CN-122023408-A - Image processing-based cryoablation needle targeting zone frosting state identification method

CN122023408ACN 122023408 ACN122023408 ACN 122023408ACN-122023408-A

Abstract

The invention discloses a method for identifying frosting state of a target area of a cryoablation needle based on image processing, and relates to the technical field of cryoablation needles. Constructing an original image sequence of a target area under a unified space-time coordinate system, outputting information of a corresponding degradation stage, obtaining a clear image sequence of the target area with high fidelity, aggregating signal subspace features through a channel gating mechanism to obtain a signal enhancement feature image set, generating a multi-scale semantic feature tensor of the target area, outputting an initial frosting probability distribution map of the target area, executing morphological constraint segmentation optimization to obtain a morphological uniform frosting area mask, obtaining a time sequence stable frosting area mask sequence, constructing a frosting state dynamic evolution model, and displaying early warning information in real time. The invention effectively suppresses the characteristic flooding problem caused by mixed noise, and obviously improves the identification degree and stability of the bottom structure information of the target area in a low-temperature imaging environment.

Inventors

ZHANG CHENG
MIAO XIAOFENG
ZHAO DEBING
ZHU CHENGYI
MAO QIANYU

Assignees

南京德文医学科技有限公司

Dates

Publication Date: 20260512
Application Date: 20260413

Claims (10)

