Method and system for post-nanosecond pulse ablation imaging evaluation

Abstract

This invention relates to the field of image assessment technology and discloses a method and system for image assessment after nanosecond pulse ablation. The method involves: acquiring images of the ablation area after nanosecond pulse ablation and dividing it into a deterministic irreversible region, an indeterminate border region, a reversible edema region, and a normal region; constructing an original pairwise set within the deterministic irreversible region and a weak pairwise set between the deterministic irreversible region and the indeterminate border region; calculating the attribute label of each voxel in the indeterminate border region and combining it with the deterministic irreversible region to output an irreversible electroporation region. This invention can accurately identify voxels in the indeterminate border region that truly constitute irreversible damage, and combine them with the deterministic irreversible region to output a complete irreversible electroporation region, providing a reliable quantitative assessment indicator of ablation integrity for clinical practice.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image evaluation, in particular to a nanosecond pulse ablation postoperative image evaluation method and system. BACKGROUND

[0002] As a new means of tumor treatment, the postoperative efficacy evaluation of nanosecond pulse ablation technology depends on the accurate identification of irreversible electroporation regions. Delayed gadolinium-enhanced MRI has become an imaging method for acute phase evaluation because it can reflect the degree of cell membrane rupture. The existing technology uses fixed threshold segmentation or single signal intensity judgment, ignoring the gradual change characteristics of nanosecond pulse electroporation damage from the central complete necrosis zone to the peripheral reversible damage zone, resulting in low edge region recognition accuracy. At the same time, the existing technology fails to utilize the spatial correlation information between the deterministic irreversible zone and the uncertain edge zone, and cannot effectively distinguish between true irreversible damage and temporary inflammatory edema. SUMMARY

[0003] The main purpose of the present application is to provide a nanosecond pulse ablation postoperative image evaluation method and system, which can accurately identify the voxels in the uncertain edge zone that truly belong to irreversible damage, and combine the deterministic irreversible zone to output the complete irreversible electroporation region, providing a reliable ablation integrity quantitative evaluation index for clinical use.

[0004] To achieve the above-mentioned purpose, the present application provides a nanosecond pulse ablation postoperative image evaluation method, comprising the following steps:

[0005] Collecting an ablation region image after nanosecond pulse ablation;

[0006] Dividing the ablation region image into a deterministic irreversible zone, an uncertain edge zone, a reversible edema zone and a normal zone;

[0007] Extracting voxel pairs that meet the spatial proximity and signal homogeneity constraints in the deterministic irreversible zone to construct an original positive pair set, and extracting voxel pairs that meet the cross-region similarity constraint between the deterministic irreversible zone and the uncertain edge zone to construct a weak positive pair set;

[0008] Calculating the strong similarity cumulative value of each voxel in the uncertain edge zone in the original positive pair set and the weak similarity cumulative value in the weak positive pair set;

[0009] Calculating the attribute label of each voxel in the uncertain edge zone based on the strong similarity cumulative value and the weak similarity cumulative value, and outputting the irreversible electroporation region in combination with the deterministic irreversible zone.

[0010] Optionally, in the first implementation manner of the first aspect of the present application, the collecting of the ablation region image after nanosecond pulse ablation comprises:

[0011] acquiring a delayed gadolinium enhanced MRI image in an acute phase after the nanosecond-pulse ablation;

[0012] performing bias field correction on the delayed gadolinium enhanced MRI image to obtain a bias field corrected image;

[0013] performing three-dimensional spatial registration on the preoperative tumor localization image and the bias field corrected image, and demarcating a spherical region with a preset expansion radius of a tip of the ablation needle as a theoretical ablation planning region to obtain an ablation region image.

[0014] Optionally, in a second implementation manner of the first aspect of the present application, the dividing the ablation region image into the certain irreversible region, the uncertain edge region, the reversible edema region and the normal region comprises:

[0015] selecting normal tumor tissue with a preset distance range from the ablation center in the ablation region image as a normal tissue reference region, and selecting a preset proportion of voxels with the highest signal intensity as a necrotic core reference region;

[0016] calculating a first signal intensity mean value and a first signal intensity standard deviation of the normal tissue reference region and a second signal intensity mean value and a second signal intensity standard deviation of the necrotic core reference region, respectively;

[0017] multiplying a difference value between the second signal intensity mean value and the first signal intensity mean value by a first preset coefficient, adding the first preset coefficient multiplied difference value to the first signal intensity mean value, and obtaining a certain irreversible threshold value;

[0018] multiplying the difference value between the second signal intensity mean value and the first signal intensity mean value by a second preset coefficient, adding the second preset coefficient multiplied difference value to the first signal intensity mean value, and obtaining an uncertain transition threshold value;

[0019] multiplying the first signal intensity standard deviation by a preset multiple, adding the preset multiple multiplied first signal intensity standard deviation to the first signal intensity mean value, and obtaining a reversible edema threshold value;

[0020] dividing the ablation region image into the certain irreversible region, the uncertain edge region, the reversible edema region and the normal region according to the certain irreversible threshold value, the uncertain transition threshold value and the reversible edema threshold value.

[0021] Optionally, in a third implementation manner of the first aspect of the present application, the dividing the ablation region image into the certain irreversible region, the uncertain edge region, the reversible edema region and the normal region according to the certain irreversible threshold value, the uncertain transition threshold value and the reversible edema threshold value comprises:

[0022] Based on the first average signal intensity and the second average signal intensity, the signal intensity of each voxel in the ablation region image is normalized to obtain the normalized signal intensity.

[0023] The normalized signal intensity is compared with the deterministic irreversible threshold, the uncertain transition threshold, and the reversible edema threshold, respectively;

[0024] Voxels with normalized signal strength greater than or equal to the deterministic irreversible threshold are marked as deterministic irreversible regions; voxels with normalized signal strength between the uncertain transition threshold and the deterministic irreversible threshold are marked as uncertain edge regions; voxels with normalized signal strength between the reversible edema threshold and the uncertain transition threshold are marked as reversible edema regions; and voxels with normalized signal strength less than the reversible edema threshold are marked as normal regions.

[0025] Optionally, in a fourth implementation of the first aspect of the present invention, the step of extracting voxel pairs satisfying spatial proximity and signal homogeneity constraints within the deterministic irreversible region to construct an original pairwise set, and extracting voxel pairs satisfying cross-regional similarity constraints between the deterministic irreversible region and the uncertain edge region to construct a weak pairwise set, includes:

[0026] Voxel pairs are extracted within the deterministic irreversible region, and voxel pairs with a spatial distance less than a first preset distance threshold and a signal strength difference less than a first preset signal threshold are selected to construct an original pair set.

[0027] Voxels in the deterministic irreversible region are selected as anchor voxels. Paired voxels with spatial distances less than a second preset distance threshold and signal strength differences less than a second preset signal threshold are searched in the uncertain edge region to construct a weak pairing set.

[0028] Optionally, in a fifth implementation of the first aspect of the present invention, selecting voxels in the deterministic irreversible region as anchor voxels, and searching in the uncertain edge region for paired voxels whose spatial distance to the anchor voxel is less than a second preset distance threshold and whose signal strength difference is less than a second preset signal threshold, to construct a weak pairing set, includes:

[0029] Traverse the voxels in the deterministic irreversible region and select them sequentially as anchor voxels to obtain the spatial coordinates and signal strength of each anchor voxel;

[0030] For each anchor voxel in the uncertain edge region, calculate the spatial distance and signal strength difference with each voxel, and select voxels that simultaneously satisfy the conditions of spatial distance less than a second preset distance threshold and signal strength difference less than a second preset signal threshold as paired voxels;

[0031] Each anchor voxel is paired with its corresponding voxel to form a voxel pair, and the normalized signal intensity difference is calculated as the gradient weight to obtain a weak pair set.

