Adaptive Threshold Segmentation Method and Device for Increased Lung CT Density Shadows

Abstract

The present invention provides an adaptive threshold segmentation method and device for pulmonary CT density-increased shadows. The method includes: obtaining a pulmonary CT image and normalizing the obtained image; binarizing the normalized image to obtain a binary mask of the lung parenchyma, and obtaining a lung parenchyma mask based on the binary mask of the lung parenchyma; marking a closed contour containing the lesion area on the normalized image; generating a binary mask according to the closed contour, performing a "logical AND" operation with the lung parenchyma mask to generate a lesion mask; using the lesion mask to extract corresponding pixel points in the normalized pulmonary CT image and performing a histogram statistics to obtain the pixel value with the highest histogram intensity; adding the pixel value with the highest histogram intensity and a predetermined positive integer offset value as the separation threshold; binarizing the normalized pulmonary CT image within the closed contour using the separation threshold to obtain a lesion area mask. The present invention can achieve fine segmentation of pulmonary images and has high stability and repeatability.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, in particular to the processing of lung CT images, and more particularly to an adaptive threshold segmentation method and device for lung CT density-enhanced shadows. Background Art

[0002] Currently, computed tomography (CT) is widely used to examine various lesions due to its high spatial resolution and clear images. Low-dose spiral CT, with its sensitivity and low radiation dose, is widely recommended for screening lung lesions, especially early-stage lung cancer. CT imaging results are also included in the clinical diagnostic criteria for COVID-19, playing a crucial role in epidemic control.

[0003] CT image data processing involves the segmentation of lung images, which includes lung parenchyma segmentation, lung airway segmentation, and lung nodule segmentation. The design of image segmentation models is mainly based on the consensus formed by multiple experts as the gold standard, such as the "STAPLE" method proposed by Simon K et al., and then the model is evaluated, tuned, and improved. However, even for experienced experts, the consensus is still formed based on subjective visual judgment. For target areas with clear edges and strong contrast with the background, it is easy to reach a consensus on the design of the segmentation gold standard. However, for ground-glass-like lesions with blurred edges and no obvious gradient, the difficulty in forming an evaluation standard is as follows: (1) The edge of the lesion is not clearly defined, and edge delineation is a difficult problem for each physician; (2) When multiple physicians work together, the recognition of the edge varies greatly, and even if a comprehensive opinion is formed, it is unstable and unconvincing; (3) It is difficult for the same physician to form a stable standard.

[0004] Unstable evaluation criteria make it difficult to draw reliable conclusions for both intra-model group evaluation and inter-model group evaluation.

[0005] Although researchers have proposed numerous medical image segmentation models, all of which have achieved good results in their respective targeted problems, for example, the Snake active contour model, as one of the most representative and practical edge segmentation models in recent years, has achieved good results in segmenting lung tumors with relatively clear edges. However, for patchy lung inflammatory changes with blurred edges and irregular shapes, edge-based segmentation methods usually fail, and the currently common evaluation criteria based on expert experience also fail.

[0006] Therefore, how to effectively avoid the instability of segmentation conclusions caused by large differences in physicians' subjective cognition, improve the robustness of image segmentation, achieve fine segmentation of lung images, and achieve high stability and repeatability is an important problem that needs to be solved. Summary of the Invention

[0007] An embodiment of the present invention provides an adaptive threshold segmentation method and device for pulmonary CT density-increased shadows to solve the problems of unstable image segmentation conclusions and poor segmentation robustness in the prior art.

[0008] According to one aspect of the present invention, an adaptive threshold segmentation method for pulmonary CT density-increased shadows is provided, and the method includes the following steps:

[0009] Obtain a pulmonary CT image with patchy density-increased shadows, and normalize the obtained pulmonary CT image;

[0010] Binarize the normalized image to obtain a binary mask of the lung parenchyma, and extract the pulmonary image based on the binary mask of the lung parenchyma to obtain a lung parenchyma mask;

[0011] Identify the closed contour marked by an image annotator on the normalized pulmonary CT image, and the closed contour includes the lesion area;

[0012] Generate a contour binary mask according to the marked closed contour, and perform a "logical AND" operation with the lung parenchyma mask to generate a lesion mask;

[0013] Use the lesion mask to extract the corresponding pixel points in the normalized pulmonary CT image and perform a histogram statistics to obtain the pixel value with the highest histogram intensity;

[0014] Sum the pixel value with the highest histogram intensity and a predetermined positive integer offset value as the separation threshold;

[0015] Binarize the normalized pulmonary CT image within the closed contour using the separation threshold to obtain a lesion area mask.

