Pulmonary nodule intraoperative positioning system based on flexible array type sensor

Abstract

The invention belongs to the technical field of medicine, and discloses a pulmonary nodule intraoperative positioning system based on a flexible array sensor, which comprises the flexible array sensor arranged at a to-be-detected part of a subject and used for collecting a compression signal and obtaining stress distribution; the nodule area detection unit is used for determining a nodule area through clustering analysis according to the stress distribution; the morphological optimization unit is used for performing morphological optimization on the determined nodule region; the feature extraction unit is used for extracting related geometric features based on the optimized nodule region; and the visualization unit is used for visually displaying the stress distribution, the optimized nodule region and related geometric features. The method can assist doctors in accurately positioning the pulmonary nodules, and meanwhile, quantitative features of the nodules are provided.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of medicine, and relates to a lung nodule intraoperative positioning system, in particular to a lung nodule intraoperative positioning system based on a flexible array sensor. BACKGROUND

[0002] In the early diagnosis and treatment of lung cancer, accurate positioning of small lung nodules that are difficult to touch during operation but found in preoperative images is the key to the success of minimally invasive surgery. To solve this problem, various technical routes have been developed in clinical practice, including the following positioning paradigms:

[0003] The first paradigm: preoperative invasive physical markers, represented by CT-guided Hook-wire; this is the current most widely used clinical "gold standard"; the core idea is to implant a metal hook wire or other physical signal into the lung through percutaneous puncture under CT guidance before operation, and anchor it near the nodule to provide path guidance during operation; the fundamental defect of this paradigm is its "invasive" nature, and the puncture process itself brings significant risks of pneumothorax, bleeding, infection, marker shedding and displacement, increasing the pain and medical costs of patients.

[0004] The second paradigm: intraoperative real-time surface detection, represented by intraoperative ultrasound (IOUS); this paradigm attempts to solve the above problems through non-invasive methods. However, due to its physical principles, there are still several problems in the application of IOUS to lung nodules: (1) extreme sensitivity to residual gas, which is the most core problem in the application of IOUS to the lung; even if the lung has collapsed, there may still be a small amount of residual gas in the alveoli or lung bullae. These tiny bubbles are the "nemesis" of ultrasound waves, causing strong acoustic shadowing or comet-tail artifacts, completely blocking the tissue behind, resulting in a large area of detection blind area, making it impossible to find nodules; (2) the invisibility of "isoechoic lesions", a significant portion of lung nodules (especially ground-glass nodules) have a tissue density very close to the surrounding collapsed lung parenchyma, resulting in little difference in acoustic impedance; on the ultrasound image, such nodules will appear as "isoechoic" with the same brightness as the background tissue, making them "invisible" on the ultrasound image and unable to be identified; (3) depth dependence and resolution limitation, high-frequency probes have high resolution but limited penetration depth; for deeply buried nodules (e.g., >2-3 cm from the lung surface), ultrasound signals are severely attenuated, and image quality will decrease dramatically, making it difficult to effectively detect; (4) highly dependent on operator experience, the image quality of IOUS is far from that of abdominal ultrasound, and surgeons need to have high ultrasound operation skills and image diagnosis experience to interpret images full of noise and artifacts, which makes the learning curve of this technology steep, and the consistency and reliability of the results are poor, making it difficult to standardize and popularize in hospitals at all levels.

[0005] The third paradigm: high-tech internal navigation through natural orifices, represented by electromagnetic navigation bronchoscopy (ENB); this paradigm represents another high-tech minimally invasive approach, like a "GPS" for the lung, which guides the catheter through the bronchial tree to the lesion. Its fundamental flaw is the "mismatch between static data and dynamic reality". The navigation path of ENB is completely based on preoperative static CT images, while factors such as patient position, anesthesia, and lung collapse during surgery can cause unpredictable deformation of lung tissue, causing the so-called "CT-to-body Divergence" (CT-to-body registration deviation) of the navigation system, thereby affecting the final positioning accuracy. In addition, its equipment and consumables are extremely expensive and complex to operate, limiting its widespread application.