[0032] Optionally, in a sixth implementation of the first aspect of the present invention, calculating the strong similarity cumulative value in the original pairwise set and the weak similarity cumulative value in the weak pairwise set for each voxel in the uncertain edge region includes:

[0033] For each target voxel in the uncertain edge region, all voxels that form a voxel pair with the target voxel are retrieved from the original pairing set to form a strong similarity neighborhood voxel set, and paired voxels starting from the anchor voxel corresponding to the target voxel are retrieved from the weak pairing set to form a weak similarity neighborhood voxel set.

[0034] For each voxel in the strong similarity neighborhood voxel set and the target voxel, a strong similarity function value based on signal intensity difference and spatial distance is calculated and accumulated to obtain a strong similarity cumulative value. For each voxel in the weak similarity neighborhood voxel set and the corresponding anchor voxel, a weak similarity function value based on signal intensity difference, spatial distance and gradient weight is calculated and accumulated to obtain a weak similarity cumulative value.

[0035] Optionally, in a seventh implementation of the first aspect of the invention, calculating the strong similarity cumulative value in the original pairwise set and the weak similarity cumulative value in the weak pairwise set for each voxel in the uncertain edge region includes:

[0036] For each target voxel in the uncertain edge region, all voxels that form a voxel pair with the target voxel are retrieved from the original pairing set to form a strong similarity neighborhood voxel set, and paired voxels starting from the anchor voxel corresponding to the target voxel are retrieved from the weak pairing set to form a weak similarity neighborhood voxel set.

[0037] For each voxel in the strong similarity neighborhood voxel set and the target voxel, a strong similarity function value based on signal intensity difference and spatial distance is calculated and accumulated to obtain a strong similarity cumulative value. For each voxel in the weak similarity neighborhood voxel set and the corresponding anchor voxel, a weak similarity function value based on signal intensity difference, spatial distance and gradient weight is calculated and accumulated to obtain a weak similarity cumulative value.

[0038] Optionally, in an eighth implementation of the first aspect of the present invention, the step of calculating a complete necrosis attribute discrimination score for each voxel in the uncertain edge region based on the normalized signal intensity and the adaptive weighting coefficient, and calculating a reversible edema attribute discrimination score, a hemorrhage attribute discrimination score, and a normal tissue attribute discrimination score based on the normalized signal intensity, the adaptive weighting coefficient, and the spatial distance to the boundary of the deterministic irreversible region, respectively, includes:

[0039] For each voxel in the uncertain edge region, calculate the shortest spatial distance to the boundary of the deterministic irreversible region, and calculate a preset percentile value as the strong similarity threshold for the cumulative strong similarity value of all voxels.

[0040] After multiplying the normalized signal intensity by the adaptive weight coefficient, an exponential decay term calculated based on the strong similarity cumulative value and the strong similarity threshold is added to obtain the complete necrosis attribute discrimination score.

[0041] Multiply the inverse value of the normalized signal strength by the inverse value of the adaptive weight coefficient, and then multiply by the first exponential decay weight calculated based on the shortest spatial distance to obtain the reversible edema attribute discrimination score.

[0042] The hemorrhage attribute discrimination score is obtained by multiplying the indication value of the signal intensity being lower than the average signal intensity of the normal tissue reference area by the local signal gradient amplitude, and then multiplying it by the second exponential decay weight calculated based on the shortest spatial distance.

[0043] The normal tissue attribute discrimination score is obtained by multiplying the exponential decay value calculated based on the normalized signal strength with the indication value that the shortest spatial distance is greater than a preset distance threshold.

[0044] This invention also provides a post-nanosecond pulse ablation imaging evaluation system, comprising:

[0045] The acquisition module is used to acquire images of the ablation area after nanosecond pulse ablation.

[0046] The segmentation module is used to divide the ablation area image into a deterministic irreversible region, an uncertain edge region, a reversible edema region, and a normal region.

[0047] The construction module is used to extract voxel pairs that satisfy spatial proximity and signal homogeneity constraints within the deterministic irreversible region to construct an original pair set, and to extract voxel pairs that satisfy cross-regional similarity constraints between the deterministic irreversible region and the uncertain edge region to construct a weak pair set.

[0048] The calculation module is used to calculate the strong similarity cumulative value in the original pairwise set and the weak similarity cumulative value in the weak pairwise set for each voxel in the uncertain edge region;

[0049] The output module is used to calculate the attribute label of each voxel in the uncertain edge region based on the strong similarity cumulative value and the weak similarity cumulative value, and output the irreversible electroporation region in combination with the deterministic irreversible region.

[0050] In summary, this invention overcomes the limitations of fixed threshold methods in adapting to individual differences in contrast agent metabolism by establishing an adaptive three-layer threshold system based on the statistical characteristics of dual reference regions. This achieves precise stratification of deterministic irreversible regions, uncertain edge regions, reversible edema regions, and normal regions. A novel hybrid orthogonal learning strategy is proposed. This strategy captures the homogeneity of completely necrotic tissue by constructing an original orthogonal set within the deterministic irreversible region, and represents the gradual transition characteristics of damage by constructing a weak orthogonal set between the deterministic irreversible region and the uncertain edge region, quantifying the damage gradient information from the core to the edge in nanosecond pulse electroporation. Based on adaptive weight coefficients calculated from strong and weak similarity cumulative values, the contribution of the two types of orthogonals is dynamically balanced according to the topological position of voxels in the hybrid orthogonal learning space. Voxels deep in the necrotic core rely on strong similarity information, while voxels in the uncertain edge region incorporate the gradual information of weak similarity, achieving continuous evaluation from deterministic to uncertain regions. By fusing multi-dimensional features such as normalized signal intensity, spatial gradient, and positive similarity through a multi-attribute discriminant function, and combining spatial consistency constraints to eliminate isolated noise points, the system accurately identifies voxels in uncertain edge regions that truly belong to irreversible damage. This is then combined with deterministic irreversible regions to output a complete irreversible electroporation region, providing reliable quantitative assessment indicators for ablation integrity in clinical practice. Attached Figure Description

[0051] Figure 1 This is a schematic diagram of the steps of a nanosecond pulse ablation postoperative image evaluation method in one embodiment of the present invention;

[0052] Figure 2 This is a structural block diagram of a nanosecond pulse ablation postoperative image evaluation system according to an embodiment of the present invention.

[0053] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0054] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0055] Reference Figure 1 This embodiment provides a method for image evaluation after nanosecond pulse ablation, including the following steps:

[0056] S1, Acquire images of the ablation area after nanosecond pulse ablation;

[0057] The acute phase, from 24 to 48 hours after nanosecond pulse ablation, was selected as the image acquisition window. During this period, the liver tissue has completed the clearance and metabolism of the contrast agent. However, in the irreversible electroporation area, the osmotic pressure difference caused by cell membrane rupture leads to the continuous retention of gadopentetate dimeglumine contrast agent in the intercellular spaces, resulting in a significant high signal characteristic in delayed gadolinium-enhanced MRI. At this time point, after intravenous injection of 0.1 mmol / kg body weight of gadopentetate dimeglumine and a 15-20 minute waiting period for the contrast agent to reach equilibrium, T1-weighted delayed gadolinium-enhanced MRI images were acquired to obtain original postoperative image data with pathologically specific contrast. The N4 bias field correction algorithm was applied to the delayed gadolinium-enhanced MRI images to effectively estimate and compensate for low-frequency artifact signals, making the signal intensity distribution of the images more realistically reflect the differences between tissues, resulting in bias field-corrected images. The preoperative tumor localization image and the delayed enhancement image after bias field correction were rigidly or non-rigidly registered in three-dimensional space. The spatial unification of images at different time points was achieved by image alignment methods based on mutual information or gradient correlation coefficient. Based on the registration results, the theoretical boundary of the spherical ablation region was constructed with the tip of the ablation needle as the center and an extension radius of 2.0 cm to 3.0 cm. The set of voxels covered by the sphere was calibrated in three-dimensional space as the ablation region image used for postoperative analysis.

