Tumor area rapid identification method and system based on cortical electroencephalogram signals

Abstract

The invention belongs to the technical field of medical imaging, and discloses a tumor area rapid identification method and system based on cortical electroencephalogram signals, and the method comprises the steps: carrying out the preprocessing of collected multi-channel ECoG signals through an interpolation filling strategy; extracting a target frequency band ECoG signal of each channel; extracting a feature sequence of each channel in a given time period as a channel feature; performing clustering analysis on each channel feature to obtain a plurality of categories; and reordering according to the feature level of each category of region, and completing tumor region boundary identification by taking the category of which the feature level meets the requirement as a tumor region. According to the invention, rapid detection and effective identification of tumor boundaries can be realized.

Description

Technical Field

[0001] The present invention belongs to the field of medical imaging technology, and relates to tumor boundary detection, and in particular to a method and system for quickly identifying tumor regions based on cortical electroencephalogram signals. Background Art

[0002] In the field of neurosurgery, accurately defining the boundaries of brain tumors is an extremely complex and crucial task. This process provides the surgical team with critical navigation information, helping doctors to accurately plan the range of resection, maximize the retention of functional compensatory areas and normal brain tissue that are critical to the patient's quality of life after surgery, and is directly related to the improvement of postoperative rehabilitation effects. For a long time, the process of defining tumor margins has mainly relied on preoperative MRI images to preliminarily calibrate the central coordinates of the tumor and its approximate shape. During surgery, surgeons need to make subjective judgments based on intuitive tissue characteristics, such as texture and color, to confirm whether they are close to the tumor boundary. However, given the unique infiltrative growth pattern of gliomas, the boundary demarcation of the infiltrative area is particularly vague and highly subjective, and there is an urgent need for more efficient and objective strategies to accurately diagnose the actual margins of the tumor.

[0003] Existing imaging technologies based on structural measurements, including MRI and intraoperative ultrasound, can reveal regional changes in tumor cell density to some extent and provide a certain reference for delineating the scope of surgery, but they face dual challenges when dealing with infiltrated areas: first, these areas often present blurred and gradual edges, making it difficult to avoid subjectivity in anatomical structure identification based on visual judgment; second, such technologies cannot effectively evaluate the real-time functional state of the cerebral cortex. In fact, even under tumor infiltration, some areas can still maintain important functions or develop new key functional areas due to functional compensation mechanisms. Preserving these areas is crucial to protecting patients' postoperative cognitive and physiological functions. Especially for those gliomas located in known functional areas, their boundary identification requires extremely high accuracy. Traditional methods are difficult to meet this high standard, increasing the risk of damaging key functional areas.

[0004] In this context, electrocorticography (ECoG), as a technology for monitoring the activity of cerebral cortical neurons, provides a new perspective for defining the boundaries of tumors and functional areas, thanks to its high sensitivity to brain functional states, especially its successful application in localizing epileptic foci and decoding language and motor areas. ECoG can not only capture electrophysiological signals with high temporal resolution in real time, but the advancement of high-density ECoG technology has greatly enhanced the spatial resolution, making it an ideal tool for intraoperative tumor localization. Studies have shown that the electrical activity patterns in the invasive areas of gliomas are significantly different from those in normal functional areas and non-invasive areas, which has opened up a new way to identify tumor boundaries using ECoG signal differences.

[0005] Therefore, how to combine cortical electroencephalogram to achieve rapid detection of tumor boundaries is a technical problem that needs to be solved. Summary of the invention

[0006] The purpose of the present invention is to address the deficiencies of the prior art and provide a method for rapid identification of tumor areas based on cortical ECoG signals. By deeply analyzing high-density ECoG signals, the differential features between tumors and normal brain areas are extracted, and a machine learning algorithm is used to accurately define the boundaries of brain tumors and their surrounding functional areas in the image.

[0007] In order to achieve the above object, the present invention adopts the following technical solutions.

