CN115375621B - Liver tumor ablation path planning method and device - Google Patents
Liver tumor ablation path planning method and device Download PDFInfo
- Publication number
- CN115375621B CN115375621B CN202210798863.2A CN202210798863A CN115375621B CN 115375621 B CN115375621 B CN 115375621B CN 202210798863 A CN202210798863 A CN 202210798863A CN 115375621 B CN115375621 B CN 115375621B
- Authority
- CN
- China
- Prior art keywords
- ablation
- tumor
- planning
- ablation path
- path
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000002679 ablation Methods 0.000 title claims abstract description 206
- 238000000034 method Methods 0.000 title claims abstract description 41
- 206010019695 Hepatic neoplasm Diseases 0.000 title claims abstract description 14
- 208000014018 liver neoplasm Diseases 0.000 title claims abstract description 14
- 206010028980 Neoplasm Diseases 0.000 claims abstract description 71
- 210000004185 liver Anatomy 0.000 claims abstract description 37
- 238000013528 artificial neural network Methods 0.000 claims abstract description 23
- 230000002787 reinforcement Effects 0.000 claims abstract description 23
- 230000002440 hepatic effect Effects 0.000 claims abstract description 15
- 230000011218 segmentation Effects 0.000 claims description 46
- 210000000056 organ Anatomy 0.000 claims description 21
- 230000003187 abdominal effect Effects 0.000 claims description 19
- 210000001519 tissue Anatomy 0.000 claims description 19
- 230000009471 action Effects 0.000 claims description 16
- 230000007246 mechanism Effects 0.000 claims description 10
- 210000004204 blood vessel Anatomy 0.000 claims description 8
- 238000012549 training Methods 0.000 claims description 8
- 210000000988 bone and bone Anatomy 0.000 claims description 7
- 230000007704 transition Effects 0.000 claims description 7
- 239000011159 matrix material Substances 0.000 claims description 6
- 230000006870 function Effects 0.000 claims description 5
- 238000005457 optimization Methods 0.000 abstract description 4
- 238000004364 calculation method Methods 0.000 description 5
- 239000003795 chemical substances by application Substances 0.000 description 5
- 238000003780 insertion Methods 0.000 description 5
- 230000037431 insertion Effects 0.000 description 5
- 238000011282 treatment Methods 0.000 description 4
- 239000002775 capsule Substances 0.000 description 3
- 230000008569 process Effects 0.000 description 3
- 238000011298 ablation treatment Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000003993 interaction Effects 0.000 description 2
- 210000001015 abdomen Anatomy 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 210000004556 brain Anatomy 0.000 description 1
- 201000011510 cancer Diseases 0.000 description 1
- 239000008358 core component Substances 0.000 description 1
- 238000006073 displacement reaction Methods 0.000 description 1
- 238000012854 evaluation process Methods 0.000 description 1
- 238000009472 formulation Methods 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 210000005228 liver tissue Anatomy 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0012—Biomedical image inspection
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B34/00—Computer-aided surgery; Manipulators or robots specially adapted for use in surgery
- A61B34/10—Computer-aided planning, simulation or modelling of surgical operations
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B34/00—Computer-aided surgery; Manipulators or robots specially adapted for use in surgery
- A61B34/20—Surgical navigation systems; Devices for tracking or guiding surgical instruments, e.g. for frameless stereotaxis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
- G06V10/267—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion by performing operations on regions, e.g. growing, shrinking or watersheds
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B34/00—Computer-aided surgery; Manipulators or robots specially adapted for use in surgery
- A61B34/10—Computer-aided planning, simulation or modelling of surgical operations
- A61B2034/101—Computer-aided simulation of surgical operations
- A61B2034/105—Modelling of the patient, e.g. for ligaments or bones
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B34/00—Computer-aided surgery; Manipulators or robots specially adapted for use in surgery
- A61B34/10—Computer-aided planning, simulation or modelling of surgical operations
- A61B2034/107—Visualisation of planned trajectories or target regions
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B34/00—Computer-aided surgery; Manipulators or robots specially adapted for use in surgery
- A61B34/20—Surgical navigation systems; Devices for tracking or guiding surgical instruments, e.g. for frameless stereotaxis
- A61B2034/2046—Tracking techniques
- A61B2034/2065—Tracking using image or pattern recognition
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10072—Tomographic images
- G06T2207/10081—Computed x-ray tomography [CT]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30056—Liver; Hepatic
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
- Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Theoretical Computer Science (AREA)
- Medical Informatics (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Surgery (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Software Systems (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Evolutionary Computation (AREA)
- Molecular Biology (AREA)
- Animal Behavior & Ethology (AREA)
- Veterinary Medicine (AREA)
- Artificial Intelligence (AREA)
- Public Health (AREA)
- Computing Systems (AREA)
- Heart & Thoracic Surgery (AREA)
- Robotics (AREA)
- Biomedical Technology (AREA)
- Multimedia (AREA)
- Radiology & Medical Imaging (AREA)
- Mathematical Physics (AREA)
- Quality & Reliability (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- Apparatus For Radiation Diagnosis (AREA)
Abstract
The invention provides a method and a device for planning a hepatic tumor ablation path. The method comprises the following steps: using a deep neural network to segment the tumor area of the input multi-phase liver CT image; determining an ablation target point in the segmented tumor area based on the ablation needle parameters; and establishing an ablation path planning constraint condition, and planning an optimal ablation path by using reinforcement learning. The invention utilizes the reinforcement learning method, automatically obtains the optimal ablation path through repeated iterative optimization, improves the liver tumor ablation efficiency, and solves the problems of low efficiency, large error, incomplete ablation range and the like in the traditional manual ablation path planning.
