. 25 6. It is an image analysis method according to any of the previous requirements, and its characteristic is; the aforementioned segmentation process involves the nuclei and nuclei of the identified cell regions. This involves defining the boundaries of the cytoplasm.
7. It is an image analysis method according to Claim 6, and its characteristic is; the first analysis mentioned. in the arm, the nucleus / cytoplasm ratio and the size of the cell, nucleus and / or cytoplasm. At least one of the shape characteristics must be identified as a morphological feature.
8. A method of image analysis according to any of the previous requirements, and its characteristic is; 35 In the second arm of analysis mentioned, edge detection is performed at the boundary of the cell, nucleus and / or cytoplasm. 19 edge and / or corner detection by applying at least one of the following processes: edge detection and / or corner detection. This involves extracting at least one image feature based on the data.
9. It is an image analysis method according to any of the previous requirements, and its characteristic is; firstly; Morphological features obtained from one analysis arm and edge features obtained from the second analysis arm. and / or by combining corner-based image features, the aforementioned combined feature 5 It is the creation of the representation as a composite feature vector.
10. It is an image analysis method according to claim 9, and its characteristic is the aforementioned combined feature. the vector by at least one machine learning and / or deep learning model By processing, cell level 10 represents the risk of malignancy for each cell region. The process involves generating a classification score.
11. It is an image analysis method according to claim 10, and its characteristic is; at the aforementioned cell level. Negative classification score, atypical urothelial cell, high-grade urothelial carcinoma suspicion and at least one of the categories of high-grade urothelial carcinoma 15 It is the association.
12. It is an image analysis method according to any of the previous requirements, and its characteristic is; The third analysis arm mentioned includes cytoplasmic overlap, color intensity, and intercellular interaction. distance, cell distribution frequency, and artifact presence characteristics at least one of which is 20 It is the determination of.
13. It is an image analysis method according to claim 12, and its characteristic is that it is obtained from the third analysis arm. The analysis outputs obtained are presented in the aforementioned combined feature representation and at the cell level. 25 as a supporting analysis output, not included in the creation of the classification score. is the creation of.
14. It is an image analysis method according to Claim 3, and its characteristic is; the aforementioned spatial thermal analysis. the coordinates of the cell-level classification scores for the relevant cell regions on the map using the information, the original full slide image is projected back onto it and the scores are 30. It is created by visually coding it.
15. It is an image analysis method according to claim 14, and its characteristic is that it is different from other cell regions. cell regions with higher cell-level classification scores mentioned It is the visually distinguishable representation of the spatial heat map. 35 16. It is an image analysis method according to any of the previous requirements, and its characteristic is; The mentioned user interface (130) includes a full slide view, spatial heat map, and cell view. Level 3 classification scores, cell locations, classification confidence values, and third-level classification scores. This involves presenting at least two of the supporting analysis outputs obtained from the analysis arm.
17. It is an image analysis method according to any of the previous requirements, and its characteristic is; During the preprocessing of the aforementioned full slide image, sample preparation color in order to reduce color differences resulting from and / or dyeing conditions It is subjected to normalization.
18. It is an image analysis method according to claim 16, and its characteristic is; the aforementioned cell level. classification scores, spatial heat map, and results from the third arm of analysis supporting analysis outcomes as distinct outcome components of a clinical practice It is presented in the report.
19. At least one processing unit (110), at least one memory communicating with the aforementioned processing unit (110). Computer-assisted cytology analysis unit (120) and at least one user interface (130) The system (100) has the feature that the processing unit (110) is digital for urine cytology sample. cell-level classification outputs through the processing of image data 20 will implement the method in one of the previous requests to ensure its creation. It is configured in this way.