[0008] The present invention provides a method for rapid tumor region identification based on cortical electroencephalogram signals, which comprises the following steps:

[0009] S1 preprocesses the collected multi-channel ECoG signals through interpolation and filling strategies;

[0010] S2 extracts the target frequency band ECoG signal of each channel;

[0011] S3 extracts the feature sequence of each channel in a given time period as the channel feature;

[0012] S4 performs cluster analysis on the characteristics of each channel to obtain several categories;

[0013] S5 reorders the regions according to their feature levels, and uses the regions whose feature levels meet the requirements as tumor regions to complete tumor region boundary identification.

[0014] In the above step S1, the purpose is to interpolate the bad conductors in all channels, so as to improve the ECoG signal.

[0015] The interpolation filling strategy in the above step S1 includes the following sub-steps:

[0016] (1) Traverse each channel, identify the abnormal channel, and add it to the bad channel list;

[0017] (2) Traverse the bad guide list and search for bad guides whose first-order neighbor channels are all in the bad guide list. All the bad guides found will skip this round of interpolation.

[0018] (3) The remaining channels in the bad guide list are used as the bad guides to be filled, and the first-order neighbor channels that are not in the bad guide list are used to interpolate and fill the bad guides to be filled;

[0019] Then return to step (1) and repeat steps (1)-(3) until all bad guides are interpolated.

[0020] In the above step (1), the bad conductor can be identified based on the characteristics of each channel. The characteristics are selected from average energy, differential entropy, kurtosis or standard deviation. These characteristics can be calculated by conventional methods based on ECoG signals.

[0021] The characteristic value of each channel can be compared with the set threshold to determine whether each channel is a bad conductor.

[0022] In one implementation, bad conductors may be identified based on the average energy of each channel; the average energy of a channel is calculated as follows:

[0023]

[0024] Among them, Pow (i) Represents the energy value in the i-th unit time; N is the number of sampling points of the data; Indicates the amplitude of the nth sampling point in the ith unit time. At this time, when the average energy of the channel is more than 5 times higher than the energy average of the global non-bad conduction channel (sensitivity can be adjusted), the channel is an abnormal channel and is marked as a bad conduction. If it is not higher than 5 times the energy average of the global non-bad conduction channel, the channel is considered to be a normal channel.

[0025] In the above steps (2)-(3), the ECoG signal of the first-order neighbor channel of the bad conductor is used to fill it. In the present invention, the first-order neighbor channel refers to the channel directly adjacent to the channel.

[0026] Specifically, it is necessary to first search for bad conductors that can be interpolated and filled through step (2), and then determine whether all the first-order neighbor channels around the bad conductor are bad conductors; if all are bad conductors, the bad conductor needs to skip this round of interpolation; if not all the first-order neighbor channels around the bad conductor are bad conductors, it indicates that the bad conductor can be interpolated and filled. Based on this, the bad conductor list is traversed to search for bad conductor channels whose first-order neighbor channels are all in the bad conductor list. All the bad conductor channels that are searched out skip this round of interpolation, and the remaining bad conductors except these channels in the bad conductor list are used as bad conductors to be filled.

[0027] In step (3), the bad conductor to be filled is filled by interpolation using the ECoG signals of the normal channels that are not in the bad conductor list. According to the previous explanation, the normal channel is not in the bad conductor list. If the first-order neighbor channels of the bad conductor to be filled are not all bad conductors, but some are bad conductors, the ECoG signals of the first-order neighbor channels that are not in the bad conductor list are used to interpolate and fill the above-mentioned bad conductor to be filled. If there is no bad conductor in the first-order neighbor channel, the ECoG signals of all the first-order neighbor channels are used to interpolate and fill the above-mentioned bad conductor to be filled.

[0028] Repeat the above steps (1)-(3) for several rounds until all bad guides are interpolated and filled.

[0029] In the above step S2, the target frequency band ECoG signal of each channel can be extracted by a bandpass filter. The bandpass filter can be a Butterworth filter, a finite impulse response filter, etc. The frequency band of the target frequency band ECoG signal extracted in this step is 4-140 Hz.