[0058] S2 divides the ablation area image into a deterministic irreversible area, an indeterminate edge area, a reversible edema area, and a normal area;

[0059] Specifically, in the ablation area images corresponding to the postoperative delayed gadolinium-enhanced MRI, two reference regions were selected based on spatial location and tissue characteristics. One type consisted of normal tumor tissue within 3.0 to 5.0 cm from the ablation center, excluding large blood vessels, preoperative necrotic areas, and tumor margins, forming a normal tissue reference region to provide a baseline signal intensity for undamaged tissue. The other type consisted of the top 20% of voxels with the highest signal intensity within the ablation area, serving as a necrotic core reference region reflecting the high retention of gadolinium contrast agent and the disintegration of the intercellular matrix structure, thus reflecting the characteristic signals of the completely irreversible electroporation zone. The mean and standard deviation of the signal intensity were calculated for both the normal tissue reference region and the necrotic core reference region. The first mean and standard deviation of the signal intensity were used to reflect the distribution characteristics of normal tissue, while the second mean and standard deviation of the signal intensity were used to quantify the enhancement degree of the necrotic core. Based on a preset first coefficient, such as 0.75, the difference between the second mean and the first mean is multiplied by the first coefficient and added to the first mean to obtain the signal intensity threshold for the deterministic irreversible region. This threshold is used to identify completely necrotic voxels whose signals are higher than those of normal tissue and approach the average value of the necrotic core. Using a second preset coefficient, such as 0.45, the above calculation logic is repeated to obtain the lower bound threshold for the uncertain transition region, which is used to delineate the blurred edge zone that slowly transitions from normal to lesion. Simultaneously, the upper bound threshold for the reversible edema region is constructed by multiplying the first signal intensity standard deviation by a preset factor, such as 2, and adding it to the first mean. This threshold is used to identify transient inflammatory areas with mild enhancement and signals still close to those of normal tissue. The signal intensity of all voxels in the ablation region is sequentially compared with the three thresholds and hierarchically classified according to a normalization strategy, thereby assigning each voxel in the image to the deterministic irreversible region, uncertain edge region, reversible edema region, or normal region.

[0060] S3. Extract voxel pairs that satisfy spatial proximity and signal homogeneity constraints within the deterministic irreversible region to construct the original pair set. Extract voxel pairs that satisfy cross-regional similarity constraints between the deterministic irreversible region and the uncertain edge region to construct the weak pair set.

[0061] It should be noted that, within the deterministic irreversible region, a set of voxels meeting the normalized signal intensity threshold is extracted. Voxel pairing is then performed within this set. For each candidate voxel pair, the Euclidean distance between their three-dimensional spatial coordinates is calculated, and the signal intensity difference between each candidate voxel pair in the original image is evaluated. When both constraints are met—the spatial distance being less than a first preset distance threshold (e.g., 5.0 mm) and the signal intensity difference being less than a first preset signal threshold (e.g., 0.15 multiplied by the standard deviation of the necrotic area)—the current voxel pair is determined to be a pair with similar signal patterns and consistent local tissue structure. This pair is then added to the original paired set to capture the highly homogeneous pathological atlas features within the completely necrotic region. Furthermore, the statistical distribution of irreversible damage is learned through a similarity kernel function. To characterize the spatial transition relationship between the necrotic core and the edge region, multiple representative voxels are selected as anchor voxels from the labeled deterministic irreversible region. With each anchor voxel as a reference, a candidate set of voxels that form a pairing relationship with it is searched in the uncertain edge region. The Euclidean distance between the candidate voxels and the anchor voxels is calculated, and the difference in their normalized signal intensities is compared. When the paired voxels satisfy the condition that the spatial distance is less than a second preset distance threshold (e.g., 10.0 mm) and the signal intensity difference is less than a second preset signal threshold (e.g., 0.30 multiplied by the standard deviation of the necrotic region), the current voxel pair is recorded as a weakly paired voxel pair under the cross-regional gradual relationship, and it is added to the weakly paired set. This reflects the multidimensional characteristics of signal gradual transition continuity, spatial adjacency, and tissue transition from the necrotic region to the edge region.

[0062] S4, calculate the strong similarity cumulative value in the original pairwise set and the weak similarity cumulative value in the weak pairwise set for each voxel in the uncertain edge region;

[0063] Specifically, subsets of voxels that have a direct pairing relationship with the current voxel to be evaluated are extracted from both the original pairing set and the weak pairing set, forming two neighborhood voxel sets. For any target voxel in the uncertain edge region, all other voxels that together form a direct pairing relationship with the current voxel are searched in the original pairing set; voxels that satisfy the pairing relationship are included to form a strong similarity neighborhood voxel set. In the weak pairing set, the target voxel is used as the anchor voxel, and all paired voxels that satisfy the anchor condition are retrieved. These paired voxels are collected to form a weak similarity neighborhood voxel set. Together, they construct the associated context reflecting the voxel's spatial organization structure. Similarity calculations are performed on each voxel in both the strong similarity neighborhood voxel set and the weak similarity neighborhood voxel set. Within the strong similarity neighborhood, the weighted relationship between signal intensity difference and spatial distance is considered, assigning higher strong similarity scores to spatially adjacent voxels with consistent signals. All scores are summed to obtain the cumulative strong similarity value of the target voxel. A larger cumulative strong similarity value indicates that the target voxel is more likely to be in a completely necrotic region with strong structural homogeneity. Simultaneously, in the weak similarity neighborhood, the degree of signal difference, spatial distance, and gradient difference on the normalized signal spectrum are evaluated between each paired voxel and the target anchor voxel. Additional gradient change weights are introduced to weight the scoring results to highlight the continuous change trend of the damage transition zone. All scoring results are summed to obtain the cumulative weak similarity value, which varies between 0 and 1. A higher value indicates that the voxel is located in an active edge region between irreversible and reversible damage, exhibiting obvious gradual transition properties. In this embodiment, the reference scale parameter for strong similarity scoring is set to 0.10×σ_c and 3.0 mm, and the reference scale parameter for weak similarity scoring is 0.20×σ_c and 8.0 mm, where σ_c is the standard deviation of signal intensity in the necrotic core region.

[0064] S5 calculates the attribute label of each voxel in the uncertain edge region based on the cumulative value of strong similarity and the cumulative value of weak similarity, and outputs the irreversible electroporation region in combination with the deterministic irreversible region.