[0016] In some embodiments of the present invention, after binarizing the normalized pulmonary CT image within the closed contour using the separation threshold, it further includes: preprocessing the normalized pulmonary CT image within the binarized closed contour, and the preprocessing includes: removing small connected components, dilation operation, and / or filling holes operation.

[0017] In some embodiments of the present invention, the closed contour is: a polygon, a circle, or an ellipse.

[0018] In some embodiments of the present invention, the obtaining of the pulmonary CT image includes CT images meeting the following conditions: 1) pulmonary CT images with patchy density-increased shadows; and 2) ground-glass lesions exist in each layer of CT slices of the same patient;

[0019] In some embodiments of the present invention, the normalization of the acquired lung CT image includes: setting the lung window width and window position to a predetermined lung window width value and a predetermined window position value, respectively, normalizing the voxel values in the window to a range of 0 to 255; and removing images of areas outside the lungs.

[0020] In some embodiments of the present invention, the predetermined lung window width is 1800HU; the predetermined window level is 300HU.

[0021] In some embodiments of the present invention, after calculating two lesion masks for the first closed contour and the second closed contour annotated by two annotators, a first segmentation area and a second segmentation area are calculated based on the two lesion masks, and an intersection-over-union (IoU) index of lesion segmentation stability is calculated based on the first segmentation area and the second segmentation area using the following formula:

[0022]

[0023] Among them, IOU represents the intersection-over-union ratio indicator, C1 and C2 are the first segmentation area and the second segmentation area respectively.

[0024] In some embodiments of the present invention, the closed contour satisfies the following conditions: contains a complete lesion; contains normal tissue as background.

[0025] In some embodiments of the present invention, the normal tissue is not less than 1 / 3 of the contour; and / or when a vascular structure with a brightness above a predetermined level passes through the lesion, the vascular structure is merged with the lesion, or the vascular structure is avoided when marking the contour.

[0026] In some embodiments of the present invention, the positive integer offset value ranges from 15 to 25.

[0027] In another aspect of the present invention, a semi-active threshold segmentation device for increased lung CT density shadows is provided. The device includes a processor and a memory, wherein the memory stores computer instructions. The processor is used to execute the computer instructions stored in the memory. When the computer instructions are executed by the processor, the device implements the steps of the method described above.

[0028] Another aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which implements the steps of the above-mentioned method when executed by a processor.

[0029] The adaptive threshold segmentation method and device for lung CT density increase shadows of the present invention can improve the robustness of image segmentation, achieve fine segmentation of lung images, and achieve higher stability and repeatability.

[0030] Additional advantages, objects, and features of the present invention will be set forth in part in the description which follows, and in part will become apparent to those having ordinary skill in the art upon examination of the following, or may be learned from practice of the present invention. The objects and other advantages of the present invention may be realized and attained by the structure particularly pointed out in the written description and claims hereof as well as the appended drawings.

[0031] Those skilled in the art will understand that the objects and advantages that can be achieved with the present invention are not limited to those specifically described above, and the above and other objects that the present invention can achieve will be more clearly understood from the following detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] The drawings described herein are for further understanding of the present invention, form a part of this application, and do not limit the present invention.

[0033] Figure 1 It is a schematic flow diagram of an adaptive threshold segmentation method for increased density shadows in lung CT images in an embodiment of the present invention.

[0034] Figure 2A and Figure 2B They are respectively examples of a normalized lung CT image and its global histogram in an embodiment of the present invention.

[0035] Figure 3A and Figure 3B They are respectively schematic diagrams of a binary mask of a lung CT image and an extracted lung parenchyma image obtained in an embodiment of the present invention.