[0030] In the above step S3, channel features may be extracted based on average energy, differential entropy, kurtosis or standard deviation of the channel.

[0031] Taking the average energy of the channel as an example, the average energy data within a unit time (which can be 0.1 seconds, 1 second, 10 seconds, etc.) can effectively distinguish tumors from normal brain tissue. The average energy calculation formula within a unit time is (1). This formula can be used to calculate an energy value for each unit time. The energy sequence of a given time period is used as the channel feature for tumor identification.

[0032] In the above step S4, for each channel feature, an unsupervised machine learning algorithm is used to cluster the channel features to obtain several categories. The unsupervised machine learning algorithm is a K-Means clustering algorithm, a hierarchical clustering algorithm, a DBSCAN algorithm or a convolutional neural network.

[0033] In the above step S5, since the label of the unsupervised algorithm is a random value, here we first count the average value of the channel features in each category area as the regional feature level based on the clustering results, and further reorder the various categories (clusters) obtained by clustering according to the regional feature level. Finally, the category with the feature level that meets the requirements (the lowest or highest feature value, related to the specific feature selected) is selected as the tumor area to complete the identification of the tumor area boundary.

[0034] The area outside the tumor area in the cluster classification result is the non-tumor area; therefore, the tumor area and the non-tumor area can also be binarized to further highlight the tumor area.

[0035] The present invention also provides a rapid tumor region identification system based on cortical EEG signals, which comprises:

[0036] A data preprocessing module is used to preprocess the collected multi-channel ECoG signals through an interpolation and filling strategy;

[0037] A target frequency band extraction module is used to extract the target frequency band ECoG signal of each channel;

[0038] A channel feature extraction module is used to extract the feature sequence of each channel within a given time as the channel feature;

[0039] Clustering module, used to perform cluster analysis on the features of each channel to obtain several categories;

[0040] The reordering module is used to reorder the feature levels of each category region, select the category with the feature level that meets the requirements as the tumor region, and complete the identification of the tumor boundary. The requirements that the feature level needs to meet here depend on the specific feature category selected.

[0041] The above-mentioned data preprocessing module, target frequency band extraction module, channel feature extraction module, clustering module, and reordering module perform operations according to the steps given in the previous tumor region rapid identification method.

[0042] Compared with the prior art, the method and system for rapid tumor area identification based on cortical EEG provided by the present invention have the following beneficial effects:

[0043] 1) Based on high-density cortical electroencephalogram, the present invention first processes abnormal data through automatic bad guide interpolation method to improve data, then extracts channel features, and then clusters and reorders according to channel features, which can realize accurate identification of tumor boundaries;

[0044] 2) The automatic guide interpolation method provided by the present invention can effectively identify and interpolate abnormal data of each channel, thereby improving the recognition accuracy of tumor boundaries;

[0045] 3) In the process of identifying tumor boundary distinction, the present invention first divides the region by clustering, and then finds the region whose characteristic value meets the requirements (the lowest or highest characteristic value, depending on the selected feature category) by re-sorting, and then determines the tumor region boundary, which can improve the accuracy of tumor boundary division. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 A schematic diagram of a method for rapid tumor region identification based on cortical electroencephalogram provided in an embodiment of the present invention;

[0047] Figure 2 A block diagram of the principle of rapid identification of tumor areas based on channel energy provided by an embodiment of the present invention;

[0048] Figure 3 An example of bad guide interpolation provided by an embodiment of the present invention;

[0049] Figure 4 The present invention provides a tumor region boundary generation process. DETAILED DESCRIPTION

[0050] The present invention is described in detail below by embodiments and application examples. It is necessary to point out that the present embodiments are only used to further illustrate the present invention and cannot be construed as limiting the scope of the present invention. Those skilled in the art in this field can make some non-essential improvements and adjustments according to the above-mentioned content of the present invention. If specific conditions are not specified in the embodiments, they are carried out according to normal conditions or conditions recommended by the manufacturer. If the manufacturer is not specified for the reagents or instruments used, they are all conventional products that can be purchased commercially.