Description
Technical Field
The invention belongs to the technical field of medical images, and particularly relates to a liver tumor ablation path planning method and device.
Background
Liver tumors are one of the most common malignant tumors in the world, and the incidence rate is increasing in recent years. According to the progress of liver tumor, different treatment methods can be clinically adopted, wherein the liver tumor ablation treatment is a minimally invasive treatment method which is widely applied clinically at present.
The main process and principle of the liver tumor ablation treatment are as follows: firstly, determining the positions of tumor focus and surrounding tissue structures in a CT image mode; then the ablation needle is inserted into the tumor position in the patient, and the high Wen Duiju focus generated by the ablation needle is ablated, so that the treatment of liver tumor is achieved. Before treatment, a doctor needs to make an ablation scheme according to the distribution situation of the tumor position and surrounding organs, blood vessels and the like, namely, determining the needle insertion path of an ablation needle and the specific ablation position. Generally, the ablation area formed by the ablation needle is an ellipsoid (which can be also be approximately a sphere for the convenience of calculation), and the ablation needles with different specifications can ablate the ellipsoid areas with different sizes. Each needle insertion, the ablation needle is required to be inserted into the patient from the skin needle insertion point and advanced to the ablation site to ablate the ellipsoidal region near the site. There are two difficulties in the formulation of ablation protocols clinically: firstly, based on manual observation and analysis, accurate calculation and planning are difficult to achieve; and secondly, depending on clinical experience of doctors, time and labor are wasted. Therefore, the invention provides an automatic ablation path planning method and device based on reinforcement learning.
Disclosure of Invention
In order to solve the problems in the prior art, the invention provides a method and a device for planning a hepatic tumor ablation path.
In order to achieve the above object, the present invention adopts the following technical scheme.
In a first aspect, the present invention provides a method for planning a hepatic tumor ablation path, comprising the steps of:
using a deep neural network to segment the tumor area of the input multi-phase liver CT image;
determining an ablation target point in the segmented tumor area based on the ablation needle parameters;
and establishing an ablation path planning constraint condition, and planning an optimal ablation path by using reinforcement learning.
Further, the method for segmenting the tumor region comprises the following steps:
constructing a neural network segmentation model based on the HR-net depth;
a training data set consisting of a multi-phase liver CT image and a corresponding label is established, and the segmentation model is trained by utilizing the data set;
and inputting the multi-phase liver CT image of the patient into a trained segmentation model, and segmenting out a tumor region.
Further, the method for determining an ablation target comprises the following steps:
a minimum circumscribed ellipsoid for the tumor region;
the major axis and the minor axis of the ellipsoid are respectively extended outwards for a certain length to obtain an ellipsoid A, the maximum diameter of A is 2L, and the volume is V 0 ；
Calculate centroid coordinate I of A 1 And do with I 1 Is a sphere with the sphere center and the maximum ablation needle length l as the radius, and the sphere volume is V 1 Wherein l is>L/2；
If V 1 ≥V 0 Then I 1 The point is the unique ablation target point;
if V 1 <V 0 And the right side l of the left end point and the left side l of the right end point on the maximum diameter of the A are respectively a first ablation target point and a second ablation target point.
Further, the method for establishing the ablation path planning safety constraint condition comprises the following steps:
constructing a 3DU-net deep neural network segmentation model based on an attention mechanism;
dividing abdominal tissue organs including blood vessels and bones from the input multi-phase liver CT images by using the trained division model;
and determining the minimum distance that the ablation path avoids the abdominal tissue organ based on the segmentation result, and obtaining the ablation path planning safety constraint condition.
Still further, the constraint further includes:
the length of the ablation path is smaller than the maximum length of the ablation needle;
the included angle between the ablation path and the liver envelope is more than 20 degrees.
Further, the method of planning an optimal ablation path using reinforcement learning includes:
s1, initializing a gain expected matrix Q (S, a), wherein S is a state, namely an ablation needle position, a is a direction vector determined by an action corresponding to S, namely an ablation needle direction under a current sub-path, and fixed parameters alpha and gamma in a Markov chain state transition equation are set;
s2, according to an epsilon greedy strategy, based on the state S of the current time t t Action a t And constraint conditions, calculating a reward function r t And the state s at the next moment t+1 ；
S3, updating Q (S) according to the Markov chain state transition equation t ,a t )：
New Q(s t ,a t )＝Q(s t ,a t )+α[r t +γ×maxQ(s t+1 ,a t+1 )-Q(s t ,a t )]
In the formula, new Q(s) t ,a t ) Is Q(s) t ,a t ) Is a new value of maxQ (s t+1 ,a t+1 ) Represented in state s t+1 The maximum expected value of income which can be obtained by executing all actions is lower;
s4, repeating the step S3 until the target point is ablated, and obtaining a path;
s5, repeating the steps S2-S4 until Q (S) t ,a t ) Local minima are reached, resulting in an optimal ablation path.