[0036] Figure 4A and Figure 4B They are respectively a CT image with the healthy area and the ground-glass opacity area marked and histogram examples of the two areas.

[0037] Figure 5A and Figure 5B They are respectively a CT image with the healthy area and the ground-glass opacity area marked and histogram examples of the two areas.

[0038] Figure 6A and Figure 6B They are respectively a CT image with two closed contours each containing a lesion area marked and histogram examples of the two areas.

[0039] Figure 7A and Figure 7B They respectively show a schematic diagram of a roughly marked closed contour in a CT image of an embodiment of the present invention and a density-increased lesion proposed based on an adaptive threshold on the basis of the closed contour.

[0040] Figure 8A and Figure 8BSchematic diagrams of lung image segmentation based on traditional manual segmentation method and semi-active threshold method of the present invention respectively.

[0041] Figure 9 Schematic diagrams of three lesions obtained by two physicians based on traditional manual segmentation method and semi-active method of the present invention respectively. Detailed implementation manners

[0042] To make the objectives, technical solutions and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below in conjunction with the implementation manners and the drawings. Herein, the illustrative implementation manners of the present invention and their descriptions are used to explain the present invention, but do not limit the present invention.

[0043] Herein, it should also be noted that in order to avoid obscuring the present invention due to unnecessary details, only the structures and / or processing steps closely related to the solution of the present invention are shown in the drawings, and other details less related to the present invention are omitted.

[0044] It should be emphasized that the term "comprising / including" when used herein refers to the presence of features, elements, steps or components, but does not exclude the presence or addition of one or more other features, elements, steps or components.

[0045] An embodiment of the present invention proposes a semi-active lesion segmentation method, which can calculate the segmentation threshold and extract the lesion area based on the region including the lesion and the background manually depicted by a physician for the patchy inflammatory density increase shadow in a lung CT, and extract the most stable segmentation target among the depictions of different physicians, which can effectively avoid the instability of the segmentation conclusion caused by the large difference in the subjective cognition of physicians.

[0046] Figure 1 The following shows a schematic flow diagram of the semi-active threshold segmentation method for the density increase shadow in a lung CT in an embodiment of the present invention. As Figure 1 shown, the method includes steps S110 to S170, wherein:

[0047] In step S110, a lung CT image is acquired and the acquired lung CT image is normalized.

[0048] An existing CT scanner can be used to acquire the lung CT image. For example, a low-dose spiral CT scanner can be used to acquire the lung CT image.

[0049] In an embodiment of the present invention, preferably, CT images showing patchy increased density shadows (such as ground-glass shadows, mixed shadows, and / or consolidation shadows, etc.) are selected as the research objects. Therefore, the CT images that meet the following conditions are used as the lung CT images to be studied: (1) Patchy increased density shadows appear on thin-slice chest CT images, including ground-glass shadows, mixed ground-glass shadows, and / or consolidation shadows. (2) Patchy increased density shadow lesions exist in each CT slice of the same patient. Correspondingly, the CT images in the following situations are not used as research objects: (1) No patchy increased density shadows are seen on thin-slice chest CT examinations; (2) There are large-area diffuse lesions, combined with other lesions, or severe motion artifacts in the patient's CT images.

[0050] In an embodiment of the present invention, after obtaining the lung CT images of a patient, for the convenience of subsequent data processing, preferably, the original images of the lung CT are normalized, such as performing gray value normalization. The original CT images usually have a large bit depth and a large numerical range, which is not conducive to the analysis and visualization of the region of interest. Therefore, in the normalization process of this embodiment of the present invention, the window width and window level of the lung window are respectively set to a predetermined lung window width value and a predetermined window level value, and the voxel values within the window are normalized to the range of 0-255. For example, the predetermined lung window width value and the predetermined window level value can be set to 1800HU and -300HU respectively.

[0051] In addition, due to the interference of structures around the lungs, such as blood vessels, bronchi, examination beds, etc., not only will the amount of irrelevant data be increased, but also the histogram, annotation contour, etc. will be interfered. Therefore, in this embodiment of the present invention, the areas outside the lungs are also removed.

