Search

CN-121980223-A - Intelligent intraoperative incisal margin evaluation system for breast cancer breast-preserving operation

CN121980223ACN 121980223 ACN121980223 ACN 121980223ACN-121980223-A

Abstract

The invention belongs to the technical field of computer-aided medical diagnosis, and discloses an intelligent intraoperative cutting edge evaluation system for breast cancer breast-preserving operation; the sequential optical coherence tomography images and the autofluorescence spectra of the incised edge tissues before and after the preset pressure stimulus are obtained, and the mechanical optical coupling response characteristics and the metabolic dynamic response characteristics are extracted, so that the multidimensional analysis of the incised edge tissues is realized. The invention adopts core technologies such as feature fusion, space consistency analysis, region iteration optimization, risk classification and the like, establishes a complete evaluation flow from micro-feature to macroscopic judgment, breakthrough introduces a topological distance concept, and combines the lesion risk degree and the space position to give objective and quantitative evaluation of incisional edge safety. The invention realizes high-precision full-incisional edge real-time evaluation, can effectively identify tiny lesions, provides immediate and objective decision support in operation for surgeons, obviously reduces the re-operation rate, and combines the treatment effect and attractive reservation.

Inventors

  • ZHOU SICHENG

Assignees

  • 北京大学第一医院(北京大学第一临床医学院)

Dates

Publication Date
20260505
Application Date
20260129

Claims (10)

