A method and system for optimal body position combination optimization of a capsule endoscope

Abstract

The application discloses a kind of optimal body position combination optimization method and system of capsule endoscope, the method is through original problem serialization→trajectory sequence probability modeling→state transition probability modeling→body position combination evaluation modeling→body position combination optimization, trajectory sequence probability modeling is realized by the mode of state transition probability modeling, a kind of value evaluation method for different body position combination is proposed, more universal optimal body position combination in probability space is realized quickly find;Particularly, the application provides two state transition probability modeling methods: statistical modeling method based on real capsule endoscopy data and probability quantization modeling method based on doctor experience space;On the one hand, the method can greatly reduce the dependence on professional gastrointestinal tract diameter doctor experience in the process of capsule endoscopy;On the other hand, the body position path optimized by the method has universality, and also lays a certain foundation for realizing future capsule endoscopy intelligent diagnosis and treatment mode.

Description

Technical Field

[0001] This invention relates to the field of capsule endoscopy, and more particularly to a method and system for finding the optimal combination of body positions for capsule endoscopy. Background Technology

[0002] Currently, gastroscopy and colonoscopy are the best means of early diagnosis and prevention of gastrointestinal tumors. However, there is a severe shortage of gastrointestinal endoscopists. Traditional gastroscopy and colonoscopy are painful, high-risk, and have poor compliance, failing to meet the public's needs for gastrointestinal tumor prevention and control. Capsule endoscopy is an innovative medical technology that has emerged in recent years, allowing for comfortable examination of the entire digestive tract. However, it cannot yet replace the painful and high-risk traditional gastroscopy and colonoscopy for the following reasons: Each capsule endoscopy examination generates more than 50,000 images, greatly increasing the workload of radiologists; due to its inherent control limitations, capsule endoscopy cannot provide sufficient coverage and precise location information for certain parts of the digestive tract, thus easily leading to missed lesions and misjudgments of anatomical locations; in particular, for capsule endoscopes without a magnetic control module, although the discomfort caused by magnetic control can be avoided, it also limits the control methods of the capsule within the digestive tract. Generally, these capsules often require the subject to make corresponding changes in body position under the guidance of a professional physician to adjust the appropriate movement of the capsule within the digestive tract, and the system automatically records the anatomical structures captured during this process. Especially in the relatively complex structure of the stomach, achieving a more comprehensive view of the structures often requires highly experienced gastroenterologists to determine the approximate position and orientation of the capsule endoscope based on real-time images transmitted by the capsule. This guidance enables the patient to make targeted postural changes and guide the movement of the capsule endoscope throughout the stomach. In future large-scale intelligent capsule endoscopy diagnostic and treatment models, the required number of gastroenterologists and the time investment are enormous and unrealistic. Considering the inherent physical structure of the human stomach, summarizing existing examination data and integrating expert experience from physicians to find a universally applicable combination of postural pathways, under the guidance of these pathways, will allow the capsule to capture images that comprehensively cover the anatomical structures within the digestive tract. This will have significant practical value. Summary of the Invention

[0003] The purpose of this invention is to provide a method for finding the optimal combination of body positions for capsule endoscopy. By using state transition probability modeling to achieve trajectory sequence probability modeling, a value evaluation method for different body position combinations is proposed to quickly find the optimal body position combination with greater universality in the probability space.

[0004] The objective of this invention is achieved through the following technical solution: a method for finding the optimal combination of patient positions for capsule endoscopy, comprising the following steps:

[0005] Original problem serialization: Describing body position combinations and anatomical structures using serialization parameters;

[0006] Trajectory sequence probabilistic modeling: Given a combination of body positions S N The anatomical structure sequence observed was Q. N By applying Markov properties and the product of probabilities, the probability of the trajectory sequence is modeled as the cumulative product of state transition probabilities.