[0052] The lung parenchyma of normal people is filled with air and has a small density, which is significantly different from the surrounding rib tissues, blood vessels, heart and other organs. During CT examination, it is reflected as the difference in voxel CT values (Hounsfield Unit). The average CT value of the whole lung of healthy adults is -886±22HU at the end of deep inspiration, while the CT value of the density-increased area is often greater than -600HU, showing a brighter area.

[0053] For better analysis and visualization of lung CT images, the present invention performs histogram conversion processing on the CT images. Figure 2A and Figure 2B The figures respectively show an example diagram of the normalized CT image and the global histogram distribution of the normalized CT image. As Figure 2A and Figure 2B shown, the healthy lungs are filled with air, and most of their gray values are concentrated within 100 (-480HU). Figure 2B In, the peak around the gray value of 150 represents the density-increased area, corresponding to the A area in Figure 2A .

[0054] In step S120 , the normalized CT image is binarized to obtain a binarized mask of the lung parenchyma, and a lung image is extracted based on the binarized mask of the lung parenchyma to obtain a lung parenchyma mask.

[0055] In this step, an empirical threshold can be selected based on the histogram of the normalized CT image, such as selecting an empirical threshold of 100, and binarizing the normalized CT image. Then, the steps of clearing the boundary, taking a connected domain with a larger area, morphological closing operation, filling holes, etc. are sequentially performed to finally obtain a binary mask of the lung parenchyma. Among them, morphological closing operation and filling holes are respectively used to fill small gaps to form a complete binary mask. Based on the binary mask of the lung parenchyma, the lung image can be extracted from the normalized CT image to obtain a lung parenchyma mask (lung parenchyma image). The binary mask of the lung parenchyma and the extracted lung parenchyma image are respectively as shown in FIG. Figure 3A and 3B shown.

[0056] In step S130 , a closed contour is marked on the normalized lung CT image with the healthy lung parenchyma as the background and the lesion area (or lesion area) as the foreground, so that the closed contour includes the lesion area.

[0057] For example, a CT image contour annotator such as a physician may use healthy lung parenchyma as the background and the diseased area (or lesion area) as the foreground to annotate a closed contour on the normalized lung CT image, so that the annotated closed contour includes the lesion area, which is used to segment the lung parenchyma, that is, to segment the patchy inflammatory density increase shadow in the lung CT image, and the patchy inflammatory density increase shadow is the lesion region of interest (ROI).

[0058] Different from the prior art, in this step, the contour annotation of the CT image can be roughly performed, and a more accurate segmentation of the lung parenchyma can be achieved based on the following steps.

[0059] Figure 4A and Figure 4B Examples of CT images with healthy areas and patchy pure ground-glass opacity areas marked, as well as histograms of the two areas. Figure 5A and Figure 5B The following are examples of CT images with healthy areas and mixed ground-glass opacity areas marked, and histograms of the two areas. Figure 4A As shown, the left and right labeled contours are typical lesion local areas and healthy local areas, respectively. Figure 4A The lesions are patchy pure ground glass density shadows. Figure 5A As shown, the left and right labeled contours are typical lesion local areas and healthy local areas, respectively. Figure 5AThe lesion is a mixed ground-glass opacity with more solid components. Figure 4A and Figure 5A have in common that the gray values of almost all pixels are lower than 200 (corresponding to HU value 205). The histogram of the healthy lung parenchyma region (excluding bronchi) is a unimodal structure with a Poisson-like distribution, and the gray-level peak is located around 20 - 30 (-1060HU - -990HU), and the gray values of more than 95% of the pixels are lower than 50 (-850HU). The gray histogram containing the lesion area shows an obvious fat-tail effect, with an increase in high-gray-value pixels and a decrease in low-gray-value pixels. Figure 5A There are more solid components in the lesion area of , and a local peak appears in the high-gray-value area on the histogram. However, the global peak is still around 20 - 30, and compared with the healthy area, the global peak has a slight right shift within 5 (35HU).