[0006] In summary, the prior art either brings safety risks due to invasiveness (Hook-wire), or has insufficient reliability due to physical principle limitations (IOUS), or is difficult to popularize due to system complexity, high cost and registration deviation (ENB). There is currently a lack of an intraoperative positioning solution that can simultaneously meet the requirements of non-invasiveness (to lung parenchyma), real-time, simple operation, low cost, and direct feedback based on tissue mechanical properties. SUMMARY

[0007] The purpose of the present application is to address the shortcomings of the prior art and provide a lung nodule intraoperative positioning system based on a flexible array sensor, which directly senses the hardness difference between the nodule and normal tissue through a flexible array sensor, and combines clustering analysis, morphological optimization, etc., to assist doctors in achieving accurate positioning of lung nodules, while providing quantitative features of the nodules.

[0008] To achieve the above purpose, the present application adopts the following technical solutions.

[0009] The present application provides a lung nodule intraoperative positioning system based on a flexible array sensor, which comprises:

[0010] A flexible array sensor arranged at a detection site of a subject for collecting a pressure signal and obtaining a stress distribution;

[0011] A nodule region detection unit for determining a nodule region through clustering analysis based on the stress distribution;

[0012] A morphological optimization unit for morphological optimization of the determined nodule region;

[0013] A feature extraction unit for extracting relevant geometric features based on the optimized nodule region;

[0014] A visualization unit for visualizing and displaying the stress distribution, the optimized nodule region and the relevant geometric features.

[0015] In one implementation, the flexible array sensor is composed of a plurality of pressure sensors arranged in an array on a flexible material, with each pressure sensor corresponding to a node. Further, the flexible array sensor is preferably an electronic skin.

[0016] In one implementation, based on the pressure signal collected by the flexible array sensor, a pressure concentration area is determined, and a two-dimensional distribution of pressure increments of the pressure concentration area is taken as the stress distribution.

[0017] In one implementation, the nodule region detection unit uses one of the K-means algorithm, Gaussian mixture model and DBSCAN algorithm for clustering analysis.

[0018] Specifically, the application adopts a Gaussian mixture model for clustering analysis, automatically divides stress data into normal and abnormal categories, generates an initial nodule mask, and further preliminarily determines a nodule region.

[0019] The Gaussian mixture model is represented as:

[0020] ;

[0021] Wherein, represents the probability density at the data point ; represents the mixing weight of the kth Gaussian distribution; represents a Gaussian distribution with a mean of and a covariance of .

[0022] Further, in order to further improve the detection accuracy of the nodule boundary, the application introduces frequency domain analysis to correct the initial clustering result. The original stress signals of the pressure sensor nodes near the nodule region boundary are converted to the frequency domain, and the nodule region boundary is corrected based on frequency domain analysis.

[0023] In a specific implementation, based on the nodule boundary determined by the initial nodule mask obtained by clustering analysis, the original stress signals collected by the pressure sensor nodes in the 3-5 pressure sensor node width annular region inside and outside the boundary contour are extracted; the original stress signal of each pressure sensor node is represented as ;

[0024] The correction of the nodule region boundary based on frequency domain analysis includes the following steps:

[0025] (1) Fast Fourier transform (FFT) is performed on the original stress signal to obtain the signal in the frequency domain :

[0026] ;

[0027] Wherein, represents the frequency; represents the pressure amplitude of the nth sampling point; N represents the total number of sampling points;

[0028] (2) Extract the main frequency of each pressure sensor node , that is, the frequency component with the highest energy in the power spectral density;

[0029] ;

[0030] (3) Perform global statistics on all pressure sensor nodes to calculate the global main frequency mean :

[0031] ;

[0032] wherein X, Y represent the number of rows and columns in the array sensor respectively;

[0033] (4) Calculate the standard deviation of the global dominant frequency :

[0034] ;

[0035] (5) Set the boundary determination frequency domain anomaly threshold:

[0036] ;

[0037] (6) For each pressure sensor node in the boundary region, if its dominant frequency , mark the pressure sensor node as an abnormal point, and adjust the nodule mask boundary; otherwise, the boundary is not adjusted, that is, the pressure sensor node is not adjusted, and the node mask is unchanged.

