CN114764809A

CN114764809A - Self-adaptive threshold segmentation method and device for lung CT (computed tomography) density increase shadow

Info

Publication number: CN114764809A
Application number: CN202110038966.4A
Authority: CN
Inventors: 苏景诗; 戴永恒; 王鹏达
Original assignee: Diankeyun Beijing Technology Co ltd
Current assignee: Diankeyun Beijing Technology Co ltd
Priority date: 2021-01-12
Filing date: 2021-01-12
Publication date: 2022-07-19

Abstract

The invention provides a self-adaptive threshold segmentation method and a device for a lung CT density increase shadow, wherein the method comprises the following steps: acquiring a lung CT image, and normalizing the acquired image; binarizing the normalized image to obtain a lung parenchymal binary mask, and obtaining a lung parenchymal mask based on the lung parenchymal binary mask; marking a closed contour containing a focus area on the normalized image; generating a binary mask according to the closed contour, and performing logical AND operation on the binary mask and the lung parenchyma mask to generate a focus mask; extracting corresponding pixel points in the normalized lung CT image by using a focus mask and performing histogram statistics to obtain a pixel value with the highest histogram intensity; summing the pixel value with the highest histogram intensity with a predetermined positive integer offset value as a split threshold; and binarizing the normalized lung CT image in the closed contour by using a separation threshold value to obtain a lesion region mask. The invention can realize the fine segmentation and higher stability and repeatability of the lung image.

Description

Self-adaptive threshold segmentation method and device for lung CT (computed tomography) density increase shadow

Technical Field

The invention relates to the technical field of image processing, in particular to lung CT image processing, and particularly relates to a self-adaptive threshold segmentation method and device for a lung CT density increase shadow.

Background

At present, the Computed Tomography (CT) technique is widely applied to various pathological changes inspection with the characteristics of high spatial resolution, clear images and the like. Low dose spiral CT is widely recommended for screening lung lesions, especially early lung cancer, because of its sensitive effect and low radiation dose. The CT image result is also included in the clinical diagnosis standard of the novel coronavirus pneumonia, and plays an important role in the epidemic situation control process.

CT image data processing involves segmentation of lung images, including lung parenchyma segmentation, lung airway segmentation, lung nodule segmentation, and the like. The design of the image segmentation model mainly takes consensus formed by a plurality of experts as a golden standard, for example, a 'STAPLE' method proposed by Simon K and the like, and then the model is evaluated, adjusted and improved. However, even with experienced experts, their consensus is still based on subjective visual judgment. For a target region with clear edge and strong contrast with the background, the design of the segmentation gold standard is easy to form consensus, while for a target region with fuzzy edge and no obvious gradient grinding-like glass shadow focus, the difficulty of forming evaluation standard is as follows: (1) the margin of the lesion is not well defined, and margin delineation is a troublesome problem for each physician; (2) when multiple doctors cooperate together, the identification difference of each person to the edge is large, and even if a comprehensive opinion is formed, the doctor is unstable and difficult to convince; (3) the same physician may have difficulty in establishing a stable standard on his own.

Unstable evaluation criteria are difficult to make reliable conclusions on both intra-model evaluation and inter-model evaluation.

Although researchers have proposed many medical image segmentation models that achieve better results in the respective problems, for example, the Snake active contour model, which is one of the most representative and practical edge segmentation models in recent years, achieves better results in segmentation of lung tumors with relatively sharp edges, the edge-based segmentation method generally fails for the pulmonary inflammatory changes of plaques with blurred edges and irregular shapes, and the currently-used evaluation criteria based on expert experience also fails.

Therefore, how to effectively avoid unstable segmentation conclusion caused by large subjective cognitive difference of doctors, improve the robustness of image segmentation, realize fine segmentation of lung images and realize higher stability and repeatability is an important problem to be solved.

Disclosure of Invention

The embodiment of the invention provides a self-adaptive threshold segmentation method and device for a lung CT density increase shadow, and aims to solve the problems of unstable image segmentation conclusion and poor segmentation robustness in the prior art.