In a second aspect, the present invention provides a hepatic tumor ablation path planning apparatus, comprising:
the tumor region segmentation module is used for segmenting the tumor region of the input multi-phase liver CT image by using the deep neural network;
the ablation target point determining module is used for determining an ablation target point in the segmented tumor area based on the ablation needle parameters;
and the ablation path planning module is used for establishing an ablation path planning constraint condition and planning an optimal ablation path by using reinforcement learning.
Further, the tumor region segmentation module is specifically configured to:
constructing a neural network segmentation model based on the HR-net depth;
a training data set consisting of a multi-phase liver CT image and a corresponding label is established, and the segmentation model is trained by utilizing the data set;
and inputting the multi-phase liver CT image of the patient into a trained segmentation model, and segmenting out a tumor region.
Further, the ablation target determination module is specifically configured to:
a minimum circumscribed ellipsoid for the tumor region;
the major axis and the minor axis of the ellipsoid are respectively extended outwards for a certain length to obtain an ellipsoid A, the maximum diameter of A is 2L, and the volume is V 0 ；
Calculate centroid coordinate I of A 1 And do with I 1 Is a sphere with the sphere center and the maximum ablation needle length l as the radius, and the sphere volume is V 1 Wherein l is>L/2；
If V 1 ≥V 0 Then I 1 The point is the unique ablation target point;
if V 1 <V 0 And the right side l of the left end point and the left side l of the right end point on the maximum diameter of the A are respectively a first ablation target point and a second ablation target point.
Further, the ablation path planning module establishes an ablation path planning safety constraint condition according to the following steps:
constructing a 3DU-net deep neural network segmentation model based on an attention mechanism;
dividing abdominal tissue organs including blood vessels and bones from the input multi-phase liver CT images by using the trained division model;
and determining the minimum distance that the ablation path avoids the abdominal tissue organ based on the segmentation result, and obtaining the ablation path planning safety constraint condition.
Compared with the prior art, the invention has the following beneficial effects.
According to the method, the depth neural network is utilized to segment the tumor area of the input multi-phase liver CT image, the ablation target point is determined in the segmented tumor area based on the ablation needle parameters, the ablation path planning constraint condition is established, and the optimal ablation path is planned by reinforcement learning, so that the full-automatic planning of the ablation path is realized. The invention utilizes the reinforcement learning method, automatically obtains the optimal ablation path through repeated iterative optimization, improves the liver tumor ablation efficiency, and solves the problems of low efficiency, large error, incomplete ablation range and the like in the traditional manual ablation path planning.
Drawings
Fig. 1 is a flowchart of a method for planning a hepatic tumor ablation path according to an embodiment of the present invention.
Fig. 2 is a schematic overall flow chart of hepatic tumor ablation path planning.
Fig. 3 is a block diagram of a hepatic tumor ablation path planning apparatus according to an embodiment of the present invention.
Detailed Description
The present invention will be further described with reference to the drawings and the detailed description below, in order to make the objects, technical solutions and advantages of the present invention more apparent. It will be apparent that the described embodiments are only some, but not all, embodiments of the invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
Fig. 1 is a flowchart of a method for planning a hepatic tumor ablation path according to an embodiment of the present invention, including the following steps:
step 101, using a deep neural network to segment tumor areas of an input multi-phase liver CT image;
102, determining an ablation target point in a segmented tumor area based on an ablation needle parameter;
and 103, establishing an ablation path planning constraint condition, and planning an optimal ablation path by using reinforcement learning.
In this embodiment, step 101 is mainly used to determine an ablation region. The ablation region refers to a tumor region, and in this embodiment, the tumor region is segmented from the input liver CT image by building a deep neural network model and using the trained model. The existing segmentation method is generally used for positioning tumor focus based on single-phase CT images only, and changes of pixel characteristics of focus at different phases are ignored, so that the problem of non-ideal positioning accuracy exists. Therefore, the method and the device for detecting the liver tumor by using the multi-phase liver CT image combined training fully utilize the image features of the tumor focus in different phases of the liver CT, and simultaneously combine the information on the time space, so that the accuracy of automatic positioning of the liver tumor area is improved, and the reliability is improved.
In this embodiment, step 102 is mainly used to determine an ablation target. The ablation target point is the position of the ablation needle in the tumor area when ablation is performed (the position of the point of the needle is generally taken as the ablation target point). In order to be able to ablate the tumor region effectively, it should be ensured that the effective ablation space of the ablation needle can completely cover the tumor region (and in general a certain safety margin is also set aside, i.e. a certain space is extended outwards for the actual tumor region). The number and optimal location of ablation targets can be designed according to the size parameters of the ablation needle and the location and size of the tumor area. The shape of the ablation needle may be approximated as an ellipsoid; although not very canonical, the actual tumor region may also be approximated as a canonical volume, such as an ellipsoid, to simplify the calculation.