[0007] State transition probability modeling: The state transition probability model P(q) was obtained through statistical modeling based on real capsule endoscopy data and probabilistic quantification modeling based on physician experience space, respectively. next |s next ,(q now ) last ,s now ), that is, the last anatomical structure in the observed anatomical structure sequence is (q now ) last The body position is s now At that time, by changing the body position to s next This makes the observed anatomical structure sequence q next The state transition probability;

[0008] Posture combination assessment: By sampling, a set of anatomical structure trajectory sequences that can be sampled under a given posture combination is obtained, and the state transition probability and coverage of a certain anatomical structure sequence are combined to obtain an assessment index that can reflect universality.

[0009] Body position combination optimization: Different body position combinations are quickly obtained through random sampling, and the optimal body position combination sequence is output by calculating the evaluation index.

[0010] The original problem is serialized, and the body position combinations and anatomical structures are described using serialization parameters as follows: Define the selectable set of body positions, S. set ={Left lateral view, Hip-high-head-low left lateral view, ...}; Define the target anatomical structure set, Q set ={esophagus, gastric fundus, cardia…}; Define the sequence of anatomical structures, Q N =(q N ,q N-1 ,…,q0), q t Let q represent the sequence of anatomical structures captured in round t. t =(q t1 ,q t2 ,..,q ti ), where q ti Let q represent the i-th anatomical structure captured in the t-th round, and q ti ∈Q set Define the sequence of body position combinations, SN =(s N ,s N-1 ,…,s0), s t Let s represent the body position chosen in round t, where s t ∈S set The anatomical structure coverage rate is defined as the ratio of the number of observed non-repeating anatomical structures to the total number of anatomical structures. Where: Set(Q) N ) indicates extracting Q N The non-repeating anatomical structure in the set, where length represents the number of elements to retrieve.

[0011] The specific probabilities of the trajectory sequence are as follows:

[0012]

[0013] Where: N is the length of the sequence, (S set ) length P(q) represents the size of the possible body position combinations. t |s t ,(q t-1 ) last ,s t-1 Let q be the state transition probability in round t. t s represents the sequence of anatomical structures observed in round t. t Indicates the body position in round t, (q) t-1 ) last s represents the last anatomical structure in the sequence of anatomical structures observed in round t-1. t-1 This indicates the body position in round t-1.

[0014] When modeling the state transition probability, the last anatomical structure in the observed anatomical structure sequence is (q now ) last The body position is s now At that time, by changing the body position to s next This makes the observed anatomical structure sequence q next The state transition probability is calculated using a statistical modeling method based on real capsule endoscopy data. This method involves target data truncation, trajectory serialization, and state transition frequency statistics to obtain the state transition probability. The specific calculation formula is as follows:

[0015]

[0016] Where, num i For the i-th state transition in the data ((q) now ) last ,s now ) i →(qnext ,s next ) i The frequency of occurrence, NUM is the frequency of all state transitions in the data;

[0017] A probabilistic quantification modeling method based on physician experience space, by labeling the likelihood levels of physician experience tables and combining two types of experiential anatomical structures and probability quantification, can achieve the ability to model any different state transition probabilities P(q). next |s next ,(q now ) last ,s now The modeling and calculation of ) are illustrated in the following empirical table:

[0018] Current position Current anatomical structure Next position Next anatomical structure Possibility of arrival probability level Left side stomach fundus Hips high, head low, left side position cardia 1 3 …… …… …… …… …… …… .

[0019] The assessment of the body position combination yields evaluation indicators that can reflect universality, specifically obtained through sampling of a given body position combination. The set of anatomical structure trajectory sequences that can be sampled and obtained. Combined with a certain anatomical structure sequence probability and coverage The specific formula is as follows:

[0020]

[0021] Among them G length The size of the set of all trajectory sequences. Indicates normalization. It can be calculated by combining the trajectory sequence probability formula with the state transition probability. The larger the value, the better the combination of body positions, and the greater the possibility of covering a more comprehensive range of anatomical structures.

[0022] The specific steps for optimizing the body position combination are as follows:

[0023] S1: State transition probability modeling. The state transition probability is obtained by combining statistical modeling methods based on real capsule endoscopy data and probability quantification methods based on the doctor's experience space according to certain weights. The weights here are adjusted according to the magnitude of the real data.