According to one aspect of the invention, an adaptive threshold segmentation method for lung CT density increase shadow is provided, which comprises the following steps:

acquiring a lung CT image with a spot-shaped density increase shadow, and normalizing the acquired lung CT image;

binarizing the normalized image to obtain a binarized mask of the lung parenchyma, and extracting a lung image based on the binarized mask of the lung parenchyma to obtain a lung parenchyma mask;

identifying a closed contour marked on the normalized lung CT image by an image marker, wherein the closed contour comprises a focus area;

generating a contour binarization mask according to the marked closed contour, and carrying out logical AND operation with the lung parenchyma mask to generate a focus mask;

extracting corresponding pixel points in the normalized lung CT image by using a focus mask, and performing histogram statistics to obtain a pixel value with the highest histogram intensity;

summing the pixel value with the highest histogram intensity with a predetermined positive integer offset value as a split threshold;

binarizing the normalized lung CT image within the closed contour using the separation threshold to obtain a lesion region mask.

In some embodiments of the present invention, after binarizing the normalized lung CT image within the closed contour using the separation threshold, the method further includes: preprocessing the normalized lung CT image in the binarized closed contour, wherein the preprocessing comprises the following steps: small connected domain removal, dilation operation and/or hole filling operation.

In some embodiments of the invention, the closed profile is: polygonal, circular or elliptical.

In some embodiments of the invention, the obtaining CT images of the lungs comprises obtaining CT images that satisfy the following criteria: 1) a lung CT image with a patch-like density enhancement image; and 2) the presence of frosty lesions in CT sections of each slice of the same patient;

in some embodiments of the present invention, the normalizing the acquired lung CT images comprises: setting the window width and the window level of a lung window as a preset window width value and a preset window level value respectively, and normalizing voxel values in the window to be in a range of 0-255; and removing images of regions outside the lungs.

In some embodiments of the invention, the predetermined lung window width value is 1800 HU; the predetermined window bit value is 300 HU.

In some embodiments of the present invention, after two masks of the lesion area are calculated for the first closed contour and the second closed contour labeled by the two markers, the first segmentation area and the second segmentation area are calculated based on the two masks of the lesion area, and the intersection-ratio index of the segmentation stability of the lesion area is calculated based on the first segmentation area and the second segmentation area by using the following formula:

The IOU represents the cross-over ratio index, and C1 and C2 are the first division area and the second division area respectively.

In some embodiments of the invention, the closed contour satisfies the following condition: comprises an intact lesion; normal tissue was included as background.

In some embodiments of the invention, the normal tissue is no less than 1/3 of the contour; and/or merging the vascular structure with a diseased part in the case that the vascular structure with the preset brightness or above passes through the lesion, or avoiding the vascular structure when marking the contour.

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

In another aspect of the present invention, a semi-active threshold segmentation apparatus for a lung CT density enhancement is provided, which includes a processor and a memory, wherein the memory stores computer instructions, and the processor is configured to execute the computer instructions stored in the memory, and when the computer instructions are executed by the processor, the apparatus implements the steps of the method as described above.

In another aspect of the present invention, a computer-readable storage medium is also provided, on which a computer program is stored, which computer program, when being executed by a processor, carries out the steps of the method as set forth above.

The self-adaptive threshold segmentation method and device for the lung CT density increase shadow can improve the robustness of image segmentation, realize fine segmentation of lung images and realize higher stability and repeatability.

Additional advantages, objects, and features of the 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 invention. The objectives and other advantages of the invention will be realized and attained by the structure particularly pointed out in the written description and claims hereof as well as the appended drawings.

It will be appreciated by those skilled in the art that the objects and advantages that can be achieved with the present invention are not limited to the specific details set forth above, and that these and other objects that can be achieved with the present invention will be more clearly understood from the detailed description that follows.

Drawings

The accompanying drawings, which are included to provide a further understanding of the invention and are incorporated in and constitute a part of this application, illustrate embodiment(s) of the invention and together with the description serve to explain the principles of the invention.