  1. 1. An intelligent intraoperative margin assessment system for breast cancer breast conservation surgery, comprising: the data acquisition module is used for acquiring a time sequence optical coherence tomography image sequence and a synchronous autofluorescence spectrum sequence of the incisional tissue before and after the preset pressure stimulation; the mechanical optical characteristic extraction module is used for acquiring the mechanical optical coupling response characteristic of each voxel according to the scattering intensity time-varying curve of each voxel in the time sequence optical coherence tomography image sequence and the displacement variation track of the interlayer boundary; The metabolic characteristic extraction module is used for obtaining metabolic dynamic response characteristics of each sampling site according to the time-varying characteristics of the oxidation-reduction ratio and the spectral kurtosis coefficient changes of the NADH sensitive wave band and the FAD sensitive wave band in the synchronous autofluorescence spectrum sequence; The lesion recognition module is used for acquiring a tissue attribute confidence vector of each position according to an original combined feature vector obtained by splicing the mechanical optical coupling response feature and the metabolic dynamic response feature of each position and combining the feature consistency degree of the space adjacent positions; The lesion classification module is used for acquiring the region reliability coefficient of each candidate lesion region according to the internal feature homogeneity of each candidate lesion region, the feature jump amplitude of the region boundary and the region compactness; The edge cutting evaluation module is used for acquiring a lesion grade label of the corresponding stable lesion region according to the layering matching result of the comprehensive feature vector of each stable lesion region and the preset lesion mode library, and evaluating the safety state of the edge cutting region based on the lesion grade label and the topological distance from the stable lesion region to the edge cutting outer boundary.
  2. 2. The system of claim 1, wherein the obtaining the mechanocoupling response characteristic for each voxel comprises: Carrying out time sequence arrangement on the scattering intensity of each voxel in the time sequence optical coherence tomography image sequence to obtain a scattering intensity time-varying curve; Identifying an intensity peak value corresponding to the pressure stimulation moment in the scattering intensity time-varying curve and a steady-state intensity value after stimulation is released, and calculating the ratio of the difference value of the intensity peak value and the steady-state intensity value to the time required for restoring to the steady-state intensity value as a scattering recovery rate; carrying out interlayer boundary identification on the time sequence optical coherence tomography image sequence, tracking the track of the displacement of each interlayer boundary before and after pressure stimulation along with the change of time, fitting the track by adopting an exponential decay model, and obtaining a deformation recovery time constant; calculating covariance of the variation of the scattering intensity of each voxel before and after pressure stimulation and the corresponding interlayer boundary displacement as a mechanical optical correlation coefficient; and forming mechanical optical coupling response characteristics of the corresponding voxels by the scattering recovery rate, the deformation recovery time constant and the mechanical optical correlation coefficient.
  3. 3. The system of claim 1, wherein said obtaining metabolic dynamic response characteristics for each sampling site comprises: determining the wavelength ranges of an NADH sensitive wave band and an FAD sensitive wave band in the synchronous autofluorescence spectrum sequence; calculating the ratio of the integral intensity of the NADH sensitive wave band to the integral intensity of the FAD sensitive wave band in the autofluorescence spectrum at each moment to be used as the oxidation-reduction ratio at the moment; Counting time-varying sequences formed by oxidation-reduction ratios at all moments before and after pressure stimulation, and calculating variation coefficients of the time-varying sequences as metabolic fluctuation amplitudes; Calculating kurtosis coefficients of the self-fluorescence spectrum at each moment, and taking the difference value between the maximum value and the minimum value of the kurtosis coefficients at all moments as spectrum morphology change degree; Extracting the product of the time required for the redox ratio to recover to the baseline level after the pressure stimulus is released and the baseline level as a metabolic recovery capacity index; And forming metabolic dynamic response characteristics of the corresponding sampling sites by the metabolic fluctuation amplitude, the spectral morphology change degree and the metabolic recovery capacity index.
  4. 4. The system of claim 1, wherein the obtaining a tissue attribute confidence vector for each location comprises: vector splicing is carried out on the mechanical optical coupling response characteristic and the metabolic dynamic response characteristic of each position, so that an original joint characteristic vector is obtained; Acquiring original joint feature vectors of all adjacent positions in a preset space neighborhood of a current position; Calculating the average value of cosine similarity between the original combined feature vector of the current position and the original combined feature vectors of all adjacent positions, and taking the average value as a local consistency coefficient; calculating the mahalanobis distance between the original joint feature vector of the current position and the preset normal tissue feature center, and taking the mahalanobis distance as the normal tissue deviation degree; calculating the mahalanobis distance between the original combined feature vector of the current position and the feature center of the preset cancerous tissue as the cancerous tissue deviation degree; and normalizing the local consistency coefficient, the reciprocal of the normal tissue deviation degree and the reciprocal of the cancerous tissue deviation degree to form a tissue attribute confidence vector of the corresponding position.
  5. 5. The system of claim 1, wherein the performing preliminary region clustering based on the tissue property confidence vector to obtain candidate lesion regions comprises: Screening the position with the component value larger than the preset canceration tendency threshold value as a canceration seed point according to the component value corresponding to the reciprocal of the canceration tissue deviation degree in the tissue attribute confidence vector of each position; expanding and growing to a space adjacent position by taking each cancerous seed point as a center; Calculating vector included angles between the position to be expanded and the average vector of the tissue attribute confidence vectors of all the positions in the current growing region, and if the vector included angles are smaller than a preset expansion angle threshold, incorporating the position to be expanded into the current growing region; And repeatedly executing the expansion growth process until no expansion position meeting the condition exists, and taking the region with the growth completed as a candidate lesion region.