[0051] Example 1

[0052] like Figure 1 and Figure 2 As shown, the method for rapid identification of tumor areas in medical images based on cortical electroencephalogram (EEG) provided in this embodiment includes the following steps:

[0053] S1 preprocesses the collected multi-channel ECoG signals through interpolation and filling strategies.

[0054] Several exploration electrodes are placed directly on the surface of the patient's cerebral cortex according to the set position, and each electrode corresponds to a channel. The embodiment collects brain electrical activity to obtain a multi-channel ECoG signal. There may be some abnormal data in the obtained multi-channel data. Therefore, it is necessary to identify and eliminate the abnormal data, and interpolate and fill in the eliminated position.

[0055] The interpolation filling strategy in this step includes the following sub-steps:

[0056] (1) Traverse each channel, identify the abnormal channels, and add them to the bad channel list.

[0057] In this embodiment, bad conductors are identified based on the average energy of each channel. The average energy of the channel is calculated by formula (1) to obtain the average energy of the channel in any unit time. When the average energy of the channel is more than 5 times higher than the energy average of the global non-bad conductor channels (sensitivity can be adjusted), the channel is an abnormal channel and is marked as a bad conductor. If it is not higher than 5 times the energy average of the global non-bad conductor channels, the channel is considered to be a normal channel.

[0058] (2) Traverse the bad guide list and search for bad guides whose first-order neighbor channels are all in the bad guide list. All the bad guides found will skip this round of interpolation.

[0059] It is necessary to first search for bad guides that can be interpolated and filled through this step, so as to determine whether all the first-order neighbor channels around the bad guides are bad guides; if all are bad guides, the bad guide needs to skip this round of interpolation; if not all the first-order neighbor channels around the bad guide are bad guides (including the case where there are no bad guides), it indicates that the bad guide can be interpolated and filled. Based on this, the bad guide list is traversed to search for bad guide channels whose first-order neighbor channels are all in the bad guide list. All the bad guide channels searched out skip this round of interpolation, and the remaining bad guides except these channels in the bad guide list are used as substitute bad guides.

[0060] (3) The remaining channels in the bad guide list are used as the bad guides to be filled, and the first-order neighbor channels that are not in the bad guide list are used to interpolate and fill the bad guides to be filled.

[0061] In this step, the bad conductor to be filled is filled by interpolation using the ECoG signal of the normal channel that is not in the bad conductor list. According to the previous explanation, the normal channel is not in the bad conductor list. If the first-order neighbor channels of the bad conductor to be filled are not all bad conductors, but there are some bad conductors, the ECoG signal of the first-order neighbor channel with normal leads that is not in the bad conductor list is used to interpolate and fill the above bad conductor. If there is no bad conductor in the first-order neighbor channel, the energy mean of all first-order neighbor channels is used to interpolate and fill the above bad conductor.

[0062] Then return to step (1) and repeat steps (1)-(3) until all bad guides are interpolated.

[0063] Figure 3 Three interpolation filling situations are given.

[0064] like Figure 3 As shown in (a), according to step (2) in the interpolation filling strategy, the first-order neighbor channels around the bad conductor are all normal leads. At this time, step (3) directly uses the ECoG signals of the four first-order neighbor channels to fill the bad conductor.

[0065] like Figure 3 As shown in (b), according to step (2) in the interpolation filling strategy, the two bad conductors are adjacent to each other; according to step (3), the two bad conductors each skip the other bad conductor and use the ECoG signals of the three first-order neighbor channels adjacent to themselves to fill the bad conductor.

[0066] like Figure 3As shown in (c), according to step (2) in the interpolation filling strategy, there are 9 bad conductors here, and the first-order neighbors around the middle bad conductor are also bad conductors. Therefore, the early interpolation will not be considered for this bad conductor; according to step (3), the 8 bad conductors around it are first interpolated and filled using the ECoG signals of the corresponding first-order neighbor channels (i.e., normal channels) that are not in the bad conductor list; then return to step (1), re-determine the bad conductor, and use these interpolated results to interpolate and fill the middle bad conductor.