In this embodiment, step 103 is mainly used to plan the optimal ablation path. Most of the existing ablation path planning at present is based on manual planning, and a doctor with years of clinical experience is relied on to locate a tumor focus. However, conventional manual ablation path planning is inefficient and may cause certain errors, resulting in incomplete ablation coverage. Therefore, the embodiment utilizes the reinforcement learning method, so that the robot automatically learns the optimal ablation path (the number of ablation needles is minimum, the adjustment times in ablation puncture are minimum, and the like) through repeated iterative optimization, thereby greatly improving the efficiency of doctors and realizing high-efficiency full-automatic path planning. Of course, the necessary ablation path constraints should also be established during reinforcement learning to ensure that the needle insertion direction does not touch untouchable abdominal tissue organs, etc.
Reinforcement learning belongs to unsupervised machine learning and comprises 5 core components: environment (Environment), agent (State), action (Action), rewards (report). Reinforcement learning refers to learning as a heuristic evaluation process in which an agent selects an action for an environment, the state of the environment changes after receiving the action, and a reinforcement signal (reward value) is generated and fed back to the agent, and the agent reselects the next action according to the reinforcement signal and the current state of the environment, wherein the selection principle is that the received positive reward value is maximized. The ablation path planning of the embodiment can be converted into an optimal strategy solving problem under the constraint condition, and the finally planned ablation path is linearly represented by a series of sub-paths, namely L obj ＝f(l 1 +l 2 +…l t + …), these sub-paths l t All can be obtained from the current point s t Status (ablation needle position) and decision action a t (spatial puncture angle of sub-path); at the same time, the sub-paths do not meet random independent conditions, and each sub-path is affected by the sub-path at the previous moment, so that an ablation path can be represented as a Markov chain. While the determination of the ablation path may be considered a markov decision process. Considering that the Markov decision process is continuously completed by a agent through interaction with the external environment (the three-dimensional space environment of the abdomen), the unpredictable interaction between the ablation needle and liver tissue and displacement deformation of the liver after being stressed are realizedExpressed as transition probabilities, such a model can well represent the randomness of the next ablation needle position.
The method for automatically obtaining the optimal ablation path through repeated iterative optimization by utilizing the reinforcement learning method improves the liver tumor ablation efficiency and solves the problems of low efficiency, large error, incomplete ablation range and the like in the traditional manual ablation path planning.
As an alternative embodiment, the method for segmenting a tumor region includes:
constructing an HR-net deep neural network segmentation model;
a training data set consisting of a multi-phase liver CT image and a corresponding label is established, and the segmentation model is trained by utilizing the data set;
and inputting the multi-phase liver CT image of the patient into a trained segmentation model, and segmenting out a tumor region.
The embodiment provides a technical scheme for dividing the tumor area. In the embodiment, the segmentation of the tumor region is realized by constructing an HR-net deep neural network segmentation model. And inputting the preprocessed CT image data and corresponding label data into an HR-net deep neural segmentation network, and performing iterative training until convergence to obtain a segmentation model for reasoning. And inputting CT image data of the patient into a trained segmentation model, and obtaining a tumor region through reasoning. HR-net (high-resolution network) is a high-resolution network, and the high-resolution subnetwork is used as a first stage, so as to gradually increase the high-resolution subnetwork to the low-resolution subnetwork, form more stages, and connect the multi-resolution subnetworks in parallel. Through multi-scale fusion, each high-resolution to low-resolution representation repeatedly receives information from other parallel representations, so that rich high-resolution representations are obtained. HR-net has very effective application in pixel-level classification, region-level classification, and image-level classification.
As an alternative embodiment, the method for determining an ablation target includes:
a minimum circumscribed ellipsoid for the tumor region;
length of the ellipsoidThe shaft and the short shaft are respectively extended outwards for a certain length to obtain an ellipsoid A, the maximum diameter of A is set to be 2L, and the volume is set to be V 0 ；
Calculate centroid coordinate I of A 1 And do with I 1 Is a sphere with the sphere center and the maximum ablation needle length l as the radius, and the sphere volume is V 1 Wherein l is>L/2；
If V 1 ≥V 0 Then I 1 The point is the unique ablation target point;
if V 1 <V 0 And the right side l of the left end point and the left side l of the right end point on the maximum diameter of the A are respectively a first ablation target point and a second ablation target point.
The embodiment provides a technical scheme for determining an ablation target point. First, to simplify the calculation, the tumor region is approximated as an ellipsoid by making the smallest circumscribed ellipsoid of the tumor region. Then, in order to leave a certain safety margin, the ellipsoid is outwards expanded into a larger ellipsoid A, and the effective ablation space of the ablation needle can be ensured to completely cover the ellipsoid A when an ablation target point is determined. The effective ablation area of the ablation needle is a sphere with the needle (tip) as the center of the sphere and the maximum ablation needle length l as the radius. According to the geometrical characteristics of the ellipsoid, the position of the ablation target point should be on the maximum diameter of the ellipsoid A. Firstly, calculating the centroid coordinate I of an ellipsoid A 1 (in fact, the center of the ellipsoid A) and then I 1 Is a sphere with a sphere center and a radius of l. The ball volume was noted as V 1 The volume of the ellipsoid A is denoted as V 0 . If V is 1 ≥V 0 I.e. the volume of the sphere is not less than the volume of A, the sphere is considered to completely cover A, i.e. only one needle is needed to achieve the purpose of ablation, thus I 1 Is the only ablation target point. If V is 1 <V 0 I.e. the volume of the sphere is smaller than the volume of a, it is considered that one sphere cannot completely cover a, and at least a second needle is required. In practice it has been found that in general one or two needles can achieve the effective ablation requirement, and therefore this embodiment defines a maximum diameter of the ellipsoid A which is less than twice the diameter of the sphere, i.e. l>L/2. Thus, two ablation targets are positioned on the maximum diameter of A and can be respectively arranged onIs positioned at two endpoints of maximum diameter, namely, at the right side l of the left endpoint and at the left side l of the right endpoint.