[0065] Specifically, the adaptive weight coefficient is calculated by proportionally calculating the cumulative strong similarity value of each voxel in the face-to-face learning space to the sum of the cumulative similarity values ​​of the two classes. This dynamically reflects the positional characteristics of the voxel in the mixed face-to-face learning space. When the voxel is in the core region of irreversible necrosis, the cumulative strong similarity value is greater than the cumulative weak similarity value, and the weight approaches 1.0. However, the weight of voxels in the uncertain region at the edge of the damage is reduced to below 0.5 because the cumulative weak similarity value is dominant, thus achieving an adaptive description of the gradual damage spectrum. For each voxel in the uncertain edge region, the complete necrosis attribute discrimination score is calculated based on its normalized signal intensity and adaptive weight coefficient. The reversible edema, hemorrhage, and normal tissue attribute discrimination scores are calculated by combining the signal intensity distribution, spatial gradient, and spatial distance from the voxel to the boundary of the irreversible region. The complete necrosis attribute mainly reflects high signal intensity, high adaptive weight, and high cumulative similarity features. The reversible edema attribute corresponds to voxels with low signal intensity, high weak similarity weight, and close to the edge of the irreversible region. The hemorrhage attribute is manifested as boundary voxels with signal intensity lower than normal tissue but large local gradient amplitude. The normal tissue attribute is mainly voxels far from the ablation center and with signal close to the baseline. The four attribute discrimination scores of each voxel are probabilistically normalized to obtain the probability distribution of each voxel under the four attributes. Spatial consistency constraints are applied to voxels with a probability of complete necrosis higher than 0.60. The proportion of irreversibly necrotic voxels in their 3×3×3 neighborhood is calculated. When the proportion is greater than 0.40, the current voxel is considered to belong to a spatially continuous necrosis cluster; otherwise, its attribute probability is reduced to suppress isolated noise points. The label with the highest attribute probability is selected from the voxels that satisfy the spatial consistency condition as the final attribute label. Voxels in uncertain edge regions labeled as completely necrotic are merged with deterministic irreversible regions to obtain the irreversible electroporation region.

[0066] In one example, images of the ablation area are acquired after nanosecond pulse ablation, including:

[0067] Delayed gadolinium-enhanced MRI images were acquired during the acute phase following nanosecond pulse ablation.

[0068] Offset field correction was performed on the delayed gadolinium-enhanced MRI images to obtain the offset field-corrected images;

[0069] The preoperative tumor localization image and the bias-field calibrated image are registered in three-dimensional space, and a spherical region with a preset extension radius from the tip of the ablation needle is calibrated as the theoretical ablation plan area to obtain the ablation area image.

[0070] In this example, based on the mechanism of cell membrane structure destruction caused by nanosecond pulse electroporation, the acute phase time window between 24 and 48 hours post-procedure was selected as the MRI image acquisition point. At this time, due to the increased permeability of the intercellular matrix in the irreversibly damaged area and its inability to repair, gadopentetate dimeglumine contrast agent remains in this area, while normal tissue has completed the metabolism and clearance of the contrast agent. Therefore, the contrast of the delayed enhancement signal during this period can reflect the true boundary of the necrotic tissue. Within the acute phase time window, after intravenous injection of 0.1 mmol / kg body weight of gadopentetate dimeglumine contrast agent, the patient waits for 15 to 20 minutes to complete the exchange and distribution balance of the contrast agent between blood and tissue. T1-weighted delayed gadolinium-enhanced MRI images are then acquired, and the scanning parameters are set as follows: slice thickness 2.0 mm to 3.0 mm, slice interval 0 mm, repetition time 3.5 ms to 4.5 ms, echo time 1.5 ms to 2.0 ms, and flip angle 10 degrees to 15 degrees, so that the image has sufficient spatial resolution to capture the subtle gradient features of the ablation boundary area. After image acquisition, the original images were corrected using an N4 bias field to eliminate background signal drift caused by magnetic field inhomogeneity. The standard deviation of the signal intensity of the corrected images decreased to less than 30% of that of the original images. Preoperative tumor localization images and postoperative delayed-enhanced images corrected with the bias field were registered in three-dimensional voxel space. Spatial alignment of preoperative and postoperative scan data was achieved through rigid registration combined with mutual information metric criteria. The ablation path and needle tip position were retrospectively reconstructed and located. Based on this, a spherical voxel mask with a spatial radius of 2.0 cm to 3.0 cm was constructed with the ablation needle tip as the center to serve as the boundary of the theoretical ablation planning area. Corresponding voxels from the MRI images were extracted within this spherical region to form the postoperative ablation area image included in the evaluation scope.

[0071] In one example, the ablation region image is divided into a deterministic irreversible region, an indeterminate border region, a reversible edema region, and a normal region, including:

[0072] In the ablation area image, normal tumor tissue within a preset distance range from the ablation center is selected as the normal tissue reference area, and the preset scale voxel with the highest signal intensity is selected as the necrotic core reference area.

[0073] Calculate the mean and standard deviation of the first signal intensity in the normal tissue reference region and the mean and standard deviation of the second signal intensity in the necrotic core reference region, respectively.

[0074] The difference between the average value of the second signal strength and the average value of the first signal strength is multiplied by a first preset coefficient and then added to the average value of the first signal strength to obtain the deterministic irreversible threshold.

[0075] The difference between the average second signal strength and the average first signal strength is multiplied by a second preset coefficient and then added to the average first signal strength to obtain the uncertain transition threshold.

[0076] The reversible edema threshold is obtained by multiplying the standard deviation of the first signal intensity by a preset factor and adding it to the mean of the first signal intensity.

[0077] Based on the deterministic irreversible threshold, the uncertain transition threshold, and the reversible edema threshold, the ablation area image is divided into a deterministic irreversible region, an uncertain edge region, a reversible edema region, and a normal region.

[0078] In this example, in delayed gadolinium-enhanced MRI images, a reference region is selected within a three-dimensional spherical region formed by the spatial coordinates of the ablation needle tip and the theoretical extension radius. The extraction criteria for the normal tissue reference region are as follows: from a ring-shaped spatial band 3.0 cm to 5.0 cm from the ablation center, excluding vascular structures (screening out signal channels with a diameter greater than 2 mm based on maximum density projection), excluding low-signal areas with necrosis in preoperative images (parts below 50% of the tumor signal mean), and excluding mixed voxels that may exist within 2 mm of the tumor edge due to blurred boundaries, thus constructing a representative normal tumor. A reference area for tissue was established. Simultaneously, within the ablation zone image, based on the high retention characteristics of the contrast agent in the irreversible electroporation necrotic area, the top 20% of voxels with the highest signal intensity were selected to construct the necrotic core reference area. This area corresponds to a tissue state where the cell membrane is completely ruptured and the contrast agent has fully penetrated, exhibiting highly consistent and high-intensity enhanced signal performance. The mean and standard deviation of signal intensity were calculated for all voxels in both the normal tissue reference area and the necrotic core reference area, respectively. The first mean and standard deviation of signal intensity were obtained, reflecting the signal distribution characteristics of unaffected tissue. The second mean and standard deviation of signal intensity were used to quantify the signal enhancement capability of the necrotic core area. By setting a first preset coefficient of 0.75, the difference between the mean of the second signal intensity and the mean of the first signal intensity is multiplied by the first preset coefficient and then added to the mean of the first signal intensity to generate a deterministic irreversible region threshold, which is used to identify voxel regions with signal intensities significantly higher than normal tissue and highly close to the necrotic core area. By setting a second preset coefficient of 0.45, the difference between the mean of the second signal intensity and the mean of the first signal intensity is multiplied by the second preset coefficient and then added to the mean of the first signal intensity to construct an uncertain transition region threshold, representing the lower bound threshold of the signal response from the normal to the lesion stage. The standard deviation of the first signal intensity is multiplied by 2 and then added to the mean of the first signal intensity, which is set as the reversible edema region threshold. The voxel signal response at the edge of the fluctuation range is captured using the 95% confidence interval of a normal distribution. The signal intensity values ​​of all voxels in the ablation area image are classified and judged. When the signal intensity is greater than or equal to the irreversible threshold, it is classified as a deterministic irreversible region. Voxels located between the uncertain threshold and the irreversible threshold are classified as uncertain edge regions. Voxels with signal intensity between the edema threshold and the uncertain threshold are classified as reversible edema regions. And those with signal intensity below the edema threshold are classified as normal regions.