[0067] S2 extracts the target frequency band ECoG signal of each channel.

[0068] The target frequency band ECoG signal of each channel can be extracted by a bandpass filter. The bandpass filter used in this embodiment is a finite impulse response filter.

[0069] The frequency band of the target frequency band ECoG signal extracted in this step is 4-140 Hz.

[0070] S3 extracts the feature sequence of each channel in a given time period as the channel feature.

[0071] In this embodiment, the unit time is 1 second, and the average energy data within 1 second can effectively distinguish tumors from normal brain tissues. The calculation formula for the average energy per unit time is as follows:

[0072]

[0073] Among them, Pow (i) Represents the energy value in the i-th unit time; N1 is the number of sampling points for 1 second of data; Represents the amplitude of the nth sampling point in the ith unit time. The above formula can be used to calculate an energy value for each unit time data. The 15-second energy sequence is used as the channel feature for tumor recognition.

[0074] Figure 4 (a) is the energy sequence of each channel obtained.

[0075] S4 performs cluster analysis on the features of each channel to obtain several categories.

[0076] For each channel feature, the K-Means clustering algorithm is used to cluster the channel features to obtain several categories. The K-Means clustering algorithm is implemented using conventional methods that have been disclosed in machine learning, see Orhan U, Hekim M, Ozer M. EEG signals classification using the K-means clustering and a multilayerperceptron neural network model [J]. Expert Systems with Applications, 2011, 38 (10): 13475-13481.

[0077] Figure 4 (b) Figure 4 The clustering results of the energy sequence obtained in (a) after K-Means clustering.

[0078] S5 reorders the regions according to their feature levels, and uses the regions whose feature levels meet the requirements as tumor regions to complete tumor region boundary identification.

[0079] Since the clustering labels of K-Means are random values, here we first count the average energy in each category area based on the clustering results (obtained by the average energy of all channels in the corresponding category, which is the characteristic level of the category area), and further reorder the categories within the entire electrode range according to the regional average energy. For this embodiment, the tumor area has the lowest energy, so the lowest energy is used as the requirement for satisfying the tumor area; finally, the category with the lowest energy is selected as the tumor area to complete the identification of the tumor area boundary. The area outside the tumor area in the cluster classification result is the non-tumor area. Based on this, binarization processing of the tumor area and the non-tumor area can also be achieved.

[0080] Figure 4 (c) Figure 4 In (b), each category area is reordered according to the regional energy within the entire electrode range (here it is represented by grayscale images with different grayscale levels). Figure 4 (d) shows the binarization results of the tumor area and non-tumor area. The area with the lowest energy is marked as the tumor area, with the value "0" (black area); the rest of the area is marked as the non-tumor area, with the value "1" (white area).

[0081] Example 2

[0082] This embodiment provides a system for rapid tumor region identification based on cortical EEG signals, which includes:

[0083] A data preprocessing module is used to preprocess the collected multi-channel ECoG signals through an interpolation and filling strategy;

[0084] A target frequency band extraction module is used to extract the target frequency band ECoG signal of each channel;

[0085] A channel feature extraction module is used to extract the feature sequence of each channel within a given time as the channel feature;

[0086] Clustering module, used to perform cluster analysis on the features of each channel to obtain several categories;

[0087] The reordering module is used to reorder the regions according to their feature levels, select the regions whose feature levels meet the requirements as the tumor regions, and complete the identification of the tumor boundaries.

[0088] The data preprocessing module, target frequency band extraction module, channel feature extraction module, clustering module, and reordering module perform operations according to the steps given in the method for rapid identification of tumor regions provided in Example 1. Among them, the target frequency band extraction module can be implemented using a finite impulse response filter, and the clustering module can be implemented using a K-Means clustering algorithm.