[0038] This helps to eliminate false classification caused by noise or local variation, and improve the boundary degree.

[0039] In an implementation manner, the morphological optimization unit comprises:

[0040] a closing operation module, configured to perform closing operation on the determined nodule region to fill small holes inside the nodule region;

[0041] an opening operation module, configured to eliminate isolated noise points outside the nodule region.

[0042] The closing operation module, for the nodule region inside, first performs expansion processing, and then performs corrosion processing, connects adjacent abnormal points, and fills small holes inside.

[0043] The opening operation module, for the nodule region processed by the closing operation unit, first performs corrosion processing, and then performs expansion processing, for eliminating isolated noise points outside the nodule region.

[0044] In an implementation manner, the geometric features extracted by the feature extraction unit include at least one of area, perimeter, equivalent diameter and circularity. In the present application, through connected domain analysis, first, based on the set area threshold, the effective nodule region with an area greater than the set threshold is screened out, and then the quantitative geometric features of the effective nodule region are calculated.

[0045] In an implementation manner, the lung nodule intraoperative positioning system based on the flexible array sensor further comprises a physical deformation correction unit for correcting the equivalent diameter.

[0046] Specifically, the physical deformation correction unit corrects the equivalent diameter by a correction function according to the nodule depth value, which is expressed as:

[0047] ;

[0048] wherein, represents the correction function; the correction function is determined by polynomial regression or by a lookup table.

[0049] The polynomial regression is quadratic polynomial regression; the correction function determined by the quadratic polynomial regression is expressed as:

[0050] ;

[0051] wherein, represents the corrected true diameter; represents the equivalent diameter before correction; represents the nodule depth value; represents the regression fitting coefficient of i-th power of and j-th power of d. It can be calculated by the least square method.

[0052] The correction function determined by the lookup table by bilinear interpolation is expressed as:

[0053] ;

[0054] wherein, represents the corrected true diameter; l , , represents the equivalent diameter currently calculated, represents the minimum value of the given equivalent diameter, represents the given diameter step, represents the current given nodule depth, represents the minimum value of the given nodule depth, represents the given depth step, represents the floor function; (h, l) represents the grid point position determined based on the current equivalent diameter; represents the true diameter experimental value at the grid point (l, h) determined by the lookup table, and the lookup table is expressed as , represents the grid size of the lookup table, represents the l-th measured diameter, represents the h-th depth; is a relative coordinate parameter determined by the grid point (l, h), .

[0055] Compared with the prior art, the lung nodule intraoperative positioning system based on the flexible array sensor has the following beneficial effects:

[0056] (1) The detection reliability and application range are fundamentally improved: the application adopts a new mechanical sensing principle, and positioning is performed by measuring the hardness difference of the tissue; hardness is a solid mechanics characteristic, is almost not affected by trace gas, and most lung nodules (especially malignant lesions) have significant hardness difference with normal lung parenchyma; therefore, the application fundamentally avoids the physical bottleneck of IOUS, significantly improves the detection success rate in a complex intraoperative environment, and expands the application range to the nodule types that cannot be detected by traditional ultrasound.

[0057] (2) The detection process is objectified and standardized, and rich quantitative information is provided: the implementation of IOUS highly depends on the experience of the operator, and the subjective interpretation of the operator to the gray-scale image full of noise leads to poor consistency of the results and a steep learning curve; moreover, the application converts the nodule identification process into an automatic and mathematical model-based objective process through Gaussian mixture model clustering and morphological optimization, can stably and repeatedly segment the nodule region from complex stress data, and completely eliminates the dependence on the "hand feeling" and "eyesight" of the operator; more importantly, the application not only realizes positioning, but also automatically outputs key quantitative parameters such as the equivalent diameter and circularity of the nodule, provides unprecedented objective data support for the intraoperative decision of the doctor (such as evaluation of the incision margin and judgment of the benign and malignant tendency), and realizes the leap from "positioning" to "quantitative analysis".