As an alternative embodiment, the method for establishing the ablation path planning safety constraint condition includes:
constructing a 3DU-net deep neural network segmentation model based on an attention mechanism;
dividing abdominal tissue organs including blood vessels and bones from the input multi-phase liver CT images by using the trained division model;
and determining the minimum distance that the ablation path avoids the abdominal tissue organ based on the segmentation result, and obtaining the ablation path planning safety constraint condition.
The embodiment provides a technical scheme for establishing the safety constraint condition of ablation path planning. The safety constraint is that the ablation needle does not damage the organs of the abdominal tissues, the path of the ablation needle is required to avoid all the organs of the abdominal tissues, and a certain safety margin is reserved, namely, the minimum distance from the organs is set. Thus, to obtain the constraints, the area occupied by the organs of the abdominal tissue must be determined. In the embodiment, the abdominal tissue organ comprising blood vessels and bones is segmented from the input multi-phase liver CT image by constructing a 3DU-net deep neural network segmentation model and utilizing the trained segmentation model. Compared with a 2D network model, the 3D network model can learn the context information of the abdominal tissue and organ in the three-dimensional space more fully, and a more accurate segmentation effect is obtained. In order to effectively extract the image features of interest, the segmentation model of the present embodiment incorporates an attention mechanism. The attention mechanism is that only some key information inputs are concerned for processing to improve the efficiency of the neural network by referring to the attention mechanism of the human brain under the condition that the computer has limited capability. The calculation of the attention mechanism can be divided into two steps: firstly, calculating the attention distribution on all input information; and secondly, calculating the weighted summation of the input information according to the attention distribution. Weighting coefficients, i.e. attention profile alpha i ＝softmax(f i W att q) representing the input vector f i Degree of correlation with the query vector q.
As an alternative embodiment, the constraint further includes:
the length of the ablation path is smaller than the maximum length of the ablation needle;
the included angle between the ablation path and the liver envelope is more than 20 degrees.
The present embodiment expands constraints on the ablation path planning. In addition to the security constraints given by the previous embodiment, this embodiment also gives some hard constraints: firstly, limiting the maximum length of an ablation path, wherein the required path length is smaller than the maximum length of an ablation needle; secondly, in order to prevent the sliding needle, the minimum value of the included angle between the direction of the ablation path and the liver capsule is limited, and the included angle is required to be larger than 20 degrees. The path here refers to a sub-path, and each sub-path is a straight line segment with unchanged needle insertion direction. In addition, the constraint conditions also include soft conditions, such as that the distance between the ablation path and the abdominal organ is as large as possible, and the included angle between the ablation path and the liver capsule is as close to 90 degrees as possible, i.e. the ablation path is perpendicular to the liver capsule as much as possible.
As an alternative embodiment, the method of planning an optimal ablation path using reinforcement learning includes:
s1, initializing a gain expected matrix Q (S, a), wherein S is a state, namely an ablation needle position, a is a direction vector determined by an action corresponding to S, namely an ablation needle direction under a current sub-path, and fixed parameters alpha and gamma in a Markov chain state transition equation are set;
s2, according to an epsilon greedy strategy, based on the state S of the current time t t Action a t And constraint conditions, calculating a reward function r t And the state s at the next moment t+1 ；
S3, updating Q (S) according to the Markov chain state transition equation t ,a t )：
New Q(s t ,a t )＝Q(s t ,a t )+α[r t +γ×maxQ(s t+1 ,a t+1 )-Q(s t ,a t )]
In the formula, new Q(s) t ,a t ) Is Q(s) t ,a t ) Is a new value of maxQ (s t+1 ,a t+1 ) Represented in state s t+1 The maximum expected value of income which can be obtained by executing all actions is lower;
s4, repeating the step S3 until the target point is ablated, and obtaining a path;
s5, repeating the steps S2-S4 until Q (S) t ,a t ) Local minima are reached, resulting in an optimal ablation path.
The present embodiment presents the execution steps of planning an optimal ablation path using reinforcement learning. Reinforcement learning draws ablation paths based on iterative laws. Step S1 initializes the gain expectation matrix Q (S, a). Step S2, calculating a reward function in the current state and the state of the next moment according to an epsilon greedy strategy. The epsilon greedy strategy is an uncertainty strategy, balances utilization and exploration, wherein the part with the maximum action value function is selected as the utilization, and the probability is still kept to find the globally optimal solution as the exploration part. For example, let e=0.1, a probability of 1-0.1=0.9 is utilized, and a probability of 0.1 is explored. That is, the probability of selecting the current maximum return is 0.9, and the probability of searching for the global optimal solution for exploration is 0.1. And S3, obtaining a desired matrix according to a Markov chain state transition equation. And repeatedly executing the step S3 until the target point is ablated, and obtaining a path. And repeating the steps until the expected profit matrix reaches a local minimum value, and obtaining the optimal ablation path.