FIG. 1 is a flowchart illustrating a method for adaptive threshold segmentation of CT density enhancement contrast in lung according to an embodiment of the present invention.

Fig. 2A and 2B are examples of a normalized lung CT image and its global histogram, respectively, in an embodiment of the invention.

Fig. 3A and 3B are schematic diagrams of a binarized mask of a lung CT image and an extracted lung parenchyma image obtained in an embodiment of the present invention, respectively.

Fig. 4A and 4B are CT images with healthy and frosted areas marked and histogram examples of the two areas, respectively.

Fig. 5A and 5B are CT images with healthy and frosted areas marked and histogram examples of the two areas, respectively.

Fig. 6A and 6B are examples of a CT image in which two closed contours each including a lesion region are marked and histograms of the two regions, respectively.

Fig. 7A and 7B are schematic diagrams respectively illustrating a roughly marked closed contour in a CT image and a density-increasing lesion proposed based on an adaptive threshold on the basis of the closed contour according to an embodiment of the present invention.

Fig. 8A and 8B are schematic diagrams illustrating lung image segmentation based on a conventional manual segmentation method and a semi-active threshold phoenix method according to the present invention, respectively.

Fig. 9 shows schematic diagrams of 3 lesions obtained by two physicians based on the conventional manual segmentation and the semi-active method according to the present invention.

Detailed Description

In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention will be described in further detail with reference to the following embodiments and accompanying drawings. The exemplary embodiments and descriptions of the present invention are provided to explain the present invention, but not to limit the present invention.

It should be noted that, in order to avoid obscuring the present invention with unnecessary details, only the structures and/or processing steps closely related to the solution according to the present invention are shown in the drawings, and other details not so related to the present invention are omitted.

It should be emphasized that the term "comprises/comprising" when used herein, is taken to specify the presence of stated features, elements, steps or components, but does not preclude the presence or addition of one or more other features, elements, steps or components.

The embodiment of the invention provides a semi-active focus segmentation method, which can be used for calculating a segmentation threshold and extracting a lesion region based on a region which is manually drawn by a doctor and contains a focus and a background aiming at a spot-shaped inflammatory density increase shadow of lung CT, extracting a segmentation target which is stable to the greatest extent in the drawing of different doctors, and effectively avoiding unstable segmentation conclusion caused by large subjective cognitive difference of the doctor.

FIG. 1 is a flowchart illustrating a semi-active threshold segmentation method for lung CT density enhancement in an embodiment of the present invention. As shown in fig. 1, the method includes steps S110 to S170, wherein:

in step S110, a lung CT image is acquired, and the acquired lung CT image is normalized.

Existing CT scanners can be used to acquire CT images of the lungs, for example, low dose helical CT scanners can be used to acquire CT images of the lungs.

In an embodiment of the present invention, a CT image of a mottled sheet-like density-increased image (e.g., a ground glass image, a confounding image, and/or a real deformation image) is preferably selected as a subject of study. Therefore, a CT image meeting the following conditions is taken as the CT image of the lung to be studied: (1) speckle-like density-increasing shadows appear on thin-layer chest CT images, including frosted glass shadows, hybrid frosted glass shadows, and/or real metamerism. (2) The CT slices of each layer of the same patient have plaque-like density-increasing shadow lesions. Accordingly, CT images of the following cases are not considered as study objects: (1) no spot-shaped density-increasing image is seen on the thin-layer chest CT examination; (2) the CT image of the patient has large-area diffuse lesion, other combined lesion or serious motion artifact.

In an embodiment of the present invention, after obtaining the lung CT image of the patient, it is preferable to perform a normalization process, such as gray value normalization, on the raw lung CT image for facilitating subsequent data processing. The original CT image generally has larger bit depth and large numerical range, and is not beneficial to analysis and visualization of an interested region, so that the lung window width and the window level are respectively set as a preset lung window width value and a preset window level value in the normalization processing, and the voxel values in the window are normalized to be in the range of 0-255. For example, the predetermined lung window width value and the predetermined window level value may be set to 1800HU and-300 HU, respectively.