[0024] S2: Perform random sampling based on the state transition probability and store the generated random trajectory data into the random sampling trajectory library;

[0025] S3: Calculate the coverage rate of each trajectory in the random sampling trajectory library, and store the body position combinations corresponding to the random sampling trajectories with a coverage rate of 100% into the deduplicated candidate body position combination library.

[0026] S4: For any combination of body positions in the pool of candidate body position combinations By traversing the state transitions with non-zero probabilities, all unique trajectory sequences are obtained. Then, using a trajectory sequence probabilistic modeling method, the probability value of each trajectory sequence is calculated. And store it in the Body Position - All Trajectory Library;

[0027] S5: By calculating any combination of body positions Evaluation value And based on the evaluation value The different body position combinations are sorted in descending order by size, and the sorting results are stored in the body position combination evaluation library.

[0028] S6: Select the top-ranked optimal body position combination from the body position combination assessment library and provide it to the endoscopist, providing the doctor with the optimal and universally applicable body position combination path.

[0029] The entire process is handled within a parallel or distributed framework.

[0030] Simultaneously, a capsule endoscopy optimal position combination optimization system based on state transition probability modeling is provided, including an original problem serialization module, a trajectory sequence probability modeling module, a state transition probability modeling module, a position combination evaluation module, and a position combination optimization module.

[0031] The original problem serialization module describes the body position combination and anatomical structure using serialization parameters;

[0032] The trajectory sequence probability modeling module provides a combination of body positions S. N The anatomical structure sequence observed was Q. N By applying Markov properties and the product of probabilities, the probability of the trajectory sequence is modeled as the cumulative product of state transition probabilities.

[0033] The state transition probability modeling module obtains the state transition probability model through statistical modeling based on real capsule endoscopy data and probabilistic quantification modeling based on physician experience space, respectively.

[0034] The body position combination assessment module obtains a set of anatomical structure trajectory sequences that can be sampled under a given body position combination by sampling, and combines the state transition probability and coverage of a certain anatomical structure sequence to obtain an assessment index that can reflect universality.

[0035] The body position combination optimization module quickly obtains different body position combinations through random sampling, and outputs the optimal body position combination sequence by calculating the evaluation index.

[0036] In addition, the present invention also provides a capsule endoscopy optimal position combination optimization system based on state transition probability modeling, including the above-mentioned original problem serialization module, trajectory sequence probability modeling module, state transition probability modeling module, position combination evaluation modeling module, and position combination optimization module.

[0037] On the other hand, the present invention can also provide a computer device, including a processor and a memory, wherein the memory is used to store a computer executable program, the processor reads the computer executable program from the memory and executes it, and the processor can realize the capsule endoscopy optimal position combination optimization method based on state transition probability modeling described in the present invention when executing the executable program.

[0038] A computer-readable storage medium storing a computer program, which, when executed by a processor, enables the optimization method for capsule endoscopy optimal position combination based on state transition probability modeling as described in this invention.

[0039] Compared with the prior art, the present invention has at least the following beneficial effects:

[0040] This invention proposes a method for finding the optimal body positioning combination for capsule endoscopy. It achieves trajectory sequence probabilistic modeling through state transition probabilistic modeling and proposes a value assessment method for different body positioning combinations, enabling the rapid search for a more universally applicable optimal body positioning combination within the probability space. Specifically, this invention employs two state transition probabilistic modeling methods: a statistical modeling method based on real capsule endoscopy data and a probabilistic quantitative modeling method based on physician experience space. On the one hand, this method can significantly reduce the reliance on the experience of physicians specializing in gastrointestinal tract diameters during capsule endoscopy examinations; on the other hand, the universality of the body positioning path optimized by this method lays a foundation for realizing a future intelligent capsule endoscopy diagnosis and treatment model. Attached Figure Description

[0041] Figure 1 This is a schematic diagram of an implementable process of the present invention.

[0042] Figure 2 This is a schematic diagram of a statistical modeling process based on real data from capsule endoscopy proposed in this invention.

[0043] Figure 3 This is a schematic diagram of a probability quantization modeling process based on the physician's experience space proposed in this invention.

