CN-122023442-A - Method, device, terminal equipment and storage medium for dividing mesenteric fascia line

CN122023442ACN 122023442 ACN122023442 ACN 122023442ACN-122023442-A

Abstract

The application discloses a method, a device, a terminal device and a storage medium for segmenting a mesenteric fascia line, belonging to the field of image processing; calculating a significance value corresponding to each pixel point in each first time-space feature map, rearranging the pixel points in the first time-space feature map according to the significance value, obtaining a plurality of scale second time-space feature maps according to the rearranged first time-space feature map, calculating a variability value corresponding to each pixel block in each second time-space feature map, rearranging the pixel blocks in the corresponding second time-space feature map according to the variability value, obtaining a plurality of scale third time-space feature maps according to the rearranged second time-space feature map, and obtaining a mesenteric fascial line segmentation result according to all the third time-space feature maps. The application can improve the segmentation efficiency of the mesenteric fascia line.

Inventors

WU XIAOJIAN
GAO FENG
CAI ZERONG
CAI DU
WU KECHENG

Assignees

中山大学附属第六医院

Dates

Publication Date: 20260512
Application Date: 20260127

Claims (10)

1. A method of segment mesenteric fascial lines comprising: obtaining a video frame of a laparoscope video, and extracting first time-space feature maps of a plurality of scales corresponding to the video frame; Calculating a significance value corresponding to each pixel point in each first time-space feature map, rearranging the pixel points corresponding to the first time-space feature map according to the significance values to obtain a corresponding first pixel point sequence, and processing the first pixel point sequence through the state space processing operator to obtain a second time-space feature map with a plurality of scales; calculating a variability value corresponding to each pixel block in each second space-time feature map, rearranging the pixel blocks corresponding to the second space-time feature map according to the variability value to obtain a corresponding second pixel point sequence, and processing each second pixel point sequence through the state space processing operator to obtain a third space-time feature map with a plurality of scales; And obtaining a mesenteric fascial line segmentation result according to all the third space-time characteristic diagrams.
2. The method for segmentation of a mesenteric fascia line according to claim 1, wherein the extracting the first temporal spatial feature map of the plurality of scales corresponding to the video frame comprises: Acquiring front and rear adjacent frames corresponding to the video frames, and respectively extracting coding features of a plurality of scales corresponding to the video frames and the front and rear adjacent frames; splicing the coding features of the video frames and the front and rear adjacent frames under the same scale to obtain splicing features of a plurality of scales; and carrying out channel attention weighting on each spliced characteristic to obtain the first time-space characteristic diagram under each scale.
3. The method for segmentation of a mesenteric fascia line according to claim 2, further comprising downsampling the video frame and the front and rear neighboring frames after the video frame and the front and rear neighboring frames are acquired to obtain downsampled results, and performing data enhancement on the downsampled results.
4. The method of claim 1, wherein calculating a saliency value corresponding to each pixel in each of the first temporal-spatial feature map, and rearranging pixels in the corresponding first temporal-spatial feature map according to the saliency values to obtain a corresponding first sequence of pixels comprises: Extracting the characteristic value of each pixel point in each first time space characteristic diagram on all channels; Calculating the average value of the characteristic values of each pixel point on all corresponding channels so as to determine the significance value corresponding to each pixel point; And sequencing all the pixels in each first time-space feature map from large to small according to the significance value to obtain a first pixel sequence corresponding to each first time-space feature map.
5. The method of claim 1, wherein calculating the variability value corresponding to each pixel block in each of the second spatiotemporal feature maps, and reordering the pixel blocks corresponding to the second spatiotemporal feature maps according to the variability values, to obtain the corresponding second sequence of pixel points comprises: calculating a significance value corresponding to each pixel point of each second space-time characteristic diagram; Calculating standard deviation values of significance values corresponding to all pixel points in each pixel block to obtain variability values corresponding to the pixel blocks; Ordering the pixel blocks in each second space-time feature map from large to small according to the variability value to obtain a corresponding pixel block sequence; and flattening the pixel blocks in each pixel block sequence to obtain a second pixel point sequence corresponding to each second space-time characteristic diagram.
6. The method of claim 1, wherein obtaining the results of the segmentation of the mesenteric fascial lines from all of the third spatiotemporal feature maps comprises: Adding each third space-time feature map with the first space-time feature map with the same scale to obtain a fourth space-time feature map with a plurality of scales; Performing point convolution processing on each fourth time-space feature map to obtain fifth time-space feature maps with a plurality of scales; And inputting a fifth space-time characteristic diagram with a plurality of scales into a preset encoder to obtain the segmentation result of the mesenteric fascial line.
7. The method for segmentation of a mesenteric fascia line according to any one of claims 1 to 6, wherein the state space processing operator comprises a first normalization layer, a bi-directional Manba layer, a second normalization layer and a channel-by-channel convolution layer, and when the state space processing operator receives any pixel point sequence, the method further comprises: normalizing the pixel point sequence through the first normalization layer to obtain a first sequence; Performing forward and backward parallel scanning on the first sequence through the bidirectional Manbab layer to obtain an enhancement sequence, and adding the pixel point sequence and the enhancement sequence to obtain a second sequence; Normalizing the second sequence through the second normalization layer to obtain a third sequence, and remapping the third sequence into a first feature map; and carrying out channel-by-channel convolution operation on the first feature map through the channel-by-channel convolution layer to obtain a second feature map, and adding the first feature map and the second feature map to obtain a space-time feature map corresponding to the pixel point sequence.
8. The mesenteric fascia segmentation device is characterized by comprising a feature extraction module, a global shape sensing module, a local variability sensing module and a segmentation result prediction module; the feature extraction module is used for obtaining a video frame of the laparoscopic video and extracting first time-space feature graphs of a plurality of scales corresponding to the video frame; the global shape sensing module is used for calculating a significance value corresponding to each pixel point in each first time space feature map, rearranging the pixel points corresponding to the first time space feature map according to the significance values to obtain a corresponding first pixel point sequence, and processing the first pixel point sequence through the state space processing operator to obtain a second time space feature map with a plurality of scales; The local variability perception module is used for calculating a variability value corresponding to each pixel block in each second space-time feature map, rearranging the pixel blocks corresponding to the second space-time feature map according to the variability value to obtain a corresponding second pixel point sequence, and processing each second pixel point sequence through the state space processing operator to obtain a third space-time feature map with a plurality of scales; and the segmentation result prediction module is used for obtaining a segmentation result of the mesenteric fascial line according to all the third space-time characteristic diagrams.
9. A terminal device comprising a processor, a memory and a computer program stored in the memory and configured to be executed by the processor, the processor implementing the method of junction mesenteric fascial line segmentation according to any one of claims 1 to 7 when executing the computer program.
10. A computer readable storage medium, characterized in that the computer readable storage medium comprises a stored computer program, wherein the computer program, when run, controls a device in which the computer readable storage medium is located to perform the method of junction mesenteric fascial line segmentation according to any one of claims 1-7.