  6. 6. The system of claim 1, wherein the obtaining the region reliability coefficients for each candidate lesion region comprises: Calculating covariance matrixes of tissue attribute confidence vectors of all positions in a candidate lesion area, taking the trace of the covariance matrixes as internal feature dispersion, and carrying out normalization processing on the reciprocal of the internal feature dispersion to obtain internal feature homogeneity; Identifying a boundary position set and a boundary outer adjacent position set of the candidate lesion region; respectively calculating Euclidean distances between tissue attribute confidence vectors of corresponding position pairs in the boundary position set and the adjacent position set outside the boundary, and taking the average value of all the Euclidean distances as the characteristic jump amplitude; calculating the ratio of the area to the perimeter of the candidate lesion area as the area compactness; and carrying out normalization processing on the weighted products of the internal feature homogeneity, the feature jump amplitude and the region compactness to obtain the region reliability coefficient of the corresponding candidate lesion region.
  7. 7. The system of claim 1, wherein iteratively optimizing the candidate lesion region based on the region reliability coefficients results in a stable lesion region, comprising: Marking a candidate lesion region with a region reliability coefficient smaller than a preset reliability threshold value as an unstable region; Judging whether other candidate lesion areas spatially adjacent to the unstable area exist or not; if the region exists, calculating the similarity between the unstable region and the tissue attribute confidence vector mean value of each spatially adjacent candidate lesion region, taking the spatially adjacent candidate lesion region with the largest similarity as an absorption region, merging the unstable region into the absorption region to form a merging region, and recalculating the region reliability coefficient of the merging region; if the deviation value is not present, calculating a deviation value between the tissue attribute confidence vector of each position in the unstable region and the mean value vector in the region, and recalculating the region reliability coefficient after removing the position of which the deviation value is larger than the preset deviation threshold value; And repeatedly executing the optimization process until the region reliability coefficients of all the candidate lesion regions are larger than a preset reliability threshold value or the iteration times reach a preset upper limit, and obtaining a stable lesion region.
  8. 8. The system of claim 1, wherein the obtaining a lesion level label corresponding to a stable lesion region comprises: calculating mean value vectors of mechanical optical coupling response characteristics and mean value vectors of metabolic dynamic response characteristics of all positions in a stable lesion region, and splicing the mean value vectors and the mean value vectors to be used as region comprehensive characteristic vectors; Matching the region comprehensive feature vector with a benign lesion feature template set in a preset lesion pattern library, and obtaining the distance between the region comprehensive feature vector and a nearest neighbor template in the benign lesion feature template set as a benign matching distance; matching the region comprehensive feature vector with a malignant lesion feature template set in a preset lesion pattern library, and obtaining the distance between the region comprehensive feature vector and a nearest neighbor template in the malignant lesion feature template set as a malignant matching distance; Calculating the ratio of the benign matching distance to the malignant matching distance as a lesion tendency index; And determining a lesion grade label corresponding to the stable lesion region according to a preset interval where the lesion tendency index is located, wherein the lesion grade label comprises low risk, medium risk and high risk.
  9. 9. The system of claim 1, wherein the method for calculating the topological distance of the stable lesion region to the outer boundary of the incisal margin comprises: Identifying the outer boundary contour line of the incisal edge tissue and the geometric center of each stable lesion area; Calculating the shortest Euclidean distance from the geometric center of the stable lesion area to the contour line of the outer boundary of the cutting edge, and taking the shortest Euclidean distance as the geometric boundary distance; Taking the geometric center of the stable lesion area as a starting point, taking the nearest point on the contour line of the outer boundary of the cutting edge as an end point, and sampling a plurality of intermediate positions at equal intervals along the connecting line direction; Calculating the accumulated sum of reciprocal corresponding component values of normal tissue deviation degrees in the tissue attribute confidence vectors of all the intermediate positions to be used as a path normal degree accumulated value; And taking the product of the geometric boundary distance and the path normal degree cumulative value as the topological distance from the corresponding stable lesion region to the external boundary of the incisal edge.
  10. 10. The system of claim 1, wherein the evaluating the safety status of the margin region based on the lesion grade label and the topological distance stabilizing the lesion region to the margin outer boundary comprises: according to the lesion grade labels, risk weights are distributed to each stable lesion area, wherein the high risk corresponds to a first weight, the medium risk corresponds to a second weight and the low risk corresponds to a third weight, and the first weight is larger than the second weight and larger than the third weight; calculating the product of the risk weight of each stable lesion region and the reciprocal of the topological distance of the stable lesion region to be used as a single region risk contribution value; Accumulating the single-region risk contribution values of all the stable lesion regions, dividing the single-region risk contribution values by the total area of the incisal edge tissues, and normalizing the single-region risk contribution values to obtain incisal edge comprehensive risk scores; If the cutting edge comprehensive risk score is larger than a preset high risk threshold, judging that the safety state of the cutting edge area is positive; and if the comprehensive risk score of the cutting edge is smaller than the preset low risk threshold, judging that the safety state of the cutting edge region is negative.