[0060] Figure 4A and Figure 5A The segmentation of the lung lesions in and is a rough segmentation with the healthy lung parenchyma as the background and the lesion area as the foreground. Figure 4A and Figure 5A The left closed contours in the corresponding two CT slices of and do not depict the fine structure of the lesion, so they include some low-density areas, making the position of the global gray peak unchanged significantly. At the same time, the shape of the high-density lesion shadow is irregular and the edge is unclear, making it difficult to depict the fine structure. Due to the rough segmentation being difficult to depict the fine structure, in the existing technology, it often requires experienced radiologists to perform it, to mark the lesion area as accurately as possible while including as little of the low-density area corresponding to the healthy part as possible. This makes the existing image segmentation not only highly dependent on experience, but also the inter-group error caused by the delineation (marking) of different annotators and the intra-group error caused by the delineation of the same annotator twice are relatively large, thus making the accuracy, stability, and repeatability of the lesion contour delineation all affected greatly and difficult to guarantee.

[0061] According to the histogram analysis of the lesion and the healthy lung parenchyma, it can be seen that if the lesion contour is roughly depicted, in the gray histogram of the included pixel points, the peak position is still approximately close to that of the healthy lung parenchyma. If the lesion contains a solid part, a small local peak may form in the high-gray-value area, and its probability distribution is much smaller than the pixel values of the adjacent background.

[0062] To confirm this judgment, Figure 6A The two closed polygon contours drawn both include the entire lesion area, only the areas of the lung parenchyma background included by them are different. As Figure 6BThe histograms of the two closed polygonal contours in the figure show that the grayscale value curves above 50 are almost identical, and the global peak positions are almost identical, but the intensities are different. Overall, the second contour includes more lung parenchyma than the first, so the histogram integral of the second contour area is larger than that of the first contour area. The increased area corresponds to the lung parenchyma, so the absolute value of the grayscale peak is increased, while the position remains unchanged.

[0063] As the above reasoning indicates, for contours independently drawn by multiple physicians, as long as the contour encompasses the entire extent of the lesion to be studied, even if it also includes portions of healthy lung parenchyma, a similar histogram distribution can be obtained. That is, the high grayscale values are nearly identical, the global peak positions overlap, and the absolute intensities differ. Based on this conclusion, the present invention devised an adaptive threshold segmentation method. Therefore, in this embodiment of the present invention, although the CT image is roughly marked with a closed contour in this step, the subsequent adaptive threshold segmentation step can achieve accurate segmentation and delineation of the lesion area.

[0064] In step S140 , a contour binary mask is generated according to the marked closed contour, and a “logical AND” operation is performed on the contour binary mask with the lung parenchyma mask to generate a lesion mask.

[0065] That is, in this step, the marked closed contour is binarized to generate a contour binarization mask; further, after performing a “logical AND” operation on the contour binarization mask and the lung parenchyma mask obtained in step S120 , a lesion mask can be generated.

[0066] In step S150, the corresponding pixel points in the normalized lung CT image are extracted using the lesion mask and histogram statistics are performed to obtain the pixel value with the highest histogram intensity.

[0067] That is, in this step, a histogram of the lesion area is obtained based on the corresponding pixel points in the lung CT image corresponding to the lesion mask, and the histogram is statistically analyzed to obtain the pixel value with the highest histogram intensity. The highest pixel value corresponds to the high grayscale value pixel in the lesion area.

[0068] In step S160, the sum of the pixel value with the highest histogram intensity and a predetermined positive integer offset value is taken as a separation threshold.

[0069] In the embodiment of the present invention, the positive integer offset value may be selected from an empirical value of 15-25, for example, 15.

[0070] In step S170 , the normalized lung CT image within the closed contour is binarized using a separation threshold to obtain a refined lesion region mask.

[0071] Optionally, after binarizing the normalized lung CT image within the closed contour using the separation threshold, preprocessing may also be performed on the normalized lung CT image within the binarized closed contour. The preprocessing includes: removing small connected components, dilation operation, and / or filling holes operation, etc. Among them, removing small connected components is used to obtain larger connected components, dilation operation is used to remove edge burrs to make the edge as smooth as possible, and filling holes is used to remove small holes in the image.

[0072] Figure 7A Fig. 4 shows a roughly marked closed contour 70A in the CT image of an embodiment of the present invention. Figure 7B Fig. 5 shows a schematic diagram of a density-increased lesion 70B obtained based on an adaptive threshold on the basis of this closed contour. As Figure 7A shown in Fig. 6 and Fig. 7B, based on the roughly marked closed contour, a fine mask of the lesion area can be obtained.