[0079] In one example, the ablation region image is divided into a deterministic irreversible region, an uncertain border region, a reversible edema region, and a normal region based on a deterministic irreversible threshold, an uncertain transition threshold, and a reversible edema threshold, including:

[0080] Based on the average of the first signal intensity and the average of the second signal intensity, the signal intensity of each voxel in the ablation region image is normalized to obtain the normalized signal intensity.

[0081] The normalized signal intensity was compared with the deterministic irreversible threshold, the uncertain transition threshold, and the reversible edema threshold, respectively.

[0082] Voxels with normalized signal intensity greater than or equal to the deterministic irreversible threshold are marked as deterministic irreversible regions; voxels with normalized signal intensity between the uncertain transition threshold and the deterministic irreversible threshold are marked as uncertain edge regions; voxels with normalized signal intensity between the reversible edema threshold and the uncertain transition threshold are marked as reversible edema regions; and voxels with normalized signal intensity less than the reversible edema threshold are marked as normal regions.

[0083] In this example, the mean signal intensity of the first signal intensity and the mean signal intensity of the second signal intensity, which correspond to the center values ​​of the signal intensity of the normal tissue reference area and the necrotic core reference area, respectively, are used as the upper and lower boundary values ​​of the normalization mapping. The original signal intensity of all voxels in the ablation area image is normalized, which transforms the problem of inconsistent signal numerical scales between different individuals due to differences in scanning parameters and contrast agent metabolism rates into a unified standard range of 0 to 1. 0 represents the baseline signal of unaffected normal tissue, while 1 represents the extreme value of the necrotic core signal where the contrast agent is highly concentrated. The normalized signal intensity value of each voxel is compared with the deterministic irreversible threshold, the indeterminate transition threshold, and the reversible edema threshold to define the signal range corresponding to different pathological stages. When the normalized signal intensity of a voxel is greater than or equal to the deterministic irreversible threshold (e.g., 0.75), it indicates that the signal of the current voxel has reached or exceeded the lower limit of the intensity characteristic of the necrotic core, and it is marked as a deterministic irreversible region, corresponding to a completely necrotic cell population. If the voxel signal intensity is between the indeterminate transition threshold (e.g., 0.45) and the deterministic irreversible threshold, the current voxel is in a gradual transition zone, containing partially necrotic and partially reversibly damaged cells, and is marked as an indeterminate edge region. When the signal intensity is between the reversible edema threshold (e.g., generated by 2 standard deviations) and the indeterminate transition threshold, it indicates that the current voxel reflects mild enhancement and its spatial location is close to the ablation boundary, which is considered to be caused by edema and is marked as a reversible damage area. If the normalized signal intensity of a voxel is lower than the reversible edema threshold, its appearance is similar to the normal tissue baseline, and it is classified as a normal tissue area.

[0084] In one example, voxel pairs satisfying spatial proximity and signal homogeneity constraints are extracted within the deterministic irreversible region to construct the original pairwise set, and voxel pairs satisfying cross-regional similarity constraints are extracted between the deterministic irreversible region and the uncertain edge region to construct the weak pairwise set, including:

[0085] Voxel pairs are extracted within the deterministic irreversible region, and voxel pairs with a spatial distance less than a first preset distance threshold and a signal intensity difference less than a first preset signal threshold are selected to construct the original pair set.

[0086] Voxels in the deterministic irreversible region are selected as anchor voxels. Paired voxels in the uncertain edge region are searched for whose spatial distance from the anchor voxel is less than a second preset distance threshold and whose signal strength difference is less than a second preset signal threshold, and a weak pairing set is constructed.

[0087] In this example, a pairwise pairing operation is performed within the deterministic irreversible region. The Euclidean distance between the three-dimensional spatial coordinates of all possible voxel pairs is calculated, and the signal intensity difference between the two voxels is evaluated. A selection process is then performed based on a first preset distance threshold (e.g., 5.0 mm) and a first preset signal difference threshold (e.g., 0.15 multiplied by the standard deviation of the signal intensity in the necrotic core region). Voxel pairs satisfying the dual constraints are retained to form the original pairwise set. The voxel pairs in the original pairwise set originate from the highly damaged region, exhibiting spatial proximity in structure and high homogeneity in signal expression. This accurately reflects the spatial structural characteristics of cell death patterns caused by concentrated energy deposition in the irreversible damage region. Meanwhile, to address the issues of signal continuity and gradual pathological changes between irreversible necrosis zones and uncertain border zones, a weak pairing set is constructed to reflect regional transition characteristics. Voxels are selected from the deterministic irreversible zone as anchor voxels, and a pairing search operation is performed in the uncertain border zone. For each anchor voxel, voxels with a spatial distance less than a second preset distance threshold (e.g., 10.0 mm) and a signal intensity difference less than a second preset signal difference threshold (e.g., 0.30 multiplied by the standard deviation of signal intensity in the necrotic core zone) are selected as pairing objects, forming cross-regional voxel pairs added to the weak pairing set.

[0088] In one example, voxels in the deterministic irreversible region are selected as anchor voxels. In the uncertain edge region, paired voxels whose spatial distance to the anchor voxel is less than a second preset distance threshold and whose signal strength difference is less than a second preset signal threshold are searched, constructing a weak pairing set, including:

[0089] Traverse the voxels in the deterministic irreversible region and select them sequentially as anchor voxels to obtain the spatial coordinates and signal strength of each anchor voxel;

[0090] For each anchor voxel, calculate the spatial distance and signal intensity difference with each voxel in the uncertain edge region, and select voxels that simultaneously satisfy the conditions of spatial distance less than a second preset distance threshold and signal intensity difference less than a second preset signal threshold as paired voxels;

[0091] Each anchor voxel is paired with its corresponding voxel to form a voxel pair, and the normalized signal intensity difference is calculated as the gradient weight to obtain a weak pair set.

[0092] In this example, a point-by-point traversal operation is performed on all voxels in the marked deterministic irreversible region, and each voxel is selected as an anchor voxel in turn. The spatial position of the anchor voxel in the 3D image coordinate system and its corresponding original or normalized signal intensity value are extracted to form the core reference features for comparison and matching. For each anchor voxel, a full-range search is performed in the set of voxels contained in the uncertain edge region. The spatial Euclidean distance and signal intensity difference between the anchor voxel and the candidate voxels are calculated one by one. The spatial distance reflects the physical proximity between tissue structures, while the signal difference measures the degree of proximity of the enhanced response between pathological states. After the calculation is completed, voxels that meet both conditions are selected to form the anchor pair. The first condition is that the spatial distance between the two voxels is less than a set second preset distance threshold, such as 10.0 mm. The second condition is that the signal intensity difference between the two is less than a second preset signal threshold, such as not exceeding 0.30 times the standard deviation of the necrotic core area. The selected paired voxels are both close to the region where the anchor voxel is located and maintain a certain degree of continuity and similarity in signal performance. After screening, each pair of anchor voxels is combined with its corresponding paired voxel to form a voxel pair. The difference in normalized signal intensity between the two pairs is calculated to reflect the relative positional relationship of the two voxels on the tissue damage gradient spectrum. This value is used as the gradient weight to obtain a similarity expression that better reflects the actual tissue transformation trend. The set of all voxel pairs consisting of anchor voxels and paired voxels that meet the above conditions constitutes the weak positive pair set.

