Search

CN-121709272-B - Method and system for predicting risk of recurrence after PELD operation with embedded enhanced attention

CN121709272BCN 121709272 BCN121709272 BCN 121709272BCN-121709272-B

Abstract

The invention provides a PELD postoperative recurrence risk prediction method and system with embedded enhanced attention, which belong to the technical field of medical image processing and artificial intelligence, and comprise the steps of constructing LiteNet network models comprising residual block groups, wherein the residual block groups comprise 4 continuous residual block stages, each residual block stage is respectively formed by connecting 2 lightweight residual blocks in series, an MAFM attention enhancement module is embedded in each lightweight residual block, and analyzing and predicting lumbar medical images to be predicted by utilizing the trained LiteNet network models. The invention can realize accurate and efficient prediction of the recurrence risk after the PELD operation by embedding the reinforced attention.

Inventors

  • WANG JILAI
  • Qiao deyang
  • WEI JIANLU
  • SONG KANGLE
  • CHENG LEI

Assignees

  • 山东大学

Dates

Publication Date
20260512
Application Date
20260213

Claims (8)

  1. 1. A method for predicting risk of recurrence after PELD surgery with enhanced attention, comprising: historical data of a PELD patient is obtained, and a training set is divided through score matching; The method comprises the steps of constructing LiteNet a network model, wherein the LiteNet network model comprises an input layer, an initial convolution module, a residual block group and an output processing module, wherein the residual block group comprises 4 continuous residual block stages, each residual block stage is formed by connecting 2 lightweight residual blocks in series, and each lightweight residual block is internally embedded with an MAFM attention enhancement module for enhancing the morphology and the characteristics of recurrent key characteristics based on a triple cooperative mechanism, the triple cooperative mechanism comprises medical channel characteristic statistical calibration, focus perception self-adaptive sampling and cross-scale constraint, and the realization of cooperative optimization of the lightweight residual blocks comprises main path optimization, first convolution sequence optimization, second convolution sequence optimization, jump connection intelligent adaptation and fusion mechanism injection, wherein the main path optimization adopts a depth separable convolution structure of 3×3 depth convolution+1×1 point-by-point convolution; Training the built LiteNet network model based on the training set, analyzing the lumbar medical image to be predicted by using the trained LiteNet network model, and outputting a PELD postoperative recurrence risk prediction result.
  2. 2. The method for predicting risk of recurrence after PELD surgery with enhanced attention as claimed in claim 1, wherein the initial convolution module comprises a convolution layer, a batch normalization layer, GELU activation functions and a max pooling layer for preliminary extraction and downsampling of image features, and the output processing module comprises an adaptive pooling layer, a dropout layer and a fully connected output layer.
  3. 3. The method for predicting recurrence risk after PELD surgery with enhanced attention as claimed in claim 1, wherein the statistical calibration of medical channel characteristics comprises calculating texture complexity of each channel based on channel variance, calculating pixel distribution asymmetry of each channel based on channel skewness to adapt to focus characteristics under low contrast of medical image, and finally weighting and fusing the channel variance and the channel skewness through a learnable parameter, and generating final channel weight through Sigmoid activation.
  4. 4. The method for predicting risk of recurrence after PELD surgery with enhanced attention of claim 1, wherein the focus perception adaptive sampling comprises locating candidate focus areas based on a feature map with enhanced attention of channels, simulating scale perception of the candidate focus areas by interpolation, calculating sampling weights, and generating final spatial weights by accumulating multi-scale sampling weights and normalizing and Sigmoid activation.
  5. 5. The method for predicting risk of post-operative recurrence of an embedded enhanced-attention PELD of claim 1, wherein said cross-scale constraint comprises constraining a scale perception range of the model by fixing a medical lesion scale prior and a learnable scale weight.
  6. 6. An in-line attention-enhancing PELD postoperative recurrence risk prediction system, comprising: The data acquisition module is configured to acquire historical data of the PELD patient and divide a training set through score matching; The model construction module is configured to construct LiteNet a network model, wherein the LiteNet network model comprises an input layer, an initial convolution module, a residual block group and an output processing module, the residual block group comprises 4 continuous residual block stages, each residual block stage is respectively formed by connecting 2 lightweight residual blocks in series, and each lightweight residual block is internally embedded with a MAFM attention enhancement module for enhancing the morphology and the characteristics of recurrent key characteristics based on a triple cooperative mechanism, the triple cooperative mechanism comprises medical channel characteristic statistical calibration, focus perception self-adaptive sampling and cross-scale constraint, and the realization of cooperative optimization of the lightweight residual blocks comprises main path optimization, first convolution sequence optimization, second convolution sequence optimization, intelligent adaptation of jump connection and fusion mechanism injection, wherein the main path optimization adopts a depth separable convolution structure of 3×3 depth convolution+1×1 point-by-point convolution; a model training module configured to train the built LiteNet network model based on the training set; and the result prediction module is configured to analyze the lumbar medical image to be predicted by using the trained LiteNet network model and output a PELD postoperative recurrence risk prediction result.
  7. 7. A computer readable storage medium having a program stored thereon, which when executed by a processor, implements the steps of the method for in-line attention-enhancing PELD postoperative recurrence risk prediction as claimed in any of claims 1-5.
  8. 8. An electronic device comprising a memory, a processor, and a program stored on the memory and executable on the processor, wherein the processor, when executing the program, performs the steps in the method for in-line enhanced-of-attention PELD postoperative recurrence risk prediction as claimed in any of claims 1-5.