[0073] The above steps S140 - S170 belong to a semi-active threshold segmentation step, or an adaptive threshold segmentation step. By determining the separation threshold based on the pixel value with the highest histogram intensity in the CT image within the roughly marked closed contour and a predetermined positive integer offset value, and more finely segmenting the area within the closed contour based on the separation threshold, a finer mask of the lesion area is obtained. That is, the semi-active threshold segmentation step of the embodiment of the present invention realizes a more accurate determination of the lesion area.

[0074] Compared with the prior art, the semi-active threshold segmentation method for lung CT density-increased shadows in the embodiment of the present invention can improve the robustness of image segmentation, achieve fine segmentation of lung images, and achieve high stability and repeatability. The semi-active threshold segmentation method of the present invention is simple, fast, and easy to implement, greatly improving the stability of the segmentation result.

[0075] Next, the Intersection-over-Union (IOU) metric is used to verify the stability of the segmentation result of the present invention. More specifically, two lesion area masks can be calculated respectively for two closed contours marked by two physicians, the first segmentation area and the second segmentation area are calculated respectively based on the two lesion area masks, and the IOU metric for the lesion area segmentation stability is calculated using the IOU metric formula based on the first segmentation area and the second segmentation area. In the embodiment of the present invention, the stability of the segmentation result is evaluated by the IOU metric:

[0076]

[0077] Among them, IOU represents the IOU metric, C1 and C2 are the first segmentation area and the second segmentation area respectively, that is, C1 and C2 are the lesion distributions obtained by adaptive threshold segmentation according to the contours drawn by two physicians respectively.

[0078] For comparison, the present invention also provides the IOU index range obtained by the traditional manual segmentation method. In this case, C1 and C2 are the lesion distributions covered by the contours drawn by two doctors, respectively.

[0079] For ease of description, the IOU obtained using the method of the present invention is referred to as "human-machine" IOU, while the IOU obtained using traditional manual segmentation is referred to as "human-human" IOU. When performing global statistics, the 95% confidence interval of the segmentation results is estimated using the non-parametric bootstrap method.

[0080] Figure 8A and Figure 8B Schematic diagrams of lung image segmentation based on a traditional manual segmentation method and a semi-active threshold segmentation method based on the present invention are respectively shown.

[0081] This example provides 200 lesion cross-sections from multiple chest CT image slices. Each lesion cross-section was segmented using both traditional manual segmentation and the semi-active segmentation method of the present invention. Two physicians (Physician A and Physician B) drew two contour lines (Contour 1 and Contour 2) for each method. The stability of the contouring was evaluated using the IoU index.

[0082] For manual segmentation, the precise division of the lesion region of interest (ROI) plays a key role in the subsequent feature engineering design and statistical analysis. In order to reduce the human errors caused by manual outlining of the ROI area (including intra-group errors caused by two outlinings by the same annotator and inter-group errors caused by outlining by different annotators) and the omission of lesion edge information, the labelme tool can be used as an annotation tool, and polygon, rectangle, circle and other shape tools can be used to draw closed contours to achieve the annotation of the region of interest. In the manual segmentation process, the physician tries to select accurate regions of interest based on experience. Since the traditional manual segmentation process is an existing technology, it will not be described here.

[0083] In the semi-active threshold segmentation process of the present invention, the labelme tool can be used as the annotation tool, or other drawing tools can also be used as the annotation tool. Two physicians can perform the annotation work separately and independently on the same set of CT image samples in the same order. For the ground-glass lesion area existing in a certain slice of the CT image, Physician A depicts the edge contour of a ground-glass shadow containing the lesion based on subjective understanding; Physician B also draws an edge contour containing the lesion. The contours drawn by Physician A and Physician B are to a certain extent arbitrary as long as they cover the entire lesion range. In addition, preferably, the drawn contour may also include a certain amount of normal lung tissue as the background (for example, the background is not less than 1 / 3, and here 1 / 3 is only an example, and the present invention is not limited to this). In addition, if there are blood vessel structures with higher brightness (such as blood vessels above a predetermined brightness) passing through the lesion, the blood vessel part can be merged with the lesion part, or the blood vessel structure can be avoided when drawing the contour. Based on the rough contours drawn by the two physicians, a finer lesion structure can be further obtained through the adaptive threshold segmentation step. Based on the obtained structure, the lesion distributions C1 and C2 obtained by adaptive threshold segmentation according to the contours drawn by the two physicians can be calculated.

