Search

CN-121987341-A - Intelligent path planning system for epileptic focus excision with multimode image fusion

CN121987341ACN 121987341 ACN121987341 ACN 121987341ACN-121987341-A

Abstract

The invention discloses an intelligent path planning system for epileptic focus excision with multimode image fusion, and belongs to the field of medical instruments. The system integrates the structural image and the functional image, realizes sub-millimeter-level accurate registration through a trans-form-based trans-modal feature fusion network, and locates a focus by combining intracranial electrophysiology data (the Dice coefficient is more than or equal to 0.9). And (3) generating a multi-target optimized path by adopting an improved A-algorithm, integrating path length, brain tissue damage and risk index (12 constraint conditions including white matter fiber bundle integrity, vascularity, functional area boundary and the like), and shortening planning time to 15-20 minutes. Constructing an anatomic-functional two-dimensional risk assessment system, and quantitatively analyzing the fiber bundle density, the blood vessel distance and the risk of the functional area. The operation navigation module supports 5G real-time transmission and dynamic registration, and automatically triggers a correction mechanism when the deviation exceeds 2mm, and the hardware configuration meets the requirement that the frame rate of three-dimensional image processing is more than or equal to 15fps.

Inventors

  • LIU JIAJUN
  • WANG YAN
  • GAO YUAN
  • DU HUIMIN

Assignees

  • 电子科技大学

Dates

Publication Date
20260508
Application Date
20260308

Claims (8)

  1. 1. The intelligent path planning system for the epileptic focus excision is characterized by comprising a multimode image acquisition module, an image fusion module, a focus positioning module, an intelligent path planning module, a risk assessment module and a surgery navigation module, wherein the multimode image acquisition module is used for synchronously acquiring structural images and functional image data of a patient, the structural images comprise MRI sequence images and CT images, the functional images comprise PET metabolic images and fMRI function activation images, the image fusion module adopts a cross-modal feature fusion algorithm based on deep learning, related features of the structural images and the functional images are extracted through a self-attention mechanism to achieve sub-millimeter level accurate registration and pixel level fusion of the multimode images, the focus positioning module is used for combining intracranial electrophysiological data to achieve three-dimensional accurate positioning of an epileptic focus based on the fused image features, the intelligent path planning module adopts a multi-objective optimization algorithm to generate an optimal surgery path, comprehensive consideration path length, brain tissue damage degree and risk index are generated through an improved A-based algorithm to meet the minimum turning angle constraint, the risk assessment module calculates the path index to quantify surgery risks, builds multi-dimensional risk including white matter fiber bundle integrity, important blood vessel distribution and brain function region boundaries, and automatically adjusts the deviation to a preset path by means of the algorithm to achieve the automatic deviation correction and the automatic deviation correction when the deviation is triggered by the automatic deviation in the preset path adjustment system.
  2. 2. The intelligent path planning system for epileptic focus resection based on multimode image fusion according to claim 1, wherein the image fusion module adopts a trans-former-based trans-modal feature fusion network, the network structure comprises an encoder and a decoder, a self-attention module is introduced into a bottleneck layer to strengthen focus region feature weights, algorithm parameters are set to be an initial learning rate of 0.001, an Adam optimizer is adopted for 500 times, and registration errors are controlled to be 0.5mm.
  3. 3. The intelligent path planning system for epileptic focus excision with multi-mode image fusion according to claim 1, wherein the focus positioning module realizes accurate focus sketch through dual-mode feature fusion, extracts a high signal area of a FLAIR sequence as an anatomical marker, performs preliminary segmentation by adopting an Otsu threshold method, simultaneously performs time-frequency analysis and positioning on three-dimensional coordinates issued by spike waves on EEG data, fuses image features and electrophysiological positioning results through a Dempster-Shafer evidence theory, and finally achieves a Dice coefficient of a focus boundary of 0.9.
  4. 4. The intelligent path planning system for epileptic focal resection with multi-mode image fusion according to claim 1, wherein an objective function of the intelligent path planning module is defined as cost=α·distance+β· Damage +γ·risk, wherein α, β, γ are weight coefficients, respectively set to 0.3, 0.4, 0.3, distance represents path length, damage quantifies cortex injury volume, risk is an integrated Risk index output by the Risk assessment module, and heuristic functions are designed as h (n) =d (n) ·w, d (n) is an euclidean Distance, and w is a Risk weighting coefficient.
  5. 5. The multimode image fused epileptic focus excision intelligent path planning system according to claim 1, wherein the risk assessment module parameters comprise an anisotropic score value of key fiber bundles such as a corticospinal bundle, an arch-shaped bundle and the like based on DTI data, fiber bundle density through which a path passes, extraction of middle cerebral artery branches by adopting CTA data, calculation of the shortest distance between the path and a blood vessel, calculation of the three-dimensional distance between a path endpoint and a boundary of a functional area by adopting an fMRI positioning language area and a motion area, and calculation of a risk index by adopting a weighted summation model, wherein the fiber bundle density weight is 40%, the blood vessel distance weight is 30%, and the functional area distance weight is 30%.
  6. 6. The intelligent path planning system for epileptic focus resection with multi-mode image fusion according to claim 1, wherein the operation navigation module adopts a 5G transmission protocol to transmit a real-time image acquired by an ultrasonic probe to a processing terminal, the dynamic registration algorithm adopts an improved mutual information method, the preoperative fusion image is taken as a reference, artifact points in the ultrasonic image are removed through a random sampling consistency algorithm, an error threshold is set to be 1.5mm, and the system automatically triggers a path adjustment mechanism when a path deviation is detected to be greater than 2 mm.
  7. 7. The intelligent path planning system for epileptic focus resection with multi-mode image fusion according to claim 1, wherein the image data acquired by the multi-mode image acquisition module accords with DICOM 3.0 standard, the layer thickness is controlled to be 0.5-1 mm, the matrix size is not lower than 512 x 512, the mri scanning comprises T1 weighted image, T2 weighted image and FLAIR sequence, the CT scanning adopts 64 rows of spiral thin-layer scanning, the PET metabolic imaging adopts 18F-FDG tracer, the EEG data sampling rate is set to 1024Hz, and the recording duration is 24 hours.
  8. 8. The multimode image fused intelligent path planning system for epileptic focus resection according to claim 1, wherein an interactive interface based on three-dimensional visualization is provided, a planned path is supported to be adjusted in real time by an operator, and an interface design integrates an image labeling function, a risk early warning function and a path simulation function.