[0093] In one example, for each voxel in the uncertain edge region, the strong similarity cumulative value in the original pairwise set and the weak similarity cumulative value in the weak pairwise set are calculated, including:

[0094] For each target voxel in the uncertain edge region, all voxels that form a voxel pair with the target voxel are retrieved from the original pair set to form a strong similarity neighborhood voxel set, and paired voxels starting from the anchor voxel corresponding to the target voxel are retrieved from the weak pair set to form a weak similarity neighborhood voxel set.

[0095] For each voxel in the strong similarity neighborhood voxel set and the target voxel, a strong similarity function value based on the signal intensity difference and spatial distance is calculated and accumulated to obtain a strong similarity cumulative value. For each voxel in the weak similarity neighborhood voxel set and the corresponding anchor voxel, a weak similarity function value based on the signal intensity difference, spatial distance and gradient weight is calculated and accumulated to obtain a weak similarity cumulative value.

[0096] In this example, a strong similarity neighborhood voxel set and a weak similarity neighborhood voxel set are constructed in two pairs of pairs, with the target voxel as the core. The process involves traversing all voxel pairs in the original pairs and searching for voxel pairs with the current target voxel as a member. Regardless of whether the voxel is the first or second member of the pair, as long as it appears in the original pairs, the corresponding voxel is considered an adjacency point forming a pair and is included in the strong similarity neighborhood voxel set. Simultaneously, with the current target voxel as the anchor voxel, all voxel pairs established with it as the starting point are searched in the weak pairs, and the corresponding paired voxels are extracted to form the weak similarity neighborhood voxel set, thus constructing a spatial connection graph describing the target voxel in the cross-regional gradient structure. For each voxel in the strong similarity neighborhood, the signal intensity difference and 3D spatial distance between it and the target voxel are calculated sequentially. Responsibility is evaluated based on a pre-defined similarity function structure. The scoring model assigns different weights to the signal difference and spatial proximity to reflect the dual constraints of structural consistency and signal homogeneity. All strong similarity scores are accumulated to form the cumulative strong similarity value of the target voxel. A larger cumulative strong similarity value indicates that the target voxel is closer to the highly consistent characteristics of the irreversible damage zone in both spatial and signal dimensions. For each paired voxel in the weak similarity neighborhood, its corresponding anchor voxel is used as a reference. The signal intensity difference, spatial distance, and difference between the normalized signal intensity and the anchor voxel are calculated. The normalized intensity difference is introduced as a gradient weight factor into the weak similarity scoring function. During the similarity calculation, the gradient weight reflects the relative position of the voxel pair on the damage spectrum distribution, thus reflecting the spatial variation trend of the gradient direction and damage depth. All weak similarity scores are summarized and accumulated to obtain the cumulative weak similarity value of the target voxel.

[0097] In one example, the attribute label for each voxel in the uncertain edge region is calculated based on the cumulative values ​​of strong and weak similarities, and the irreversible electroporation region is output by combining the deterministic irreversible region, including:

[0098] The adaptive weighting coefficient is calculated by the ratio of the cumulative strong similarity value to the sum of the cumulative strong similarity value and the cumulative weak similarity value.

[0099] For each voxel in the uncertain edge region, the discrimination score for complete necrosis is calculated based on the normalized signal intensity and the adaptive weighting coefficient. The discrimination scores for reversible edema, hemorrhage, and normal tissue are calculated based on the normalized signal intensity, the adaptive weighting coefficient, and the spatial distance to the boundary of the deterministic irreversible region, respectively.

[0100] Based on the discrimination scores of complete necrosis, reversible edema, hemorrhage, and normal tissue, spatial consistency constraints are applied to voxels with a probability of complete necrosis greater than a preset probability threshold to determine whether the proportion of irreversible voxels in their neighborhood meets the preset proportion threshold.

[0101] The attribute corresponding to the highest attribute probability value among the voxels that meet the preset percentage threshold is selected as the attribute label of the voxel. The voxels with the attribute label of complete necrosis in the uncertain edge region are merged with the deterministic irreversible region to obtain the irreversible electroporation region.

[0102] In this example, the cumulative strong similarity and cumulative weak similarity values ​​of each voxel in the uncertain edge region are fused. The adaptive weight coefficient of the current voxel is calculated by dividing the cumulative strong similarity value of the voxel by the sum of its strong and weak similarity values. The weight coefficient is used to dynamically adjust the degree of dependence on different types of positive information in attribute discrimination. If a voxel is mainly composed of strong similarity neighborhoods, the weight coefficient is close to 1.0, indicating that its attribute discrimination should rely more on the structural consistency information of the deterministic irreversible region. If a voxel is mainly composed of weak similarity neighborhoods, the weight coefficient decreases, indicating that the voxel is more likely to be in the lesion gradient zone. Based on the adaptive weighting coefficients and the normalized signal intensity value of each voxel, the discrimination score for complete necrosis is calculated. In the calculation, the simultaneous occurrence of high signal values ​​and high weight values ​​will lead to a higher necrosis tendency score, thus reflecting the typical irreversible characteristics of voxels in terms of structure and signal. At the same time, spatial distance information to the boundary of the deterministic irreversible region is introduced as a spatial location reference, and it participates in the calculation of the discrimination scores of the other three types of attributes together with the normalized signal intensity and the adaptive weighting coefficients. The discrimination score for reversible edema is high in the region with low normalized signal value, weak similarity, and close to the irreversible boundary. The discrimination score for hemorrhage is high when the signal is slightly lower than the mean of normal tissue, the local gradient changes strongly, and it is close to the edge of the injury. The discrimination score for normal tissue is high when the signal is close to the normal baseline, the weight is low, and it is far from the ablation center. Normalization is performed on the discrimination scores of the four attribute categories to construct a probability distribution, which is then used to assess the probability that each voxel belongs to a different tissue attribute. Based on this, a spatial consistency constraint is applied to all voxels whose probability of complete necrosis attribute exceeds a preset threshold (e.g., 0.60). That is, a 3×3×3 voxel neighborhood is constructed with the voxel as the center, and the number of voxels marked as irreversible necrosis attribute in the neighborhood is counted and the proportion is calculated. When the proportion exceeds the preset threshold (e.g., 0.40), the voxel is determined to belong to a spatially continuous necrotic structural unit. If the proportion is insufficient, it is considered an isolated misclassification, and its necrosis probability weight is reduced. The category with the highest attribute probability is selected from the voxels that meet the spatial consistency constraint as the final label, and the uncertain edge region voxels classified as completely necrosis are merged with the original set of deterministic irreversible region voxels to obtain the irreversible electroporation region.

[0103] In one example, for each voxel in the uncertain boundary region, a complete necrosis attribute discrimination score is calculated based on normalized signal intensity and adaptive weighting coefficients. Reversible edema attribute discrimination scores, hemorrhage attribute discrimination scores, and normal tissue attribute discrimination scores are calculated based on normalized signal intensity, adaptive weighting coefficients, and spatial distance to the boundary of the deterministic irreversible region, respectively.

[0104] For each voxel in the uncertain edge region, calculate the shortest spatial distance to the boundary of the deterministic irreversible region, and calculate the preset percentile value of the strong similarity cumulative value of all voxels as the strong similarity threshold.