[0058] (3) Higher dynamic positioning accuracy and cost-effectiveness: the application is a real-time detection technology, which directly measures the physical characteristics of the lung tissue at the current time and in the current state, and does not have the problem of registration deviation; in addition, the application can actively calibrate the measurement error caused by the compression of the tissue through the physical deformation correction unit, and realizes high accuracy that cannot be achieved by existing surface detection technologies; in terms of cost, the system composition (flexible sensor + standard processing unit) of the application is much simpler than ENB or high-end intraoperative ultrasound equipment, has extremely high cost-effectiveness, and is conducive to popularization and promotion in hospitals at all levels.

[0059] (4) The safety is significantly improved, and the clinical process is greatly simplified: the application is completely non-invasive to the lung parenchyma, and fundamentally eliminates the safety risk caused by puncture; at the same time, the application completely integrates the positioning process into the surgical operation, and becomes a "one-stop" operation independently completed by the surgeon, greatly simplifies the clinical workflow, shortens the operation time, and reduces the pain and medical cost of the patient. BRIEF DESCRIPTION OF DRAWINGS

[0060] Figure 1 The phantom model.

[0061] Figure 2 A schematic diagram of a lung nodule intraoperative positioning system based on a flexible array sensor provided by the present application.

[0062] Figure 3 Results of nodule processing using the lung nodule intraoperative positioning system; wherein (a) is a stress distribution map, (b) is a clustering analysis result based on a Gaussian mixture model, (c) is a result of correcting the nodule region boundary based on frequency domain analysis, and (d) is a result of nodule region frequency domain projection after morphological optimization.

[0063] Figure 4 Results of nodule true diameter, equivalent diameter and corrected diameter analysis; wherein (a) is a Bland-Altman analysis result of corrected diameter and true diameter, and (b) is a comparison chart of nodule true diameter, equivalent diameter and corrected diameter.

[0064] Figure 5 Nodule detection visualization results.

[0065] Figure 6 Nodule diameter trend analysis results. DETAILED DESCRIPTION

[0066] The technical solutions of the embodiments of the present application will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of the present application.

[0067] The phantom model used in the following embodiments is a pig lung model, in which gel beads (0.25 cm-2.0 cm) are embedded, and the depth is measured (0.5 cm-3.0 cm) by scratching with a cotton swab, as shown in Figure 1 .

[0068] Embodiment 1

[0069] This embodiment provides a lung nodule intraoperative positioning system based on a flexible array sensor, as shown in Figure 2 which includes a flexible array sensor, a nodule region detection unit, a morphological optimization unit, a feature extraction unit, a physical deformation correction unit and a visualization unit.

[0070] 1. Flexible array sensor

[0071] The flexible array sensor is arranged at the detection site of the subject, used for collecting pressure signals and obtaining stress distribution.

[0072] In this embodiment, the significant difference in elastic modulus between lung nodules and normal lung parenchyma is utilized. The elastic modulus of normal collapsed lung tissue (Etissue) is low, while the elastic modulus of lung nodules (Enodule) is significantly higher, i.e. Enodule >> Etissue. Due to the greater stiffness of the nodule, its deformation is much smaller than that of the surrounding normal tissue. This results in a stress concentration zone on the surface of the flexible array sensor directly above the nodule, where the tissue is unable to effectively unload pressure.

[0073] The flexible array sensor is composed of a plurality of pressure sensors arranged in an array on a flexible material. In this embodiment, an electronic skin is used as the flexible array sensor. Specifically, the electronic skin uses the ink-present technology humanoid fingertip pressure sensor, with an array size of 12x8 (96 sensing units) and a thickness of only 0.025 mm, providing high flexibility and biocompatibility. Each pressure sensor corresponds to a node. The flexible array sensor is used to collect pressure signals and display them in real time, as shown in Figure 1

[0074] The basic physical principle of the present application is to jointly utilize the significant difference in elastic modulus between lung nodules and normal lung parenchyma on a macroscopic level and the frequency change caused by the stiffness discontinuity on a microscopic stress wave level (reflected in the increase in high-frequency components when stress waves propagate at the edge of the nodule due to scattering and reflection).