[0089] Those skilled in the art will appreciate that the embodiments herein are intended to help readers understand the principles of the present invention, and should be understood that the protection scope of the present invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific variations and combinations that do not deviate from the essence of the present invention based on the technical revelations disclosed by the present invention, and these variations and combinations are still within the protection scope of the present invention.

Claims

1. A method for rapid tumor region identification based on cortical EEG signals, characterized in that: The following steps are involved: S1 preprocesses the collected multi-channel ECoG signals through interpolation and filling strategies; S2 extracts the target frequency band ECoG signal of each channel; S3 extracts the feature sequence of each channel in a given time period as the channel feature; S4 performs cluster analysis on the characteristics of each channel to obtain several categories; S5 reorders the regions according to their feature levels, and uses the regions whose feature levels meet the requirements as tumor regions to complete tumor region boundary identification.

2. The method for rapid tumor area identification based on cortical EEG signals according to claim 1, characterized in that: The interpolation filling strategy in step S1 includes the following sub-steps: (1) Traverse each channel, identify the abnormal channel, and add it to the bad channel list; (2) Traverse the bad guide list and search for bad guides whose first-order neighbor channels are all in the bad guide list. All the bad guides found will skip this round of interpolation. (3) The remaining channels in the bad guide list are used as the bad guides to be filled, and the first-order neighbor channels that are not in the bad guide list are used to interpolate and fill the bad guides to be filled; Then return to step (1) and repeat steps (1)-(3) until all bad guides are interpolated.

3. The method for rapid tumor area identification based on cortical EEG signals according to claim 2, characterized in that: In step (1), bad conductors are identified based on the characteristics of each channel; the characteristics are selected from average energy, differential entropy, kurtosis or standard deviation.

4. The method for rapid tumor area identification based on cortical EEG signals according to claim 3, characterized in that: Bad conductors are identified based on the average energy of each channel; when the average energy of a channel is more than 5 times higher than the global average energy of non-bad conductor channels, the channel is an abnormal channel and is marked as a bad conductor.

5. The method for rapid tumor region identification based on cortical EEG signals according to claim 1, characterized in that: In step S2, the target frequency band ECoG signal of each channel is extracted through a bandpass filter.

6. The method for rapid tumor region identification based on cortical EEG signals according to claim 5, characterized in that: The bandpass filter is a Butterworth filter or a finite impulse response filter.

7. The method for rapid tumor region identification based on cortical EEG signals according to claim 1, characterized in that: In step S3, channel features are extracted based on the average energy, differential entropy, kurtosis or standard deviation of the channel.

8. The method for rapid tumor region identification based on cortical EEG signals according to claim 1, characterized in that: In step S4, an unsupervised machine learning algorithm is used to cluster the channel features to obtain several categories.

9. The method for rapid tumor region identification based on cortical EEG signals according to claim 8, characterized in that: The unsupervised machine learning algorithm is K-Means algorithm, hierarchical clustering algorithm, DBSCAN algorithm or convolutional neural network.

10. The method for rapid tumor region identification based on cortical EEG signals according to any one of claims 1 to 9, characterized in that: In step S5, based on the clustering result, the average value of the channel features in each category area is counted as the regional feature level, and the categories obtained by clustering are further reordered according to the regional feature level.

11. The method for rapid tumor region identification based on cortical EEG signals according to claim 10, characterized in that: In step S5, the area outside the tumor area in the cluster classification result is the non-tumor area; the tumor area and the non-tumor area are binarized.

12. A rapid tumor region identification system based on cortical EEG signals, characterized in that: include: A data preprocessing module is used to preprocess the collected multi-channel ECoG signals through an interpolation and filling strategy; A target frequency band extraction module is used to extract the target frequency band ECoG signal of each channel; A channel feature extraction module is used to extract the feature sequence of each channel within a given time as the channel feature; Clustering module, used to perform cluster analysis on the features of each channel to obtain several categories; The reordering module is used to reorder the regions according to their feature levels, select the regions whose feature levels meet the requirements as the tumor regions, and complete the identification of the tumor boundaries.

Images