Fig. 3 is a schematic diagram of a hepatic tumor ablation path planning apparatus according to an embodiment of the present invention, the apparatus includes:
the tumor region segmentation module 11 is used for segmenting the tumor region of the input multi-phase liver CT image by using a deep neural network;
an ablation target determination module 12, configured to determine an ablation target in the segmented tumor region based on the ablation needle parameter;
the ablation path planning module 13 is configured to establish an ablation path planning constraint condition and plan an optimal ablation path by reinforcement learning.
The device of this embodiment may be used to implement the technical solution of the method embodiment shown in fig. 1, and its implementation principle and technical effects are similar, and are not described here again. As well as the latter embodiments, will not be explained again.
As an alternative embodiment, the tumor region segmentation module 11 is specifically configured to:
constructing a neural network segmentation model based on the HR-net depth;
a training data set consisting of a multi-phase liver CT image and a corresponding label is established, and the segmentation model is trained by utilizing the data set;
and inputting the multi-phase liver CT image of the patient into a trained segmentation model, and segmenting out a tumor region.
As an alternative embodiment, the ablation target determination module 12 is specifically configured to:
a minimum circumscribed ellipsoid for the tumor region;
the major axis and the minor axis of the ellipsoid are respectively extended outwards for a certain length to obtain an ellipsoid A, the maximum diameter of A is 2L, and the volume is V 0 ；
Calculate centroid coordinate I of A 1 And do with I 1 Is a sphere with the sphere center and the maximum ablation needle length l as the radius, and the sphere volume is V 1 Wherein l is>L/2；
If V 1 ≥V 0 Then I 1 The point is the unique ablation target point;
if V 1 <V 0 And the right side l of the left end point and the left side l of the right end point on the maximum diameter of the A are respectively a first ablation target point and a second ablation target point.
As an alternative embodiment, the ablation path planning module 13 establishes the ablation path planning safety constraint as follows:
constructing a 3DU-net deep neural network segmentation model based on an attention mechanism;
dividing abdominal tissue organs including blood vessels and bones from the input multi-phase liver CT images by using the trained division model;
and determining the minimum distance that the ablation path avoids the abdominal tissue organ based on the segmentation result, and obtaining the ablation path planning safety constraint condition.
The foregoing is merely illustrative of the present invention, and the present invention is not limited thereto, and any changes or substitutions easily contemplated by those skilled in the art within the scope of the present invention should be included in the present invention. Therefore, the protection scope of the invention is subject to the protection scope of the claims.
Claims (8)
1. A method for planning a hepatic tumor ablation path, comprising the steps of:
using a deep neural network to segment the tumor area of the input multi-phase liver CT image;
determining an ablation target point in the segmented tumor area based on the ablation needle parameters;
establishing an ablation path planning constraint condition, and planning an optimal ablation path by using reinforcement learning;
the method for determining the ablation target comprises the following steps:
a minimum circumscribed ellipsoid for the tumor region;
the major axis and the minor axis of the ellipsoid are respectively extended outwards for a certain length to obtain an ellipsoid A, the maximum diameter of A is 2L, and the volume is V 0 ；
Calculate centroid coordinate I of A 1 And do with I 1 Is a sphere with the sphere center and the maximum ablation needle length l as the radius, and the sphere volume is V 1 Wherein l is>L/2；
If V 1 ≥V 0 Then I 1 The point is the unique ablation target point;
if V 1 <V 0 And the right side l of the left end point and the left side l of the right end point on the maximum diameter of the A are respectively a first ablation target point and a second ablation target point.
2. The method of planning a hepatic tumor ablation path according to claim 1, wherein the method of segmenting the tumor region comprises:
constructing a neural network segmentation model based on the HR-net depth;
a training data set consisting of a multi-phase liver CT image and a corresponding label is established, and the segmentation model is trained by utilizing the data set;
and inputting the multi-phase liver CT image of the patient into a trained segmentation model, and segmenting out a tumor region.
3. The method of claim 1, wherein establishing an ablation path planning safety constraint comprises:
constructing a 3D U-net deep neural network segmentation model based on an attention mechanism;
dividing abdominal tissue organs including blood vessels and bones from the input multi-phase liver CT images by using the trained division model;
and determining the minimum distance that the ablation path avoids the abdominal tissue organ based on the segmentation result, and obtaining the ablation path planning safety constraint condition.
4. A method of planning a hepatic tumor ablation path according to claim 3, wherein the constraints further comprise:
the length of the ablation path is smaller than the maximum length of the ablation needle;
the included angle between the ablation path and the liver envelope is more than 20 degrees.