[0075] When the flexible array sensor is attached to the lung surface and a relatively uniform background pressure Papplied is applied, the tissue below the flexible array sensor will deform, collecting two-dimensional stress distribution data caused by the hardness difference of the nodule, forming an XxY stress matrix Praw. Due to the greater stiffness of the nodule, its deformation is much smaller than that of the surrounding normal tissue. This results in a stress concentration zone on the surface of the flexible array sensor directly above the nodule, where the tissue is unable to effectively unload pressure. The pressure value P(x,y) in this area will be significantly higher than the surrounding area. The pressure increment ΔP(x,y) is represented as:

[0076] ;

[0077] where, Pbackground represents the average pressure of the background area (normal tissue).

[0078] is a function related to the difference in elastic modulus, the depth of nodule burial d, and the size of the nodule S.

[0079] In this embodiment, the two-dimensional distribution of ΔP(x,y) is used as the stress distribution (as shown in Figure 3 ​(a) is shown, the position, size and shape of the nodule are inverted by precisely measuring and analyzing the two-dimensional distribution of ΔP(x, y).

[0080] 2. Nodule region detection unit

[0081] The nodule region detection unit is used to determine the nodule region by cluster analysis according to the stress distribution.

[0082] The nodule region detection unit first performs median filtering on the collected original stress matrix Praw to remove potential salt and pepper noise, and obtains a preprocessed pressure matrix Pproc.

[0083] The present application uses a Gaussian mixture model to perform unsupervised clustering on pressure distribution data, automatically divides the stress data into normal and abnormal categories, and generates an initial nodule mask. It is assumed that the pressure data is composed of two distributions: a background distribution representing normal tissue and a high stress distribution representing a nodule. The GMM models the overall pressure distribution p( ) as a weighted sum of K=2 Gaussian components.

[0084] The Gaussian mixture model is represented as:

[0085] ;

[0086] wherein, denotes the probability density at data point ; denotes the mixing weight of the kth Gaussian distribution; denotes the kth Gaussian distribution with mean and covariance .

[0087] The parameters ( , , ) are iteratively solved by the expectation maximization (EM) algorithm, and the posterior probability of each sensor node belonging to the nodule component is calculated ; if ( denotes the set posterior probability threshold), the sensor node is marked as "1" (nodule), otherwise it is marked as "0" (background), so that each data point is classified into the class with the maximum posterior probability, realizing the automatic segmentation of high stress region and background region (as shown in (b) of Figure 3 , generating a preliminary binary mask region M GMM .

[0088] In order to further improve the detection accuracy of the nodule boundary, the application introduces frequency domain analysis to correct the initial clustering result. The original stress signals corresponding to the pressure sensor nodes near the boundary of the nodule region are converted to the frequency domain, and the boundary of the nodule region is corrected based on the frequency domain analysis.

[0089] In a specific implementation, based on the nodule boundary determined by the initial nodule mask obtained by clustering analysis, the original stress signals collected by the pressure sensor nodes in the 3-5 pressure sensor node wide annular region inside and outside the boundary contour are extracted; the original stress signal of each pressure sensor node is represented as ;

[0090] The correction of the nodule region boundary based on frequency domain analysis includes the following steps:

[0091] (1) Fast Fourier transform (FFT) is performed on the original stress signal to obtain the signal in the frequency domain :

[0092] ;

[0093] Wherein, represents the frequency; represents the pressure amplitude of the nth sampling point; N represents the total number of sampling points;

[0094] (2) Extract the main frequency of each pressure sensor node , that is, the frequency component with the highest energy in the power spectral density;

[0095] ;

[0096] (3) Global statistics are performed on all pressure sensor nodes to calculate the global main frequency mean :

[0097] ;

[0098] Wherein, X and Y represent the number of rows and columns in the array sensor respectively; in this embodiment, X=12, Y=8.