1. The method for identifying the frosting state of the target area of the cryoablation needle based on image processing is characterized by comprising the following steps of: Acquiring multi-mode original visual data of a target area in the cryoablation operation process, performing time synchronous registration and space geometric correction, and constructing an original image sequence of the target area under a unified space-time coordinate system; generating a degradation stage label based on the original image sequence of the target area, and outputting corresponding degradation stage information; inputting the original image sequence of the target area and the degradation phase information into a phase self-adaptive DiffUIR model together to obtain a clear image sequence of the high-fidelity target area; Performing reversible residual hourglass mapping treatment on the clear image sequence of the high-fidelity target area, decomposing the potential feature space into a signal subspace and a noise subspace, and aggregating the signal subspace features through a channel gating mechanism to obtain a signal enhancement feature atlas; Inputting the signal enhancement feature atlas into a weak feature multi-scale coding network to generate a target area multi-scale semantic feature tensor; inputting the multi-scale semantic feature tensor of the target area into an iterative topologically consistent segmentation decoding network, and outputting an initial frosting probability distribution map of the target area; Constructing a morphology prior scheduler, applying connectivity constraint, curvature smoothness constraint and volume increase constraint output by the morphology prior scheduler to an initial frosting probability distribution map of a target area, and executing morphology constraint segmentation optimization to obtain a morphology consistent frosting area mask; And performing optical flow reverse mapping comparison and abnormal fine adjustment on the morphological uniform frosting area mask and the morphological uniform frosting area mask of the previous frame to obtain a time sequence stable frosting area mask sequence, constructing a frosting state dynamic evolution model, performing data interaction on the frosting state dynamic evolution model and a surgery real-time monitoring system, and displaying early warning information in real time.
2. The method for identifying the frosting state of a target area of a cryoablation needle based on image processing according to claim 1, wherein the multi-modal raw visual data of the target area comprises synchronous endoscope images, ultrasonic images, optical coherence tomography data and corresponding temperature-time curves.
3. The method for identifying the frosting state of a target area of a cryoablation needle based on image processing according to claim 1, wherein the generating a degradation stage label based on an original image sequence of the target area comprises the following steps: Extracting degradation characterization quantity from original images of each frame of target area, and constructing degradation characteristic vector composed of fog response quantity, refraction enhancement response quantity and ice crystal high reflection response quantity; Constructing an input sequence of a degradation stage detection network by using continuous frame degradation characteristic vectors, inputting the input sequence into the degradation stage detection network, and outputting degradation stage prediction vectors which correspond to the original image of the target region and are constrained by unidirectional thermodynamic phase change monotonicity; generating a degradation stage label corresponding to the original image of the target area based on the degradation stage prediction vector, and performing time sequence smoothing on the degradation stage label to obtain a smoothed degradation stage label; And outputting corresponding degradation stage information based on the smoothed degradation stage label and the degradation stage prediction vector.
4. The method for identifying the frosting state of a target area of a cryoablation needle based on image processing according to claim 1, wherein the step of inputting the original image sequence of the target area and the degradation phase information into the phase adaptive DiffUIR model comprises the steps of: constructing a diffusion coefficient for the original image of the target area of the t frame based on the smoothed degradation stage label corresponding to the original image of the target area of the t frame; performing a forward diffusion process on the original image of the t frame targeting region based on the diffusion coefficient to generate a diffusion state sequence; Carrying out noise estimation processing on each diffusion state in the diffusion state sequence based on the degradation stage information, and generating a corresponding noise estimation result and an accumulated retention coefficient; Constructing a denoising intensity adjusting coefficient in the reverse diffusion process based on the information of the degradation stage of the t frame; performing a reverse diffusion process based on the noise estimation result, the accumulated retention coefficient and the denoising intensity adjusting coefficient to obtain a clear image reconstruction result corresponding to the original image of the t frame targeting region; And outputting the reconstruction result of each frame of clear image frame by frame to construct a high-fidelity target area clear image sequence.
5. The method for identifying the frosting state of the cryoablation needle target area based on image processing according to claim 1, wherein the reversible residual hourglass mapping processing is performed on the clear image sequence of the high-fidelity target area, and the method comprises the following steps: Inputting a clear image reconstruction result in a clear image sequence of the high-fidelity target area into a reversible residual error hourglass mapping processing unit to generate a corresponding initial clear characteristic image; performing reversible residual error hourglass mapping processing on the initial clear feature map to obtain corresponding bottleneck potential features; Decomposing the potential feature space based on the bottleneck potential feature to obtain a signal subspace feature and a noise subspace feature; Constructing a channel gating coefficient based on the signal subspace characteristics and the noise subspace characteristics, enhancing the signal subspace characteristics by using the channel gating coefficient, and simultaneously inhibiting the noise subspace characteristics to obtain gating signal characteristics; and performing hourglass decoding reconstruction on the t frame gating signal characteristics to obtain a t frame signal enhancement characteristic atlas, outputting the frame-by-frame signal enhancement characteristic atlas of each frame, and constructing a total signal enhancement characteristic atlas.
6. The method for identifying frosting state of a cryoablation needle targeting zone based on image processing according to claim 1, wherein the inputting of the signal enhancement feature atlas into the weak feature multi-scale coding network comprises the following steps: inputting the signal enhancement feature atlas into a weak feature multi-scale coding network, and extracting fractional step response aiming at the signal enhancement feature atlas of each scale level to obtain phase change micro texture features of corresponding scales; Generating a space offset field based on the phase change micro-texture characteristics, and performing deformable convolutional coding on the signal enhancement characteristic map by using the space offset field to obtain morphological sensing characteristics of the self-adaptive fit ice crystal growth morphology; And performing cross-scale non-local attention interaction on the morphological perception features of each scale, fusing the global topological context with the local micro texture, and generating a target area multi-scale semantic feature tensor.
7. The method for identifying the frosting state of a target area of a cryoablation needle based on image processing according to claim 1, wherein the step of inputting the multi-scale semantic feature tensor of the target area into an iterative topologically consistent segmentation decoding network comprises the following steps: inputting the multi-scale semantic feature tensor of the target area into an iterative topological consistent segmentation decoding network, and performing cross-scale feature cascade decoding and feature resolution reduction to obtain full-resolution decoding features consistent with the spatial scale of the original image of the target area; Extracting structural topology attributes of the full-resolution decoding features, constructing a boundary perception gating mechanism, and executing topology feature correction on the full-resolution decoding features to obtain topology consistency boundary enhancement features; mapping the topological consistency boundary enhancement features into a single-channel probability space, and generating an initial frosting probability distribution map of the target area.
8. The method for identifying the frosting state of a cryoablation needle target zone based on image processing according to claim 1, wherein the applying connectivity constraint, curvature smoothness constraint and volume increase constraint output by the morphology prior scheduler to the initial frosting probability distribution map of the target zone comprises: Constructing a morphology prior scheduler, and generating a corresponding physical morphology prior field, a connectivity constraint coefficient, a curvature smooth constraint coefficient and a volume increase constraint coefficient according to an initial frosting probability distribution map of a target area; Performing connectivity constraint optimization on the initial frosting probability distribution map of the target area based on the connectivity constraint coefficient and the physical form prior field to obtain a connectivity constraint probability map; Executing curvature smoothness constraint optimization on the connectivity constraint probability map based on the curvature smoothness constraint coefficient to obtain a curvature smoothness probability map; Performing volume increase constraint optimization on the curvature smooth probability map based on the volume increase constraint coefficient to obtain a morphological uniform frosting area mask; Outputting the morphological uniform frosting area mask of each frame by frame to construct a morphological uniform frosting area mask sequence.
9. The method for identifying the frosting state of a cryoablation needle targeting zone based on image processing according to claim 1, wherein the performing optical flow reverse mapping comparison and abnormal fine tuning on the morphological uniform frosting zone mask and the morphological uniform frosting zone mask of the previous frame to obtain a time sequence stable frosting zone mask sequence and construct a frosting state dynamic evolution model comprises the following steps: Performing optical flow reverse mapping comparison and abnormal fine adjustment on the morphological uniform frosting area mask and the morphological uniform frosting area mask of the previous frame to obtain a time sequence stable frosting area mask sequence; based on a time sequence stable frosting area mask sequence, constructing a frosting three-dimensional state model, and calculating a frosting quantitative index set of a target zone; Generating a frosting state dynamic evolution model based on the frosting three-dimensional state model and the frosting quantitative index set, performing data interaction on the frosting state dynamic evolution model and the operation real-time monitoring system, and executing monitoring intervention early warning.
10. The method for identifying the frosting state of a cryoablation needle targeting zone based on image processing according to claim 9, wherein the monitoring intervention pre-warning comprises the following steps: When the frosting expansion speed is greater than a preset safe expansion speed threshold value or the minimum space distance is smaller than a preset safe distance threshold value, generating a surgery monitoring early warning signal containing a refrigeration and decompression instruction or a device shutdown instruction, and sending the surgery monitoring early warning signal to a surgery real-time monitoring system to trigger corresponding critical highlight rendering and forced device cutoff intervention; When the frosting expansion speed is smaller than or equal to a preset safe expansion speed threshold value, and the minimum space distance is larger than or equal to a preset safe distance threshold value, generating a rendering monitoring signal and sending the rendering monitoring signal to a real-time operation monitoring system to display the frosting range and the frosting speed in real time.