Description

Method and system for predicting risk of recurrence after PELD operation with embedded enhanced attention Technical Field The invention belongs to the technical field of medical image processing and artificial intelligence, and particularly relates to a PELD postoperative recurrence risk prediction method and system with embedded enhanced attention. Background The statements in this section merely provide background information related to the present disclosure and may not necessarily constitute prior art. Percutaneous Endoscopic Lumbar Discectomy (PELD) is the mainstream minimally invasive surgery for treating Lumbar Disc Herniation (LDH), has the advantages of small muscle damage, small bleeding in surgery and the like, but the postoperative recurrence problem (r-LDH) still seriously affects the treatment effect. The existing research shows that the occurrence rate of r-LDH is 5% -18%, and the treatment prognosis of postoperative recurrence patients is often poor, so that the accurate preoperative recurrence risk assessment and personalized treatment scheme establishment have important clinical significance. At present, the prediction of r-LDH mainly depends on subjective judgment (such as the herniated disc size, modic change and the like) of MRI images by doctors and basic clinical information (such as age and BMI), and has the problems of strong subjectivity in evaluation and single prediction dimension. In this regard, existing deep learning models (e.g., resNet and ResNet) and traditional attention enhancement models, while applied in image classification tasks, have significant limitations in r-LDH prediction scenarios: (1) Model redundancy, namely, the parameters of the deep learning model ResNet with excellent performance in the prior art can reach tens of millions, so that the calculation cost is high when the analysis operation of the medical image is processed, and the model is difficult to be deployed on low-calculation-force equipment of a basic medical institution. (2) The prediction accuracy is low, and in the prior art, in order to improve the attention to a key region in a medical image, the attention is generally enhanced by adopting a traditional CBAM paradigm which depends on pooling +MLP +convolution. However, when the method is used for processing lumbar MRI small sample data, fine pathological features (such as intervertebral disc degeneration potential signals) are easily lost due to feature compression, and the method cannot be adapted to the characteristics of low contrast of medical images and fixed focus scale. Disclosure of Invention In order to overcome the defects in the prior art, the invention provides the method and the system for predicting the risk of recurrence after PELD operation with embedded enhanced attention, and the accurate and efficient prediction of the risk of recurrence after PELD operation can be realized through embedded enhanced attention. To achieve the above object, one or more embodiments of the present invention provide the following technical solutions: the first aspect of the present invention provides a method for predicting risk of recurrence after PELD surgery with enhanced attention. A method for predicting risk of recurrence after PELD surgery with enhanced attention in-line, comprising: historical data of a PELD patient is obtained, and a training set is divided through score matching; Constructing LiteNet an 18 network model, wherein the LiteNet network model comprises an input layer, an initial convolution module, a residual block group and an output processing module, wherein the residual block group comprises 4 continuous residual block stages, each residual block stage is respectively formed by connecting 2 lightweight residual blocks in series, and each lightweight residual block is internally embedded with an MAFM attention enhancement module for enhancing the morphology and the characteristics of recurrent key characteristics based on a triple cooperative mechanism; Training the built LiteNet network model based on the training set, analyzing the lumbar medical image to be predicted by using the trained LiteNet network model, and outputting a PELD postoperative recurrence risk prediction result. The initial convolution module comprises a convolution layer, a batch normalization layer, GELU activation functions and a maximum pooling layer, and is used for preliminary extraction and downsampling of image features, and the output processing module comprises an adaptive pooling layer, a dropout layer and a fully-connected output layer. Further, the triple synergy mechanisms include medical channel feature statistical calibration, lesion perception adaptive sampling, and cross-scale constraints. The medical channel feature statistical calibration method comprises the steps of calculating texture complexity of each channel based on channel variance, calculating pixel distribution asymmetry of each channel based on channel skewness to adapt