[0099] (4) Calculate the global main frequency standard deviation :

[0100] ;

[0101] (5) According to the 3σ criterion, set the boundary frequency domain abnormal threshold value:

[0102] ;

[0103] (6) For each pressure sensor node in the boundary region, if its main frequency If yes, the pressure sensor node is marked as an outlier (as shown in (c) of FIG. 10), and the boundary of the nodule mask is adjusted; otherwise, the boundary is not adjusted, i.e., the pressure sensor node is not adjusted, and the node mask is not changed. Figure 3

[0104] This helps to eliminate false classification due to noise or local variation, and improves the boundary degree. For any one sensor node, if the node is in the nodule boundary region, and , then ; otherwise .

[0105] Through this step, the preliminary nodule mask is generated; for each pressure sensor node, if (belonging to the nodule component has a higher probability) and (local high-frequency component is increased), it is marked as “1” (nodule), otherwise as “0” (background), to generate the final preliminary nodule mask region .

[0106] 3. Morphological optimization unit

[0107] The morphological optimization unit is used to perform morphological optimization on the determined nodule region. In the embodiment, the nodule mask region is removed by the morphological optimization unit, and internal holes are filled.

[0108] The morphological optimization unit includes a closing operation module and an opening operation module.

[0109] The closing operation module is used to perform closing operation on the determined nodule region to fill small holes inside the nodule region.

[0110] In a specific implementation manner, the closing operation module, for the inside of the nodule region, first performs dilation processing, and then performs erosion processing to connect adjacent outlier points and fill small internal holes. This process can be represented as:

[0111] ;

[0112] represents that the preliminary nodule mask is first subjected to dilation processing, and then the dilation processing is subjected to erosion processing, which is used to fill small holes inside the nodule region. B is a preset structure element:

[0113] .

[0114] The opening operation module is used to eliminate isolated noise points outside the nodule region.

[0115] ​The opening operation module, for the nodular region processed by the closing operation unit, first performs erosion processing, and then performs inflation processing, for eliminating isolated noise points outside the nodular region. This process can be represented as:

[0116] ;

[0117] It is represented that the nodular region processed by the closing operation unit is first subjected to erosion processing, and then subjected to inflation processing, for eliminating isolated noise points outside the nodular region.

[0118] The nodular mask region after morphological optimization is projected to the thermal map in the frequency domain as shown in (d) of FIG. 6. Figure 3

[0119] 4. Feature extraction unit

[0120] The feature extraction unit is used to extract relevant geometric features based on the optimized nodular region. The geometric features extracted by the feature extraction unit include area, perimeter, equivalent diameter, and circularity.

[0121] In this embodiment, based on the final nodular mask , first, the connected domain area is calculated by connected domain labeling and calling the regionprops method; based on the set area threshold, the effective nodular region with an area greater than the set threshold is screened out; and then the quantitative geometric features of the effective nodular region are calculated.

[0122] Area (Area): refers to the total number of pixel points in the mask region.

[0123] Perimeter (Perimeter): refers to the contour length of the mask region.

[0124] Equivalent diameter (Measure Diameter): .

[0125] Circularity (Circularity): ; the circularity value ranges from 0 to 1, and the closer to 1, the more circular.

[0126] 5. Physical deformation correction unit

[0127] Because soft tissue will deform when under pressure, when the depth effect is not considered, it will cause the nodular projection size measured by the sensor on the surface to be systematically larger than its true size . Moreover, the deviation is strongly related to the buried depth d of the nodule.

[0128]

[0129] ​​To solve this problem, the application constructs a physical deformation correction unit for correcting the equivalent diameter.

[0130] Through in vitro phantom experiments (using different sizes and different depths of silicone gel to simulate nodules) or animal experiments, a set of data points is systematically collected Based on the data set, a correction function is constructed by using polynomial regression or establishing a lookup table , which is expressed as:

[0131] ;

[0132] Wherein, represents the correction function.

[0133] Here, the polynomial regression is a quadratic polynomial regression; the correction function determined by the quadratic polynomial regression is expressed as:

[0134] ;

[0135] Wherein, represents the corrected true diameter; represents the equivalent diameter before correction; represents the nodule depth value; represents the i-th power of d, the j-th power of d regression fitting coefficient. It can be calculated by the least square method.