5. The method of planning a hepatic tumor ablation path according to claim 1, wherein the method of planning an optimal ablation path using reinforcement learning comprises:
s1, initializing a gain expected matrix Q (S, a), wherein S is a state, namely an ablation needle position, a is a direction vector determined by an action corresponding to S, namely an ablation needle direction under a current sub-path, and fixed parameters alpha and gamma in a Markov chain state transition equation are set;
s2, according to an epsilon greedy strategy, based on the state S of the current time t t Action a t And constraint conditions, calculating a reward function r t And the state s at the next moment t+1 ；
S3, turning according to the state of the Markov chainEquation is shifted, and Q(s) is updated as follows t ,a t )：
New Q(s t ,a t )＝Q(s t ,a t )+α[r t +γ×maxQ(s t+1 ,a t+1 )-Q(s t ,a t )]
In the formula, new Q(s) t ,a t ) Is Q(s) t ,a t ) Is a new value of maxQ (s t+1 ,a t+1 ) Represented in state s t+1 The maximum expected value of income which can be obtained by executing all actions is lower;
s4, repeating the step S3 until the target point is ablated, and obtaining a path;
s5, repeating the steps S2-S4 until Q (S) t ,a t ) Local minima are reached, resulting in an optimal ablation path.
6. A hepatic tumor ablation path planning apparatus, comprising:
the tumor region segmentation module is used for segmenting the tumor region of the input multi-phase liver CT image by using the deep neural network;
the ablation target point determining module is used for determining an ablation target point in the segmented tumor area based on the ablation needle parameters;
the ablation path planning module is used for establishing an ablation path planning constraint condition and planning an optimal ablation path by reinforcement learning;
the method for determining the ablation target comprises the following steps:
a minimum circumscribed ellipsoid for the tumor region;
the major axis and the minor axis of the ellipsoid are respectively extended outwards for a certain length to obtain an ellipsoid A, the maximum diameter of A is 2L, and the volume is V 0 ；
Calculate centroid coordinate I of A 1 And do with I 1 Is a sphere with the sphere center and the maximum ablation needle length l as the radius, and the sphere volume is V 1 Wherein l is>L/2；
If V 1 ≥V 0 Then I 1 The point is the unique ablation target point;
if V 1 <V 0 And the right side l of the left end point and the left side l of the right end point on the maximum diameter of the A are respectively a first ablation target point and a second ablation target point.
7. The hepatic tumor ablation path planning apparatus of claim 6, wherein the tumor region segmentation module is specifically configured to:
constructing a neural network segmentation model based on the HR-net depth;
a training data set consisting of a multi-phase liver CT image and a corresponding label is established, and the segmentation model is trained by utilizing the data set;
and inputting the multi-phase liver CT image of the patient into a trained segmentation model, and segmenting out a tumor region.
8. The liver tumor ablation path planning apparatus of claim 6, wherein the ablation path planning module establishes the ablation path planning safety constraints by:
constructing a 3D U-net deep neural network segmentation model based on an attention mechanism;
dividing abdominal tissue organs including blood vessels and bones from the input multi-phase liver CT images by using the trained division model;
and determining the minimum distance that the ablation path avoids the abdominal tissue organ based on the segmentation result, and obtaining the ablation path planning safety constraint condition.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210798863.2A CN115375621B (en) | 2022-07-08 | 2022-07-08 | Liver tumor ablation path planning method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210798863.2A CN115375621B (en) | 2022-07-08 | 2022-07-08 | Liver tumor ablation path planning method and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115375621A CN115375621A (en) | 2022-11-22 |
| CN115375621B true CN115375621B (en) | 2023-09-05 |
Family
ID=84061576
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210798863.2A Active CN115375621B (en) | 2022-07-08 | 2022-07-08 | Liver tumor ablation path planning method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115375621B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116936033B (en) * | 2023-09-18 | 2023-11-17 | 海杰亚（北京）医疗器械有限公司 | Percutaneous ablation operation planning method, percutaneous ablation operation planning device, percutaneous ablation operation planning medium and percutaneous ablation operation planning equipment |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103970988A (en) * | 2014-04-14 | 2014-08-06 | 中国人民解放军总医院 | Ablation needle insertion path planning method and system |
| CN105286988A (en) * | 2015-10-12 | 2016-02-03 | 北京工业大学 | CT image-guided liver tumor thermal ablation needle location and navigation system |
| CN105997245A (en) * | 2016-01-28 | 2016-10-12 | 杭州奥视图像技术有限公司 | Method for precisely simulating radiofrequency ablation technology by utilizing ellipsoid to cover tumor |
| CN109961443A (en) * | 2019-03-25 | 2019-07-02 | 北京理工大学 | Liver neoplasm dividing method and device based on the guidance of more phase CT images |
| CN110570924A (en) * | 2019-08-26 | 2019-12-13 | 天津大学 | Tumor ablation-oriented remote diagnosis and treatment system and method |
| CN112258514A (en) * | 2020-11-20 | 2021-01-22 | 福州大学 | Segmentation method of pulmonary blood vessels of CT (computed tomography) image |