[0105] After multiplying the normalized signal strength by the adaptive weight coefficient, and adding the exponential decay term calculated based on the strong similarity accumulation value and the strong similarity threshold, the complete necrosis attribute discrimination score is obtained.

[0106] Multiply the inverse value of the normalized signal strength by the inverse value of the adaptive weight coefficient, and then multiply by the first exponential decay weight calculated based on the shortest spatial distance to obtain the reversible edema attribute discrimination score.

[0107] The hemorrhage attribute discrimination score is obtained by multiplying the indication value of the signal intensity being lower than the mean signal intensity of the normal tissue reference area by the local signal gradient amplitude, and then multiplying it by the second exponential decay weight calculated based on the shortest spatial distance.

[0108] The normal tissue attribute discrimination score is obtained by multiplying the exponential decay value calculated based on the normalized signal strength with the indication value that the shortest spatial distance is greater than a preset distance threshold.

[0109] In this example, the shortest spatial distance to the boundary of the deterministic irreversible region is calculated for each voxel to be discriminated. The shortest spatial distance, as a geometric feature reflecting the physical proximity of the voxel to the necrotic core, can effectively encode spatial topological information. At the same time, the cumulative strong similarity values ​​of all voxels in the entire ablation region image are statistically sorted, and the 75th percentile value is taken as the strong similarity threshold to define the enhanced reference standard when judging the complete necrosis attribute. For each voxel, four attribute discrimination scores are calculated. For the complete necrosis attribute, the product of normalized signal intensity and adaptive weighting coefficient is used as the basic expression metric. An exponential decay term based on the current voxel's accumulated strong similarity value and a preset strong similarity threshold is introduced to represent the voxel's response intensity in the direction of structural homogeneity. The two parts are summed to obtain the complete necrosis attribute discrimination score. For the reversible edema attribute, the inverse value of the normalized signal intensity (1 minus the normalized signal intensity) is multiplied by the inverse value of the adaptive weighting coefficient (1 minus the weighting coefficient) to form a basic weak similarity expression component. This is then multiplied by an exponential decay function constructed based on the voxel's distance to the irreversible boundary, resulting in higher edema scores for voxels closer to the boundary, thus completing the edema attribute discrimination score calculation. For the outflow... For hemorrhage attribute determination, the system determines whether the current voxel signal intensity is lower than the mean of the normal tissue reference area. If so, the indicator value is 1; otherwise, it is 0. The indicator value is multiplied by the local signal gradient amplitude corresponding to the voxel and multiplied by a second exponential decay weight based on distance to enhance the influence of locally high gradient voxels near the spatial center in hemorrhage discrimination, thus obtaining a hemorrhage attribute score. For normal tissue attribute discrimination, the exponential decay function value of the normalized signal intensity is used as the basic expression, emphasizing the importance of the signal intensity being close to the normal baseline. At the same time, it determines whether the shortest distance of the current voxel is greater than a preset spatial distance threshold. If it is greater, the indicator value is 1; otherwise, it is 0. The above two items are multiplied to suppress misidentification cases where the spatial location is close to the ablation core area but the signal is close to normal, thus obtaining a normal tissue attribute discrimination score.

[0110] Reference Figure 2 This embodiment provides a post-nanosecond pulse ablation image evaluation system, including:

[0111] Acquisition module 1 is used to acquire images of the ablation area after nanosecond pulse ablation.

[0112] The segmentation module 2 is used to divide the ablation area image into a deterministic irreversible region, an indeterminate edge region, a reversible edema region, and a normal region.

[0113] Module 3 is used to extract voxel pairs that satisfy spatial proximity and signal homogeneity constraints within the deterministic irreversible region to construct the original pair set, and to extract voxel pairs that satisfy cross-regional similarity constraints between the deterministic irreversible region and the uncertain edge region to construct the weak pair set.

[0114] Calculation module 4 is used to calculate the strong similarity cumulative value in the original pairwise set and the weak similarity cumulative value in the weak pairwise set for each voxel in the uncertain edge region;

[0115] Output module 5 is used to calculate the attribute label of each voxel in the uncertain edge region based on the cumulative value of strong similarity and the cumulative value of weak similarity, and output the irreversible electroporation region in combination with the deterministic irreversible region.

[0116] In this embodiment, the specific implementation of each unit in the above system embodiment is described in the above method embodiment, and will not be repeated here.

[0117] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, system, article, or method that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, system, article, or method. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, system, article, or method that includes that element.

[0118] The above description is only a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A method for postoperative image evaluation of nanosecond pulse ablation, characterized in that, include: Acquire images of the ablation area after nanosecond pulse ablation; The ablation area image is divided into a deterministic irreversible region, an indeterminate edge region, a reversible edema region, and a normal region. Within the deterministic irreversible region, voxel pairs satisfying spatial proximity and signal homogeneity constraints are extracted to construct an original pair set. Between the deterministic irreversible region and the uncertain edge region, voxel pairs satisfying cross-regional similarity constraints are extracted to construct a weak pair set. For each voxel in the uncertain edge region, calculate the strong similarity cumulative value in the original pairwise set and the weak similarity cumulative value in the weak pairwise set; The attribute label of each voxel in the uncertain edge region is calculated based on the strong similarity cumulative value and the weak similarity cumulative value, and the irreversible electroporation region is output in combination with the deterministic irreversible region.

2. The method for postoperative image evaluation of nanosecond pulse ablation according to claim 1, characterized in that, The acquisition of images of the ablation area after nanosecond pulse ablation includes: Delayed gadolinium-enhanced MRI images were acquired during the acute phase following nanosecond pulse ablation. Bias field correction is performed on the delayed gadolinium-enhanced MRI image to obtain a bias field-corrected image; The preoperative tumor localization image and the bias-field calibrated image are registered in three-dimensional space, and a spherical region with a preset extension radius from the tip of the ablation needle is calibrated as the theoretical ablation plan region to obtain the ablation region image.

3. The method for postoperative image evaluation of nanosecond pulse ablation according to claim 1, characterized in that, The process of dividing the ablation area image into a deterministic irreversible region, an indeterminate edge region, a reversible edema region, and a normal region includes: In the ablation area image, normal tumor tissue within a preset distance range from the ablation center is selected as the normal tissue reference area, and a preset scale voxel with the highest signal intensity is selected as the necrotic core reference area. Calculate the mean and standard deviation of the first signal intensity in the normal tissue reference region and the mean and standard deviation of the second signal intensity in the necrotic core reference region, respectively. The difference between the second signal strength mean and the first signal strength mean is multiplied by a first preset coefficient and then added to the first signal strength mean to obtain the deterministic irreversible threshold. The difference between the second average signal strength and the first average signal strength is multiplied by a second preset coefficient and then added to the first average signal strength to obtain an uncertain transition threshold, wherein the first preset coefficient is greater than the second preset coefficient. The reversible edema threshold is obtained by multiplying the standard deviation of the first signal intensity by a preset factor and adding it to the mean of the first signal intensity. Based on the deterministic irreversible threshold, the uncertain transition threshold, and the reversible edema threshold, the ablation region image is divided into a deterministic irreversible region, an uncertain edge region, a reversible edema region, and a normal region.