Description

Intelligent intraoperative incisal margin evaluation system for breast cancer breast-preserving operation Technical Field The invention relates to the technical field of computer-aided medical diagnosis, in particular to an intelligent intraoperative cutting edge evaluation system for breast cancer breast-preserving surgery. Background There are some clinical problems to be solved in the existing incisional edge evaluation technology in breast cancer breast-conserving surgery. Traditional dependency on intra-operative frozen sections and post-operative routine pathology is faced with serious time efficiency bottlenecks, and surgeons often wait for pathology results before operating tables, which not only prolongs anesthesia time, but also increases infection risk. Even experienced pathologists cannot avoid sampling errors, and typical intra-operative evaluations can only examine very small parts of the incisal margin, as if a sea needle were trying to find tiny lesions in a broad incisal margin, resulting in massive tiny infiltrates and missed catheter in situ cancer extensions. In clinical practice, there is often a significant difference in interpretation of the same slice between pathologists, and the lack of a standardized quantitative evaluation system makes the results highly dependent on personal experience and subjective judgment. Traditional morphological observation cannot capture the functional abnormality and biomechanical change of tissues, just like judging icebergs by appearance only, and cannot see the danger of underwater hiding. This limitation of single modality assessment is particularly evident when dealing with complex cases, such as small She Jinrun and multifocal lesions, often misjudged by scattered distribution and atypical manifestations. In the operating room, surgeons are faced with the choice of removing too large a range that detracts from the aesthetic effect and function and too small a range that may leave cancer cells behind. This results in the patient who has found a positive incision in the postoperative pathology being faced again with the scalpel, experiencing additional physical trauma, psychological stress and economic burden, while medical resources are being excessively consumed. Secondly, the geometric distance measurement ignores the spatial variation of tissue characteristics, and the safety distance requirement of high-risk and low-risk lesions cannot be distinguished, so that the concept of 'safety cutting edge' is vague in practice. In view of the above, the present invention proposes an intelligent intraoperative margin assessment system for breast cancer breast-conserving surgery to solve the above-mentioned problems. Disclosure of Invention In order to overcome the defects in the prior art and achieve the purposes, the invention provides the following technical scheme that the intelligent intraoperative incisal margin evaluation system for breast cancer breast-preserving surgery comprises: the data acquisition module is used for acquiring a time sequence optical coherence tomography image sequence and a synchronous autofluorescence spectrum sequence of the incisional tissue before and after the preset pressure stimulation; the mechanical optical characteristic extraction module is used for acquiring the mechanical optical coupling response characteristic of each voxel according to the scattering intensity time-varying curve of each voxel in the time sequence optical coherence tomography image sequence and the displacement variation track of the interlayer boundary; The metabolic characteristic extraction module is used for obtaining metabolic dynamic response characteristics of each sampling site according to the time-varying characteristics of the oxidation-reduction ratio and the spectral kurtosis coefficient changes of the NADH sensitive wave band and the FAD sensitive wave band in the synchronous autofluorescence spectrum sequence; The lesion recognition module is used for acquiring a tissue attribute confidence vector of each position according to an original combined feature vector obtained by splicing the mechanical optical coupling response feature and the metabolic dynamic response feature of each position and combining the feature consistency degree of the space adjacent positions; The lesion classification module is used for acquiring the region reliability coefficient of each candidate lesion region according to the internal feature homogeneity of each candidate lesion region, the feature jump amplitude of the region boundary and the region compactness; The cutting edge evaluation module is used for acquiring a lesion grade label of the corresponding stable lesion region according to the layering matching result of the comprehensive feature vector of each stable lesion region and the preset lesion mode library, and evaluating the safety state of the cutting edge region based on the topological distance between the lesion grade label and the ex