[0136] The lookup table method establishes a discrete input-output mapping relationship based on experimental data, and realizes fast response to any input through table lookup and interpolation. This method avoids the assumption limitation of the parametric model, and directly establishes the correction relationship based on experimental data:

[0137] Lookup table construction:

[0138] ;

[0139] Wherein, represents the grid size of the lookup table, represents the lth measured diameter, represents the hth depth. The constructed lookup table is shown in Table 1.

[0140] Table 1 Nodule size correction lookup table

[0141]

[0142] Through the lookup table, the correction function determined by the bilinear interpolation is expressed as:

[0143] ; ​

[0144] wherein, represents the corrected true diameter; l , , represents the current calculated equivalent diameter, represents the given equivalent diameter minimum value, represents the given diameter step (generally 0.5 cm, which can be increased in precision as appropriate), represents the current given nodule depth, represents the given nodule depth minimum value, represents the given depth step (generally 0.5 cm, which can be increased in precision as appropriate), represents the floor function; (h, l) represents the grid point position determined based on the current equivalent diameter; represents the true diameter experimental value at the grid point (l, h) determined by the lookup table, which is represented as is a relative coordinate parameter determined by the grid point (l, h), .

[0145] Finally, the performance of the two correction methods is evaluated by Bland-Altman analysis:

[0146] ;

[0147] wherein, , ; , represents the true diameter of the qth sample and the corrected true diameter, respectively; represents the depth of the qth nodule; Q represents the number of all samples.

[0148] Figure 4 (a) in FIG. 8 shows the diameter correction effect of the 0.5 cm nodule at different depths obtained by the method of the embodiment using the correction function constructed by the above quadratic polynomial. It can be found from the figure that most of the scatter points fall within the 0.95 consistency boundary, indicating that the diameter correction result and the true result have consistency. Figure 4 (b) in FIG. 8 shows the comparison diagram of the equivalent diameter, the true diameter and the corrected diameter of the nodule. Overall, the corrected diameter is closer to the true diameter.

[0149] 6. Visualization unit

[0150] The visualization unit is used to visualize the stress distribution, the optimized nodule region and the related geometric features.

[0151] The stress distribution refers to the two-dimensional distribution or / and three-dimensional distribution of the real-time measured stress data, such as Figure 5as shown.

[0152] The optimized nodule region refers to the nodule mask region after morphological optimization. The nodule mask region can be mapped to the frequency domain, so as to obtain a corresponding heat map, as shown. Figure 5

[0153] The relevant geometric features can be trend analysis on the nodule area, the true diameter of the nodule, etc., as shown. Figure 6

[0154] Those skilled in the art will realize that the embodiments described herein are for the purpose of illustrating the principles of the present application and should not be construed as limiting the scope of the present application. Those skilled in the art can make various other specific modifications and combinations according to the technical inspiration disclosed in the present application without departing from the spirit of the present application, and these modifications and combinations are still within the scope of protection of the present application.​​

Claims

1. A flexible array sensor based intraoperative lung nodule localization system, comprising: The system comprises: a flexible array sensor arranged at a part of a subject to be detected for collecting a pressure signal to obtain a stress distribution; a nodule region detection unit configured to determine a nodule region by cluster analysis based on the stress distribution; a morphology optimization unit configured to perform morphology optimization on the determined nodule region; a feature extraction unit configured to extract relevant geometric features based on the optimized nodule region; and a visualization unit configured to visually display the stress distribution, the optimized nodule region and the relevant geometric features.

2. The flexible array-based sensor-based intraoperative lung nodule localization system of claim 1, wherein, The flexible array sensor is composed of a plurality of pressure sensors arranged in an array on a flexible material.

3. The flexible array-based sensor-based intraoperative lung nodule localization system of claim 1, wherein, Based on the collected pressure signals, pressure concentration zones are identified, and the pressure increment in these zones is used as the basis for determination. The two-dimensional distribution is used as the stress distribution.