In addition, since the interference of the structures around the lung, such as blood vessels, bronchi, examining tables, etc., not only increases the amount of irrelevant data, but also interferes with the histogram, labeled contour, etc., the embodiment of the present invention also removes the regions outside the lung.

The lung parenchyma of normal people is filled with air and has small density, and is obviously different from the surrounding rib tissues, blood vessels, heart and other organs. The difference in voxel CT values (Hounsfield Unit) is shown in the CT examination, the mean CT value of healthy adult whole lung at the end of deep inspiration is-886 + -22 HU, while the CT value of the density-increased area is often greater than-600 HU, showing as a higher-intensity area.

In order to better analyze and visualize the lung CT image, the invention carries out histogram conversion processing on the CT image. Fig. 2A and 2B illustrate exemplary normalized CT images and the global histogram distribution of the normalized CT images, respectively. As shown in FIGS. 2A and 2B, healthy lungs are filled with air, the gray values of which are mostly concentrated within 100(-480 HU). In fig. 2B, a peak around 150 gradation values indicates a density increasing region, and corresponds to the region a in fig. 2A.

And step S120, carrying out binarization on the normalized CT image to obtain a binarization mask of the lung parenchyma, and extracting the lung image based on the binarization mask of the lung parenchyma to obtain a lung parenchyma mask.

In this step, an empirical threshold may be selected based on the histogram of the normalized CT image, and binarization may be performed on the normalized CT image if the empirical threshold is selected to be 100. And then sequentially carrying out the steps of removing the boundary, taking a communication domain with a large area, carrying out morphological closed operation, filling holes and the like, and finally obtaining the binary mask of the lung parenchyma. The morphological closing operation and the hole filling are respectively used for filling small gaps so as to form a complete binary mask. The lung parenchymal-based binary mask can be used for extracting a lung image from the normalized CT image to obtain a lung parenchymal mask (lung parenchymal image). The binarized mask of the lung parenchyma and the extracted lung parenchyma image are shown in fig. 3A and 3B, respectively.

In step S130, a closed contour is labeled on the normalized lung CT image with the healthy lung parenchyma as a background and the lesion region (or called lesion region) as a foreground, so that the closed contour includes the lesion region.

For example, a person who performs contour labeling on a CT image by a physician or the like marks a closed contour on the normalized CT image of the lung with the healthy lung parenchyma as the background and the lesion region (or called lesion region) as the foreground, so that the marked closed contour includes the lesion region, and is used for performing segmentation of the lung parenchyma, that is, segmenting a region of increased plaque inflammatory density in the CT image of the lung, where the region of increased plaque inflammatory density is a region of interest (ROI).

Unlike the prior art, in this step, the contour labeling of the CT image may be a rough labeling, and a more accurate segmentation of the lung parenchyma may be achieved based on the following steps.

Fig. 4A and 4B are an example of a CT image in which a healthy region and a plaque-like pure ground glass shadow region are marked, and histograms of the two regions, respectively. Fig. 5A and 5B are a CT image in which a healthy region and a hybrid ground glass shadow region are marked, and an example of histograms of the two regions, respectively. As shown in fig. 4A, the left and right side labeled outlines are typical local areas of lesion and healthy local areas, respectively, wherein the lesion of fig. 4A is a patch-like pure ground glass density shadow. As shown in fig. 5A, the left and right labeled contours are typical local lesion areas and healthy local areas, respectively, wherein the lesion in fig. 5A is a mixed frosted glass density shadow with a large real variation component, and the common point in fig. 4A and 5A is that the gray scale value of almost all pixels is lower than 200 (corresponding to HU value 205). The histogram of the healthy lung parenchyma region (not including the bronchus) is a unimodal structure of Poisson-like distribution, the gray level peak is about 20-30 (-1060HU to-990 HU), and the gray level value of more than 95% of pixels is lower than 50 (-850 HU). And the gray level histogram containing the lesion area shows obvious fat tail effect, the number of pixels with high gray level is increased, and the number of pixels with low gray level is reduced. The lesion area of fig. 5A has more real components, and local peaks appear in the high gray value areas on the histogram, however, the global peak is still around 20-30, and the global peak is slightly shifted to the right by less than 5(35HU) compared to the healthy area.