[0084] Figure 9 The lesion area masks obtained based on manual drawing (segmentation) and the adaptive drawing (segmentation) of the present invention are shown respectively. As Figure 9 shown, for the same lesion, the lesion areas obtained by different physicians through manual segmentation vary greatly, so it is difficult to stably and finely extract the lesion structure. However, by using the semi-active threshold segmentation method (or called the lesion structure segmentation method) of the present invention, whether Physician A or Physician B performs the rough contour drawing, a finer and more consistent lesion structure can finally be obtained.

[0085] Based on the above experiments, the IOU index obtained by the semi-active threshold segmentation method proposed by the present invention on the test samples is 0.81 (95% confidence interval: 0.78, 0.84), which is much better than the subjective segmentation results of physicians. The ground-glass opacity, which represents the increased lung density lesions, is one of the main lesions in lung diseases, and the fuzzy edge brings great inconvenience to physicians in quantitatively analyzing the lesion volume. For example, the IOU obtained by the subjective annotations of two physicians using the traditional manual segmentation method is only 0.51 (95% confidence interval: 0.51, 0.58), depending on the physician's style. That is to say, through the designed semi-active threshold segmentation method, fine lesion structures can be stably and robustly extracted, which can be used as the basis for quantitatively evaluating the condition and has a fundamental significance for precise and personalized treatment.

[0086] From the above, it can be seen that the adaptive threshold segmentation method for increased lung CT density shadows proposed in the present invention is a quick method that achieves the lesion segmentation effect of maximizing the suppression of human factors. This method has important clinical significance. Patients may visit multiple doctors to evaluate the progression of the disease or the classification of the disease throughout the course of their illness. In order to achieve the grand goal of modern precision and personalized treatment, the stability of disease assessment under multi-person collaboration is a primary problem that needs to be solved urgently. For example, the progressive change in the volume of a certain local affected tissue is one of the key indicators for quantitatively assessing the degree and speed of disease progression, and the lack of a segmentation standard for lesions with fuzzy edges will cause a huge variance in the manual segmentation results, which loses the reference value. Therefore, the proposal of this method has strong practical significance for the quantitative assessment of disease progression, quantitative assessment of benign and malignant disease, and stable extraction of statistical modeling parameters of lesions.

[0087] The method of the present invention is a robust semi-active adaptive threshold segmentation method, which can achieve stable and fine segmentation of patchy density-increased shadows with blurred edges in lung CT images, removing obstacles for future quantitative evaluation and statistical modeling.

[0088] Corresponding to the aforementioned method, the present invention also provides a semi-active threshold segmentation device for increased lung CT density shadows, the device comprising a processor and a memory, the memory storing computer instructions, the processor being used to execute the computer instructions stored in the memory, and when the computer instructions are executed by the processor, the device implements the steps of the method described above.

[0089] The present invention also relates to a storage medium on which computer program code can be stored, which can implement various embodiments of the method of the present invention when the program code is executed. The storage medium can be a tangible storage medium, such as an optical disk, random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, register, hard disk, removable disk, CD-ROM or any other form of tangible storage medium known in the technical field.

[0090] It should be understood that the present invention is not limited to the specific configurations and processes described above and illustrated in the figures. For the sake of brevity, a detailed description of known methods is omitted. In the above embodiments, several specific steps are described and illustrated as examples. However, the method of the present invention is not limited to the specific steps described and illustrated. Those skilled in the art may make various changes, modifications, and additions, or change the order of the steps after understanding the spirit of the present invention.