Description

Image processing-based cryoablation needle targeting zone frosting state identification method Technical Field The invention relates to the technical field of cryoablation needles, in particular to a method for identifying frosting state of a target area of a cryoablation needle based on image processing. Background Along with the wide application of the minimally invasive cryoablation technology in tumor treatment and precise surgery, how to accurately sense the frosting range of a target area and the dynamic evolution state of the target area in real time in surgery has become a key technical problem for guaranteeing the safety and effectiveness of the surgery. In the actual cryoablation process, the visual information of the target area collected by the imaging equipment is often in an extremely low-temperature environment and is influenced by the coagulation of tissue fluid, phase-change refraction and specular scattering of ice crystals, so that the image shows a highly complex and dynamically-changing mixed degradation characteristic along with time. In the prior art, a processing method for a medical image degradation problem mostly depends on an image enhancement or denoising algorithm for a single degradation type, for example, a defogging and denoising method based on a histogram equalization, a Retinex model or a convolutional neural network, the method generally assumes that the degradation type is relatively stable in space or time, is difficult to cope with dynamic hybrid degradation phenomena generated by coupling of multiple physical mechanisms in a cryoablation process, and when an operation enters different stages, image degradation characteristics are obviously changed, so that an original model cannot be effectively generalized, and the condition of characteristic information loss or recognition failure occurs, so that critical frosting boundary information is covered in complex noise, and the accuracy of subsequent analysis is seriously affected. In the aspect of recognition and segmentation of frosted areas, the existing mainstream methods are mostly based on a pixel-level semantic segmentation network of deep learning, mainly rely on local gray scale, texture or semantic features for classification and judgment, lack modeling capability of biological heat transfer science and phase change physical laws in the cryoablation process, and easily generate the problems of discontinuous segmentation results, void generation inside or false detection of isolated plaques in non-frosted areas under the condition of extremely low contrast between soft tissues of a target area and the frosted areas, namely obvious topological structure errors and non-physical form artifacts. Disclosure of Invention The invention aims to provide an image processing-based identification method for the frosting state of a cryoablation needle targeting region, which can effectively inhibit the characteristic flooding problem caused by mixed noise and remarkably improve the identification degree and the stability of the bottom structure information of the targeting region in a low-temperature imaging environment. According to the embodiment of the invention, the identification method for the frosting state of the cryoablation needle targeting area based on image processing comprises the following steps: Acquiring multi-mode original visual data of a target area in the cryoablation operation process, performing time synchronous registration and space geometric correction, and constructing an original image sequence of the target area under a unified space-time coordinate system; generating a degradation stage label based on the original image sequence of the target area, and outputting corresponding degradation stage information; inputting the original image sequence of the target area and the degradation phase information into a phase self-adaptive DiffUIR model together to obtain a clear image sequence of the high-fidelity target area; Performing reversible residual hourglass mapping treatment on the clear image sequence of the high-fidelity target area, decomposing the potential feature space into a signal subspace and a noise subspace, and aggregating the signal subspace features through a channel gating mechanism to obtain a signal enhancement feature atlas; Inputting the signal enhancement feature atlas into a weak feature multi-scale coding network to generate a target area multi-scale semantic feature tensor; inputting the multi-scale semantic feature tensor of the target area into an iterative topologically consistent segmentation decoding network, and outputting an initial frosting probability distribution map of the target area; Constructing a morphology prior scheduler, applying connectivity constraint, curvature smoothness constraint and volume increase constraint output by the morphology prior scheduler to an initial frosting probability distribution map of a target area, and executing morphology constrai