The segmentation of the lung lesions in fig. 4A and 5A is a rough segmentation with healthy lung parenchyma as background and lesion areas as foreground. The left-side closed contour in both of the CT slices of fig. 4A and 5A does not depict the fine structure of the lesion, and therefore includes a portion of the low density region, so that the position of the global intensity peak does not change significantly. Meanwhile, the shape of the high-density lesion shadow is irregular, the edge is not clear, and a fine structure is difficult to draw. Since the rough segmentation is difficult to draw a fine structure, in the prior art, an image specialist with abundant experience is often required to perform the segmentation to accurately mark a focus region as much as possible without including a low-density region corresponding to a healthy part as much as possible, so that the conventional image segmentation is not only dependent on experience too much, but also inter-group errors caused by sketching (labeling) of different labeling personnel and intra-group errors caused by two sketching of the same labeling personnel are both large, and the accuracy, stability and repeatability of focus contour drawing are greatly influenced and difficult to guarantee.

As can be seen from the histogram analysis of the lesion and the healthy lung parenchyma, if the lesion contour is roughly described, the peak position in the gray histogram including the pixel points is still approximately close to the healthy lung parenchyma, and if the lesion contains a real-variant portion, a small local peak may be formed in a high gray value region, and the probability distribution thereof is much smaller than the pixel value near the background.

To confirm this determination, both of the closed polygon outlines depicted in fig. 6A contain the entire area of the lesion, except that the areas of the two containing the background area of the lung parenchyma are different. As shown in the histogram of the two closed polygon outlines in fig. 6B, the gray value curves almost coincide at the portion higher than 50, the positions of the global peaks are almost the same, and the intensities are different. In general, the second contour includes more lung parenchyma regions than the first contour, so the histogram integral of the second contour region is larger than that of the first contour region, and the increased portion corresponds to the lung parenchyma, so the absolute value of the gray peak is raised and the position is not changed.

As can be inferred from the above, similar histogram distributions, i.e., the high-tone values are almost overlapped, the global peak positions are overlapped, and the absolute intensities are different, can be obtained for the contour independently drawn by a plurality of physicians as long as the contour includes the entire range of the lesion to be studied, and even if the contour includes a partial healthy lung parenchyma range. Formally based on the conclusion, the invention designs an adaptive threshold segmentation method. Therefore, in the embodiment of the present invention, although the closed contour is roughly marked on the CT image in this step, the accurate segmentation and delineation of the lesion region can be achieved by the following adaptive threshold segmentation step.

In step S140, a contour binarization mask is generated according to the labeled closed contour, and a logical and operation is performed with the lung parenchyma mask, thereby generating a lesion mask.

That is, in this step, binarizing the labeled closed contour 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 may be generated.

In step S150, a focus mask is used to extract corresponding pixel points in the normalized lung CT image and perform histogram statistics, thereby obtaining a pixel value with the highest histogram intensity.

That is, in this step, a focus area histogram is obtained based on corresponding pixel points in the lung CT image corresponding to the focus mask, and statistics is performed on the histogram to obtain a pixel value with the highest histogram intensity, where the highest pixel value corresponds to a high-gray-value pixel of the lesion area.

In step S160, the pixel value with the highest histogram intensity is summed with a predetermined positive integer offset value as the partition threshold.

In an embodiment of the present invention, the positive integer offset value may be selected from an empirical value of 15-25, such as 15.

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

Optionally, after binarizing the normalized lung CT image within the closed contour by using the separation threshold, preprocessing may be further performed on the normalized lung CT image within the binarized closed contour, where the preprocessing includes: small connected domain removing, expanding operation and/or hole filling operation and the like. The small connected domain removal is used for obtaining a connected domain with a large area, the expansion operation is used for removing edge burrs, so that the edge is smooth as much as possible, and the filling of the hole is used for removing a small hole in the image.