[0091] Those skilled in the art will appreciate that the various exemplary components, systems, and methods described in conjunction with the embodiments disclosed herein can be implemented in hardware, software, or a combination of both. Whether hardware or software is used depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of the present invention. When implemented in hardware, it may be, for example, an electronic circuit, an application-specific integrated circuit (ASIC), appropriate firmware, a plug-in, a function card, etc. When implemented in software, the elements of the present invention are programs or code segments used to perform the required tasks. The program or code segment may be stored in a machine-readable medium or transmitted over a transmission medium or communication link via a data signal carried in a carrier wave. "Machine-readable medium" may include any medium capable of storing or transmitting information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, ROM, flash memory, erasable ROM (EROM), floppy disks, CD-ROMs, optical disks, hard disks, fiber optic media, radio frequency (RF) links, etc. The code segments may be downloaded via a computer network such as the Internet or an intranet.

[0092] It should also be noted that the exemplary embodiments described herein describe methods or systems based on a series of steps or devices. However, the present invention is not limited to the order of the steps described above. In other words, the steps may be performed in the order described in the embodiments, or in a different order, or several steps may be performed simultaneously.

[0093] In the present invention, features described and / or illustrated for one embodiment may be used in the same or similar manner in one or more other embodiments, and / or combined with or replace features of other embodiments.

[0094] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention are intended to be within the scope of protection of the present invention.

Claims

1. An adaptive threshold segmentation method for pulmonary CT density increased shadows, characterized in that The method includes the following steps: Obtain a lung CT image with patchy density increase, and normalize the obtained lung CT image; Binarize the normalized image to obtain a binary mask of the lung parenchyma, and extract the lung image based on the binary mask of the lung parenchyma to obtain a lung parenchyma mask; Identify the closed contour marked by the image annotator on the normalized lung CT image, where the closed contour includes the lesion area; Generate a contour binary mask according to the marked closed contour, and perform a "logical AND" operation with the lung parenchyma mask to generate a lesion mask; Use the lesion mask to extract the corresponding pixel points in the normalized lung CT image and perform a histogram statistics to obtain the pixel value with the highest histogram intensity; Add the pixel value with the highest histogram intensity and a predetermined positive integer offset value as the segmentation threshold; Use the segmentation threshold to binarize the normalized lung CT image within the closed contour to obtain a lesion area mask; Wherein, the method further includes: Calculate two lesion area masks respectively for the first closed contour and the second closed contour marked by two annotators, calculate the first segmentation area and the second segmentation area respectively based on the two lesion area masks, and calculate the intersection over union index of the lesion area segmentation stability using the following formula based on the first segmentation area and the second segmentation area: ； Wherein, IOU represents the intersection over union index, and C1 and C2 are the first segmentation area and the second segmentation area respectively.

2. The method according to claim 1, characterized in that, After using the segmentation threshold to binarize the normalized lung CT image within the closed contour, it further includes: Preprocess the normalized lung CT image within the binarized closed contour, and the preprocessing includes: removing small connected components, dilation operation and / or filling holes operation.

3. The method according to claim 1, wherein The closed contour is: a polygon, a circle or an ellipse.

4. The method according to claim 1, wherein The normalization of the obtained lung CT image includes: Set the lung window width and window level to a predetermined lung window width value and a predetermined window level value respectively, and normalize the voxel values within the window to the range of 0-255; and Remove the image outside the lungs.

5. The method according to claim 4, wherein The predetermined lung window width value is 1800 HU; The predetermined window level value is -300 HU.

6. The method according to claim 1, wherein The closed contour satisfies the following conditions: Contains the complete lesion; Contains normal tissue as the background.

7. The method according to claim 6, wherein The normal tissue is not less than 1 / 3 of the contour; and / or In the case where there is a vascular structure with a predetermined brightness or above passing through the lesion, merge the vascular structure with the lesion part, or avoid the vascular structure when marking the contour.

8. The method according to claim 1, wherein The range of the positive integer offset value is 15-25.

9. An adaptive threshold segmentation device for lung CT density increased shadows, the device comprising a processor and a memory, characterized in that, The memory stores computer instructions, and the processor is used to execute the computer instructions stored in the memory. When the computer instructions are executed by the processor, the device implements the steps of the method according to any one of claims 1-8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps of the method according to any one of claims 1 to 8.

Images