| CN113456219A (en) * | 2021-06-15 | 2021-10-01 | 艾瑞迈迪医疗科技（北京）有限公司 | Liver cancer radio frequency ablation operation path planning method and device based on CT image |
| CN113952030A (en) * | 2021-10-28 | 2022-01-21 | 北京深睿博联科技有限责任公司 | Planning method and device for needle insertion path and ablation position of radio-frequency electrode |
| CN114451990A (en) * | 2020-11-09 | 2022-05-10 | 北京理工大学 | Intraoperative dynamic path optimization method and device for tumor thermal ablation |
-
2022
- 2022-07-08 CN CN202210798863.2A patent/CN115375621B/en active Active
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103970988A (en) * | 2014-04-14 | 2014-08-06 | 中国人民解放军总医院 | Ablation needle insertion path planning method and system |
| CN105286988A (en) * | 2015-10-12 | 2016-02-03 | 北京工业大学 | CT image-guided liver tumor thermal ablation needle location and navigation system |
| CN105997245A (en) * | 2016-01-28 | 2016-10-12 | 杭州奥视图像技术有限公司 | Method for precisely simulating radiofrequency ablation technology by utilizing ellipsoid to cover tumor |
| CN109961443A (en) * | 2019-03-25 | 2019-07-02 | 北京理工大学 | Liver neoplasm dividing method and device based on the guidance of more phase CT images |
| CN110570924A (en) * | 2019-08-26 | 2019-12-13 | 天津大学 | Tumor ablation-oriented remote diagnosis and treatment system and method |
| CN114451990A (en) * | 2020-11-09 | 2022-05-10 | 北京理工大学 | Intraoperative dynamic path optimization method and device for tumor thermal ablation |
| CN112258514A (en) * | 2020-11-20 | 2021-01-22 | 福州大学 | Segmentation method of pulmonary blood vessels of CT (computed tomography) image |
| CN113456219A (en) * | 2021-06-15 | 2021-10-01 | 艾瑞迈迪医疗科技（北京）有限公司 | Liver cancer radio frequency ablation operation path planning method and device based on CT image |
| CN113952030A (en) * | 2021-10-28 | 2022-01-21 | 北京深睿博联科技有限责任公司 | Planning method and device for needle insertion path and ablation position of radio-frequency electrode |
Non-Patent Citations (1)
| Title |
|---|
| Robot-assisted flexible needle insertion using universal distributional deep reinforcement learning;Xiaoyu Tan 等;《International Journal of Computer Assisted Radiology and Surgery》;正文"Methods"部分 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115375621A (en) | 2022-11-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| JP5039295B2 (en) | Imaging system for use in medical intervention procedures | |
| US8233681B2 (en) | Methods, systems, and computer program products for hierarchical registration between a blood vessel and tissue surface model for a subject and a blood vessel and tissue surface image for the subject | |
| US20070167784A1 (en) | Real-time Elastic Registration to Determine Temporal Evolution of Internal Tissues for Image-Guided Interventions | |
| Zhang et al. | Multi‐needle localization with attention U‐net in US‐guided HDR prostate brachytherapy | |
| CN112862833B (en) | Blood vessel segmentation method, electronic device and storage medium | |
| Zheng et al. | Multi-part modeling and segmentation of left atrium in C-arm CT for image-guided ablation of atrial fibrillation | |
| CN113693725B (en) | Needle insertion path planning method, device, equipment and storage medium | |
| CN112244883B (en) | Method and system for extracting left auricle data parameters based on CT image | |
| CN113952030B (en) | Planning method and device for needle insertion path and ablation position of radio frequency electrode | |
| CN115375621B (en) | Liver tumor ablation path planning method and device | |
| Smit et al. | Pelvis: Atlas-based surgical planning for oncological pelvic surgery | |
| CN115294124B (en) | Ultrasonic puncture guiding planning system based on multi-mode medical image registration | |
| US20230223129A1 (en) | Automatic planning method and device for tissue ablation | |
| CN114845642A (en) | Intelligent measurement assistance for ultrasound imaging and associated devices, systems, and methods | |
| CN114831729B (en) | Left auricle plugging simulation system for fusion of echocardiography and CT multi-mode images | |
| CN115530973B (en) | Ablation visualization method and system | |
| JP6039876B2 (en) | System and method for measuring left ventricular torsion | |
| CN103340602B (en) | Method for optimizing optimal viewing angle of multiple branch interesting blood vessel section | |
| CN112043377B (en) | Ultrasound visual field simulation auxiliary ablation path planning method and system for any section of CT | |
| CN117197346A (en) | Three-dimensional ultrasonic imaging processing method, system, electronic device and readable storage medium | |
| Guo et al. | Ultrasound frame-to-volume registration via deep learning for interventional guidance | |
| CN115486935B (en) | Ablation determination method and system | |
| Mitra et al. | A thin-plate spline based multimodal prostate registration with optimal correspondences | |
| CN114419032A (en) | Method and device for segmenting the endocardium and/or the epicardium of the left ventricle of the heart | |
| Mao et al. | Direct 3d ultrasound fusion for transesophageal echocardiography |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20240307 Address after: 100039 No. 69, Yongding Road, Beijing, Haidian District Patentee after: Third Medical Center, General Hospital of the Chinese PLA Country or region after: China Address before: 100039 No. 69, Yongding Road, Beijing, Haidian District Patentee before: Wang Guisheng Country or region before: China Patentee before: Chen Xiaoxia Patentee before: Hua Rongrong |
|
| TR01 | Transfer of patent right |