FIG. 7A shows a roughly labeled closed contour 70A in a CT image in accordance with one embodiment of the present invention. Fig. 7B shows a schematic diagram of a densitometric lesion 70B based on an adaptive threshold based on the closed contour. As shown in fig. 7A and 7B, a fine mask of the lesion area may be obtained based on the marked rough closed contour.

As above steps S140-S170 belong to a semi-active thresholding step, or adaptive thresholding step, which obtains a finer lesion region mask by determining a partition threshold based on the pixel value with the highest histogram intensity of the CT image in the roughly labeled closed contour and a predetermined positive integer offset value, and more finely partitioning the region in the closed contour based on the partition threshold. That is, the semi-active thresholding step of embodiments of the present invention enables more accurate determination of the lesion region.

Compared with the prior art, the semi-active threshold segmentation method for the lung CT density increase shadow can improve the robustness of image segmentation, realize fine segmentation of lung images and realize higher stability and repeatability. The semi-active threshold segmentation method is simple, quick and easy to realize, and greatly improves the stability of the segmentation result.

Next, the stability of the segmentation result of the present invention is verified by using an Intersection-over-Union (IOU) index. More specifically, two lesion masks may be calculated for two closed contours labeled by two physicians, a first segmentation area and a second segmentation area may be calculated based on the two lesion masks, and an intersection ratio index of lesion segmentation stability may be calculated based on the first segmentation area and the second segmentation area by using an intersection ratio index formula. In the embodiment of the invention, the stability of the segmentation result is evaluated by an IOU index:

the IOU represents the cross-over ratio index, and 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.

For comparison, the invention also provides the IOU index range obtained by adopting the traditional manual segmentation mode. At this time, C1 and C2 are the distribution of the lesion covered by the outlines drawn by the two physicians, respectively.

For convenience of description, the IOU obtained based on the method of the present invention is referred to as a "man-machine" IOU, and the IOU obtained based on the conventional manual segmentation manner is referred to as a "man-human" IOU. In global statistics, the 95% confidence interval estimation of the segmentation results is done using a non-parametric bootstrap method.

Fig. 8A and 8B are schematic diagrams illustrating lung image segmentation based on a conventional manual segmentation method and a semi-active threshold segmentation method according to the present invention, respectively.

The embodiment of the invention provides 200 lesion cross sections from a plurality of thoracic cavity CT image slices, a traditional manual segmentation mode and a semi-active segmentation mode of the invention are adopted for each lesion cross section, and two contour lines (a contour 1 and a contour 2) are respectively drawn by two doctors (a doctor A and a doctor B) in each method. The stability of the contouring was evaluated by the IOU index.

For a manual segmentation mode, accurate division of a focus region of interest (ROI) plays a key role in later-stage feature engineering design and statistical analysis. In order to reduce human errors (including intra-group errors caused by two times of drawing by the same marking person and inter-group errors caused by drawing by different drawing persons) and omission of focus edge information caused by manually drawing the ROI, a labelme tool can be used as a marking tool, and closed contours can be drawn by using polygonal, rectangular, circular and other shape tools, so that the marking of the ROI is realized. In manual segmentation procedures, the physician tries to select the exact region of interest based on experience. Since the conventional manual segmentation process is the prior art, it is not described herein again.