4. The flexible array-based sensor-based intraoperative lung nodule localization system of claim 3, wherein, The nodule region detection unit uses one of K-means algorithm, Gaussian mixture model and DBSCAN algorithm for cluster analysis.

5. The flexible array-based sensor-based intraoperative lung nodule localization system of claim 4, wherein, When the Gaussian mixture model is used for cluster analysis, it is expressed as: ； wherein, represents a probability density at a data point ; represents a mixing weight of the kth Gaussian distribution; represents a Gaussian distribution with mean and covariance .

6. The flexible array-based sensor-based intraoperative lung nodule localization system of any of claims 1 to 5, wherein, The original pressure signals of the pressure sensor nodes near the boundary of the nodule region are converted to the frequency domain, and the boundary of the nodule region is corrected based on frequency domain analysis.

7. The flexible array-based sensor-based intraoperative lung nodule localization system of claim 6, wherein, Based on the nodule boundary determined by the initial nodule mask obtained by the cluster analysis, the original stress signals collected by the pressure sensor nodes in the annular region with a width of 3-5 nodes inside and outside the boundary contour are extracted; the original stress signal of each pressure sensor node is represented as ; The correction of the boundary of the nodule region based on frequency domain analysis includes the following steps: (1) The original stress signal is subjected to fast Fourier transform to obtain the signal in frequency domain : ； wherein denotes the frequency; denotes the pressure amplitude at the nth sampling point; N denotes the total number of sampling points; (2) extracting a dominant frequency of each pressure sensor node ; ； (3) Global statistics are performed on all pressure sensor nodes to calculate the global dominant frequency mean : ； wherein X and Y represent the number of rows and the number of columns in the array sensor, respectively; (4) Calculate the global frequency standard deviation : ； (5) Set the boundary judgment frequency domain anomaly threshold value: ； (6) For each pressure sensor node in the boundary region, if its dominant frequency is greater than 0.5 Hz, then mark this pressure sensor node as an outlier and adjust the nodule mask boundary; otherwise, the boundary is not adjusted.

8. The flexible array-based sensor-based intraoperative lung nodule localization system of claim 1, wherein, The morphology optimization unit includes: a closing operation module configured to perform closing operation on the determined nodule region to fill small holes inside the nodule region; and an opening operation module configured to eliminate isolated noise points outside the nodule region.

9. The flexible array-based sensor-based intraoperative lung nodule localization system of claim 1, wherein, The feature extraction unit first filters out effective nodule regions based on a set area threshold value, and then extracts geometric features of the effective nodule regions. The geometric features include at least one of area, perimeter, equivalent diameter and circularity.

10. The flexible array-based sensor-based intraoperative lung nodule localization system of claim 9, wherein, The system further comprises a physical deformation correction unit configured to correct the equivalent diameter.

11. The flexible array-based sensor-based intraoperative lung nodule localization system of claim 10, wherein, The physical deformation correction unit corrects the equivalent diameter by a correction function according to the nodule depth value, which is expressed as: ； wherein denotes a correction function; the correction function is determined by a polynomial regression or by a look-up table.

12. The intraoperative positioning system for lung nodules based on a flexible array sensor according to claim 11, wherein the polynomial regression is a quadratic polynomial regression; and the correction function determined by the quadratic polynomial regression is expressed as: ； wherein, represents the corrected true diameter; represents the uncorrected equivalent diameter; represents the nodule depth value; represents the regression fit coefficient for i-th power of i, j-th power of d; the correction function determined by the lookup table is expressed as: ； wherein, represents the corrected true diameter; l , , represents the currently calculated equivalent diameter, represents the given equivalent diameter minimum value, represents the given diameter step size, represents the current given nodule depth, represents the given nodule depth minimum value, represents the given depth step size, represents the floor function; (h, l) represents the grid point position determined based on the current equivalent diameter; represents the true diameter experimental value at grid point (l, h) determined by a lookup table, which is represented as , represents the grid size of the lookup table, represents the lth measured diameter, represents the hth depth; is a relative coordinate parameter determined by grid point (l, h), .

Images