Description

Intelligent path planning system for epileptic focus excision with multimode image fusion Technical Field The invention relates to medical equipment, in particular to an intelligent path planning system for epileptic focus excision with multimode image fusion. Background Epilepsy is a common neurological disorder, and about 30% of epileptic patients develop drug-refractory epilepsy, requiring surgical removal of lesions. Accurate practice of epileptic focal resections relies on comprehensive assessment of focal location, functional zone boundaries, and surgical path, but the prior art systems have significant limitations. In imaging applications, traditional surgical planning relies primarily on single structure images, such as MRI or CT. Although the images can clearly show the anatomical structure of the brain, the metabolic activity boundary between the brain functional area and the focus is difficult to accurately identify, so that the risk of positioning deviation of the functional area is increased. The existing research shows that the error of focus positioning can reach 3-5mm only by relying on structural images, and the operation accuracy is seriously affected. In the path planning link, the current neurosurgery operation mostly adopts a manual design mode, and the path selection is carried out by relying on experience of an operator. The method is high in subjectivity and low in planning efficiency, the average planning time is as long as 60-90 minutes, and the quick decision requirement of complex cases is difficult to meet. More importantly, the difference of experience levels of different operators leads to poor consistency of planning schemes and influences the controllability of operation quality. In terms of risk assessment, the prior art performs mechanical collision detection based only on anatomical structures, ignoring the potential risk of damage to the brain function network by the surgical path. For example, while current commercial image navigation systems such as Medtronic StealthStation provide real-time position tracking, the impact of the path on key functional areas such as language, motion, etc. cannot be quantified, resulting in a post-operative neurological dysfunction incidence of still up to 8-15%. The method comprises the following steps of (1) cutting structural images and functional information, and not realizing anatomic-functional fusion positioning, (2) low planning efficiency, time-consuming manual path design, poor consistency and obvious influence of experience of operators, and (3) one-sided risk assessment, namely only focusing on anatomic damage and lacking quantitative prediction of functional damage. The technical bottlenecks limit the accuracy and safety of epileptic surgery, and an integrated solution integrating multi-modal images, intelligent path optimization and functional risk assessment is needed to be constructed. Disclosure of Invention The invention aims to provide an intelligent path planning system for epileptic focus resection by multi-mode image fusion, which aims to solve the technical problems of single image information, low path planning efficiency and one-sided risk assessment in the prior art. The intelligent path planning system for the epileptic focus excision comprises a multimode image acquisition module, an image fusion module, a focus positioning module, an intelligent path planning module, a risk assessment module and a surgery navigation module, wherein the multimode image acquisition module is used for synchronously acquiring structural images and functional image data of a patient, the structural images comprise MRI sequence images and CT images, the functional images comprise PET metabolic images and fMRI function activation images, the image fusion module adopts a cross-modal feature fusion algorithm based on deep learning, related features of the structural images and the functional images are extracted through a self-attention mechanism to achieve sub-millimeter level accurate registration and pixel level fusion of the multimode images, the focus positioning module is used for carrying out three-dimensional accurate positioning of an epileptic focus by combining intracranial electrophysiological data based on the fused image features, the intelligent path planning module adopts a multi-objective optimization algorithm to generate an optimal surgery path, the comprehensive consideration path length, brain tissue damage degree and risk index are generated through improving an A-type algorithm, the risk assessment module calculates the path index to quantify surgery risk, comprises white matter fiber bundle integrity, important blood vessel distribution and brain function region boundary multidimensional risk, and automatically adjusts the deviation to a real-time threshold value when the deviation is corrected by the algorithm in a support system, and the real-time deviation is triggered when the deviation is corrected to the p