4. The method for postoperative image evaluation of nanosecond pulse ablation according to claim 3, characterized in that, The step of dividing the ablation region image into a deterministic irreversible region, an uncertain edge region, a reversible edema region, and a normal region based on the deterministic irreversible threshold, the uncertain transition threshold, and the reversible edema threshold includes: Based on the first average signal intensity and the second average signal intensity, the signal intensity of each voxel in the ablation region image is normalized to obtain the normalized signal intensity. The normalized signal intensity is compared with the deterministic irreversible threshold, the uncertain transition threshold, and the reversible edema threshold, respectively; Voxels with normalized signal strength greater than or equal to the deterministic irreversible threshold are marked as deterministic irreversible regions; voxels with normalized signal strength between the uncertain transition threshold and the deterministic irreversible threshold are marked as uncertain edge regions; voxels with normalized signal strength between the reversible edema threshold and the uncertain transition threshold are marked as reversible edema regions; and voxels with normalized signal strength less than the reversible edema threshold are marked as normal regions.

5. The method for post-nanosecond pulse ablation image evaluation according to claim 4, characterized in that, The step of extracting voxel pairs satisfying spatial proximity and signal homogeneity constraints within the deterministic irreversible region to construct an original pairwise set, and extracting voxel pairs satisfying cross-regional similarity constraints between the deterministic irreversible region and the uncertain edge region to construct a weak pairwise set, includes: Voxel pairs are extracted within the deterministic irreversible region, and voxel pairs with a spatial distance less than a first preset distance threshold and a signal intensity difference less than a first preset signal threshold are selected to construct an original pair set. Voxels in the deterministic irreversible region are selected as anchor voxels. Paired voxels with spatial distances less than a second preset distance threshold and signal strength differences less than a second preset signal threshold are searched in the uncertain edge region to construct a weak pairing set.

6. The method for postoperative image evaluation of nanosecond pulse ablation according to claim 5, characterized in that, The process involves selecting voxels in the deterministic irreversible region as anchor voxels, searching in the uncertain edge region for paired voxels whose spatial distance to the anchor voxel is less than a second preset distance threshold and whose signal strength difference is less than a second preset signal threshold, and constructing a weak pairing set, including: Traverse the voxels in the deterministic irreversible region and select them sequentially as anchor voxels to obtain the spatial coordinates and signal strength of each anchor voxel; For each anchor voxel in the uncertain edge region, calculate the spatial distance and signal strength difference with each voxel, and select voxels that simultaneously satisfy the conditions of spatial distance less than a second preset distance threshold and signal strength difference less than a second preset signal threshold as paired voxels; Each anchor voxel is paired with its corresponding voxel to form a voxel pair, and the normalized signal intensity difference is calculated as the gradient weight to obtain a weak pair set.

7. The method for post-nanosecond pulse ablation image evaluation according to claim 6, characterized in that, The calculation of the strong similarity cumulative value in the original pairwise set and the weak similarity cumulative value in the weak pairwise set for each voxel in the uncertain edge region includes: For each target voxel in the uncertain edge region, all voxels that form a voxel pair with the target voxel are retrieved from the original pairing set to form a strong similarity neighborhood voxel set, and paired voxels starting from the anchor voxel corresponding to the target voxel are retrieved from the weak pairing set to form a weak similarity neighborhood voxel set. For each voxel in the strong similarity neighborhood voxel set and the target voxel, a strong similarity function value based on signal intensity difference and spatial distance is calculated and accumulated to obtain a strong similarity cumulative value. For each voxel in the weak similarity neighborhood voxel set and the corresponding anchor voxel, a weak similarity function value based on signal intensity difference, spatial distance and gradient weight is calculated and accumulated to obtain a weak similarity cumulative value.

8. The method for postoperative image evaluation of nanosecond pulse ablation according to claim 7, characterized in that, The step of calculating the attribute label of each voxel in the uncertain edge region based on the accumulated strong similarity value and the accumulated weak similarity value, and outputting the irreversible electroporation region in combination with the deterministic irreversible region, includes: The adaptive weighting coefficient is calculated by the ratio of the cumulative strong similarity value to the sum of the cumulative strong similarity value and the cumulative weak similarity value. For each voxel in the uncertain edge region, a complete necrosis attribute discrimination score is calculated based on the normalized signal intensity and the adaptive weighting coefficient. A reversible edema attribute discrimination score, a hemorrhage attribute discrimination score, and a normal tissue attribute discrimination score are calculated based on the normalized signal intensity, the adaptive weighting coefficient, and the spatial distance to the boundary of the deterministic irreversible region, respectively. Based on the complete necrosis attribute discrimination score, the reversible edema attribute discrimination score, the hemorrhage attribute discrimination score, and the normal tissue attribute discrimination score, a spatial consistency constraint is applied to voxels with a complete necrosis attribute probability greater than a preset probability threshold to determine whether the proportion of irreversible voxels in their neighborhood meets a preset proportion threshold. The attribute corresponding to the highest attribute probability value among the voxels that meet the preset percentage threshold is selected as the attribute label of the voxel. The voxels with the attribute label of complete necrosis in the uncertain edge region are merged with the deterministic irreversible region to obtain the irreversible electroporation region.

9. The method for postoperative image evaluation of nanosecond pulse ablation according to claim 8, characterized in that, The step of calculating a complete necrosis attribute discrimination score for each voxel in the uncertain edge region based on the normalized signal intensity and the adaptive weighting coefficient, and calculating a reversible edema attribute discrimination score, a hemorrhage attribute discrimination score, and a normal tissue attribute discrimination score based on the normalized signal intensity, the adaptive weighting coefficient, and the spatial distance to the boundary of the deterministic irreversible region, respectively, includes: For each voxel in the uncertain edge region, calculate the shortest spatial distance to the boundary of the deterministic irreversible region, and calculate a preset percentile value as the strong similarity threshold for the cumulative strong similarity value of all voxels. After multiplying the normalized signal intensity by the adaptive weight coefficient, an exponential decay term calculated based on the strong similarity cumulative value and the strong similarity threshold is added to obtain the complete necrosis attribute discrimination score. Multiply the inverse value of the normalized signal strength by the inverse value of the adaptive weight coefficient, and then multiply by the first exponential decay weight calculated based on the shortest spatial distance to obtain the reversible edema attribute discrimination score. The hemorrhage attribute discrimination score is obtained by multiplying the indication value of the signal intensity being lower than the average signal intensity of the normal tissue reference area by the local signal gradient amplitude, and then multiplying it by the second exponential decay weight calculated based on the shortest spatial distance. The normal tissue attribute discrimination score is obtained by multiplying the exponential decay value calculated based on the normalized signal strength with the indication value that the shortest spatial distance is greater than a preset distance threshold.

10. A postoperative image evaluation system for nanosecond pulse ablation, characterized in that, The steps for implementing the post-nanosecond pulse ablation image evaluation method according to any one of claims 1 to 9 include: The acquisition module is used to acquire images of the ablation area after nanosecond pulse ablation. The segmentation module is used to divide the ablation area image into a deterministic irreversible region, an uncertain edge region, a reversible edema region, and a normal region. The construction module is used to extract voxel pairs that satisfy spatial proximity and signal homogeneity constraints within the deterministic irreversible region to construct an original pair set, and to extract voxel pairs that satisfy cross-regional similarity constraints between the deterministic irreversible region and the uncertain edge region to construct a weak pair set. The calculation module is used to calculate the strong similarity cumulative value in the original pairwise set and the weak similarity cumulative value in the weak pairwise set for each voxel in the uncertain edge region; The output module is used to calculate the attribute label of each voxel in the uncertain edge region based on the strong similarity cumulative value and the weak similarity cumulative value, and output the irreversible electroporation region in combination with the deterministic irreversible region.

Images