Description

Method, device, terminal equipment and storage medium for dividing mesenteric fascia line Technical Field The present application relates to the field of image processing, and in particular, to a method, an apparatus, a terminal device, and a storage medium for segmentation of a mesenteric fascia line. Background The mesenteric fascia (Toldt) line is an important marker for reaching the fusion fascia and entering the fusion fascia space, and can be used for clearly distinguishing the anatomical structure of the intestinal tract from the abdominal wall in a laparoscopic video, so that accurate identification and cutting of the Toldt line is important. In order to solve the above problems, some of the prior art guides the attention mechanism based on the fractal characteristics of the laparoscopic video image, so as to enhance the distinguishing capability of the attention mechanism to different anatomical structures in the laparoscopic video, so as to improve the accuracy of segmenting Toldt's line, but the attention calculation adopted by the method has higher complexity, and requires a large amount of computational resources during training and reasoning, so that the calculation efficiency in the Toldt's line segmentation process is reduced. Disclosure of Invention The application provides a method, a device, terminal equipment and a storage medium for segmenting a mesenteric fascia line, which can solve the problem of how to improve the segmentation efficiency of Toldt lines in the prior art. Some embodiments of the application provide a method of segmentation of a mesenteric fascia line comprising: obtaining a video frame of a laparoscope video, and extracting first time-space feature maps of a plurality of scales corresponding to the video frame; Calculating a significance value corresponding to each pixel point in each first time-space feature map, rearranging the pixel points corresponding to the first time-space feature map according to the significance values to obtain a corresponding first pixel point sequence, and processing the first pixel point sequence through the state space processing operator to obtain a second time-space feature map with a plurality of scales; calculating a variability value corresponding to each pixel block in each second space-time feature map, rearranging the pixel blocks corresponding to the second space-time feature map according to the variability value to obtain a corresponding second pixel point sequence, and processing each second pixel point sequence through the state space processing operator to obtain a third space-time feature map with a plurality of scales; And obtaining a mesenteric fascial line segmentation result according to all the third space-time characteristic diagrams. Compared with the prior art, the method has the advantages that firstly, through extracting the first time space characteristic diagram of the dry scale of the video frame in the laparoscopic video and calculating the pixel level saliency value, the model can distinguish Toldt's line related area from a large area background area in the encoding stage, key characteristics with anatomical significance are highlighted from the source, and the interference of irrelevant information on subsequent calculation is reduced. Secondly, rearranging pixel points based on significance values, modeling features according to significance sequences by utilizing a state space processing operator, enabling computing resources to be preferentially used in Toldt's line high-correlation areas, and realizing efficient modeling of linear structure global dependency relationship under the condition of not introducing high-complexity attention computation, so that the real-time performance of data processing is improved. Further, through calculating the standard deviation of the pixel block saliency value, rearrangement and reinforcement processing are carried out on the region with severe local saliency change, so that the key region with fuzzy boundary and easy misclassification is effectively highlighted, the whole calculation redundancy is reduced, and the whole efficiency of Toldt's line segmentation in the laparoscopic video is improved. Further, the extracting a first time-space feature map of a plurality of scales corresponding to the video frame includes: Acquiring front and rear adjacent frames corresponding to the video frames, and respectively extracting coding features of a plurality of scales corresponding to the video frames and the front and rear adjacent frames; splicing the coding features of the video frames and the front and rear adjacent frames under the same scale to obtain splicing features of a plurality of scales; and carrying out channel attention weighting on each spliced characteristic to obtain the first time-space characteristic diagram under each scale. Compared with the prior art, the embodiment has the advantages that the multi-scale coding characteristics of the current video f