In the semi-active threshold segmentation process, a labelme tool can be used as a marking tool, and other delineation tools can also be used as the marking tool. Two doctors can respectively and independently label the same CT image sample set in the same sequence. For a ground glass lesion area in a certain slice of a CT image, a doctor A traces an edge contour of a ground glass shadow containing a focus based on subjective recognition; physician B also draws an edge profile that contains the lesion. The contours drawn by physician a and physician B are somewhat arbitrary, as long as they encompass the entire range of the lesion. Furthermore, the drawn outline may preferably contain some normal lung tissue as a background (the background is not less than 1/3, for example, 1/3 is only an example, and the invention is not limited thereto). In addition, if a vessel structure with a high brightness (e.g., a blood vessel with a brightness higher than a predetermined brightness) passes through the lesion, the blood vessel portion may be merged with the lesion portion, or the contour may be drawn to avoid the blood vessel structure. Based on the rough contours drawn by both physicians, finer lesion structures can be further obtained by an adaptive thresholding step. Based on the obtained structure, lesion distributions C1 and C2 segmented by adaptive thresholding according to contours drawn by two physicians can be calculated.

Fig. 9 shows lesion area masks based on manual rendering (segmentation) and adaptive rendering (segmentation) of the present invention, respectively. As shown in fig. 9, the lesion regions obtained by manual segmentation by different physicians for the same lesion are greatly different, and thus it is difficult to extract the lesion structure stably and finely. By adopting the semi-active threshold segmentation method (or called as lesion structure segmentation method) of the invention, no matter doctor A or doctor B carries out rough contour drawing, a finer and highly consistent lesion structure can be obtained finally.

Based on the above experiments, the IOU index obtained by the semi-active threshold segmentation method provided by the invention on the test sample is 0.81 (95% confidence interval: 0.78, 0.84), which is greatly superior to the subjective segmentation result of the physician. The lung density-increasing lesion represented by frosted glass shadow is one of the main lesions of lung lesion, and the fuzzy edge brings great inconvenience to doctors for quantitatively analyzing the lesion volume, for example, IOU obtained by subjective marking of two doctors using the traditional manual segmentation method is only 0.51 (95% confidence interval: 0.51, 0.58), according to different styles of the doctors. In other words, by means of the designed semi-active threshold segmentation method, a fine focus structure can be stably and robustly extracted and used as a basis for quantitatively evaluating the state of an illness, and the method has foundation significance for precise and personalized treatment.

Therefore, the self-adaptive threshold segmentation method for the lung CT density increase shadow is a quick method, and achieves the focus segmentation effect of maximally inhibiting human factors. The method has important clinical significance, patients possibly visit a plurality of doctors to evaluate the disease progress or the disease grading in the whole disease course, and the stability problem of the disease evaluation under the cooperation of a plurality of persons is taken as a primary problem to be urgently solved in order to realize the ambitious goal of modern precise and personalized treatment. For example, the progressive change of the volume of a certain local affected tissue is one of the key indexes for quantitatively evaluating the degree and speed of disease progression, and the lack of the segmentation standard of the edge fuzzy lesion causes huge variance of the manual segmentation result, so that the reference value is lost. Therefore, the method has strong practical significance for the work of quantitative evaluation of disease progress, quantitative evaluation of disease quality and malignancy, stable extraction of disease statistical modeling parameters and the like.

The method is a robust semi-active self-adaptive threshold segmentation method, can realize stable and fine segmentation on the patch density increase image with fuzzy edges in the lung CT image, and clears obstacles for quantitative evaluation and statistical modeling in the future.

In accordance with the foregoing method, the present invention further provides a semi-active thresholding device for pulmonary CT densitometric enhancement, comprising a processor and a memory, the memory having stored therein computer instructions for executing computer instructions stored in the memory, the device implementing the steps of the foregoing method when the computer instructions are executed by the processor.

The present invention also relates to a storage medium on which computer program code may be stored, which when executed may implement various embodiments of the method of the present invention, and which may be a tangible storage medium such as an optical disk, a Random Access Memory (RAM), a memory, a Read Only Memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a hard disk, a removable disk, a CD-ROM, or any other form of tangible storage medium known in the art.

It is to be understood that the invention is not limited to the specific arrangements and instrumentality described above and shown in the drawings. A detailed description of known methods is omitted herein for the sake of brevity. In the above embodiments, several specific steps are described and shown as examples. However, the method processes of the present invention are not limited to the specific steps described and illustrated, and those skilled in the art can make various changes, modifications and additions or change the order between the steps after comprehending the spirit of the present invention.

Those of ordinary skill in the art will appreciate that the various illustrative components, systems, and methods described in connection with the embodiments disclosed herein may be implemented as hardware, software, or combinations thereof. Whether this is done in hardware or software depends upon the particular application and design constraints imposed on the solution. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present invention. When implemented in hardware, it may be, for example, an electronic circuit, an Application Specific Integrated Circuit (ASIC), suitable firmware, plug-in, function card, or the like. When implemented in software, the elements of the invention are the programs or code segments used to perform the required tasks. The program or code segments may be stored in a machine-readable medium or transmitted by a data signal carried in a carrier wave over a transmission medium or a communication link. A "machine-readable medium" may include any medium that can store or transfer information. Examples of a machine-readable medium 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, and so forth. The code segments may be downloaded via computer networks such as the internet, intranet, etc.

It should also be noted that the exemplary embodiments mentioned in this patent describe some methods or systems based on a series of steps or devices. However, the present invention is not limited to the order of the above-described steps, that is, the steps may be performed in the order mentioned in the embodiments, may be performed in an order different from the order in the embodiments, or may be performed simultaneously.

Features that are described and/or illustrated with respect to one embodiment may be used in the same way or in a similar way in one or more other embodiments and/or in combination with or instead of the features of the other embodiments in the present invention.

The above description is only a preferred embodiment of the present invention, and is not intended to limit the present invention, and various modifications and changes may be made to the embodiment of the present invention by those skilled in the art. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for adaptive threshold segmentation of pulmonary CT densitometric enhancement images, the method comprising the steps of:

acquiring a lung CT image with a spot-shaped density increasing image, and normalizing the acquired lung CT image;

generating a contour binarization mask according to the marked closed contour, and performing logical AND operation with the lung parenchyma mask to generate a focus mask;

extracting corresponding pixel points in the normalized lung CT image by using a focus mask and performing histogram statistics to obtain a pixel value with the highest histogram intensity;

2. The method of claim 1, wherein after binarizing the normalized lung CT image within the closed contour using the separation threshold, further comprising:

preprocessing the normalized lung CT image in the binarized closed contour, wherein the preprocessing comprises the following steps: small connected domain removal, dilation operation and/or hole filling operation.

3. The method of claim 1,

the closed contour is: polygonal, circular or elliptical.

4. The method of claim 1, wherein the normalizing the acquired lung CT images comprises:

setting the window width and the window level of the lung window as a preset lung window width value and a preset window level value respectively, and normalizing voxel values in the window to be in an interval range of 0-255; and

images of the regions outside the lungs are removed.

5. The method of claim 1,

the predetermined lung window width value is 1800 HU;

the predetermined window bit value is-300 HU.

6. The method of claim 1, further comprising:

calculating two masks of the lesion area aiming at the first closed contour and the second closed contour marked by the two markers respectively, calculating a first segmentation area and a second segmentation area respectively based on the two masks of the lesion area, and calculating an intersection ratio index of the segmentation stability of the lesion area based on the first segmentation area and the second segmentation area by using the following formula:

7. The method of claim 1, wherein the closed contour satisfies the following condition:

comprises an intact lesion;

normal tissue was included as background.

8. The method of claim 6,

1/3 where the normal tissue is not less than the contour; and/or

In the case where a vascular structure having a brightness above a predetermined brightness passes through a lesion, the vascular structure is merged with a lesion portion, or the vascular structure is avoided when a contour is labeled.

9. The method of claim 1,

the positive integer offset value ranges from 15 to 25.

10. An apparatus for adaptive threshold segmentation of CT densitometric shadows of the lungs, comprising a processor and a memory, wherein the memory has stored therein computer instructions for executing computer instructions stored in the memory, which when executed by the processor, performs the steps of the method of any of claims 1-9.

11. A computer-readable storage medium, on which a computer program is stored which, when being executed by a processor, carries out the steps of the method according to any one of claims 1 to 9.