[0044] Figure 4 This is a schematic diagram of the optimal body position combination search process proposed in this invention. Detailed Implementation

[0045] This invention proposes an optimal body positioning method for capsule endoscopy. It constructs a trajectory sequence probability model through state transition probabilistic modeling and proposes a value assessment method for different body positioning combinations, enabling the rapid identification of more universally applicable optimal body positioning combinations within a probability space. Specifically, this invention proposes two state transition probabilistic modeling methods: a statistical modeling method based on real capsule endoscopy data and a probabilistic quantitative modeling method based on physician experience space. On the one hand, this method can significantly reduce the reliance on the experience of physicians specializing in gastrointestinal tract diameters during capsule endoscopy examinations; on the other hand, the universality of the body positioning paths optimized by this method lays a foundation for realizing a future intelligent capsule endoscopy diagnosis and treatment model.

[0046] like Figure 1 As shown, this invention provides a method for finding the optimal combination of patient positions in a capsule endoscopy, comprising the following steps (taking the stomach space as an example):

[0047] A1: Serialization of the original problem

[0048] ① The set of selectable body positions, S set = {Left side position, hips high and head low left side position, left side half-support position, prone position, supine position, right side position, right side half-support position}, a total of 7 body position movements.

[0049] ② Target anatomical structure set, Q set = {Esophagus, fundus, cardia, fundobody junction, upper body of stomach, lower body of stomach, gastric angle, antrum, pyloric canal, duodenum}, a total of 10 anatomical structures.

[0050] ③ Anatomical structure sequence, Q N =(q N ,q N-1 ,…,q0), q t Let q represent the sequence of anatomical structures captured in round t. t =(q t1 ,q t2 ,..,q ti ), where q ti Let q represent the i-th anatomical structure captured in the t-th round, and q ti ∈Q set .

[0051] ④ Postural combination sequence, S N =(s N ,s N-1 ,…,s0), s t Let s represent the body position chosen in round t, where s t ∈S set .

[0052] ⑤ Anatomical structure coverage, defined as the ratio of the number of observed non-repeating anatomical structures to the total number of anatomical structures, i.e. Where: Set(Q) N ) indicates extracting Q N The non-repeating anatomical structure in the set, where length represents the number of elements to retrieve.

[0053] A2: Probabilistic Modeling of Trajectory Sequences

[0054] B1: Denote the observed anatomical structure sequence Q. N The probability is P(Q) N ), then in the body position combination sequence S N From the conditional probability formula, we have:

[0055] P(Q N )=P(Q N |S N )·P(S N (1)

[0056] B2: First consider P(Q) in equation (1) N |S N (The rest of the text is missing.)

[0057] P(Q N |S N )=P(q N ,q N-1 ,…,q0|s N ,s N-1 ,…,s0) (2)

[0058] Because of s n and q n They always appear in pairs, so equation (2) can be written as

[0059] P(Q N |S N )=P(q N ,s N , q N-1 ,s N-1 ,…,q0,s0) (3)

[0060] Expanding equation (3) further using the conditional probability formula, we have:

[0061] P(Q N |S N )=P(q N ,s N ,q N-1 ,s N-1 ,…,q0,s0)

[0062] =P(q) N ,sN |q N-1 ,s N-1 ,…,q0,s0)·P(q N-1 ,s N-1 ,…,q0,s0)

[0063] =P(q) N ,s N |q N-1 ,s N-1 ,…,q0,s0)·P(q N-1 ,s N-1 |q N-2 ,s N-2 …,q0,s0)

[0064] P(q N-2 ,s N-2 …,q0,s0)

[0065] Further simplification yields:

[0066]

[0067] Since the initial state q0,s0 of the test subject swallowing the capsule is generally fixed as (s0 = left side, q0 = esophagus), that is, the test subject swallows the capsule while maintaining the [left side position], and the observed initial anatomical structure is the [esophagus], therefore, P(q0,s0) = 1 in equation (4), that is:

[0068]

[0069] B3: Continue to solve for P(q) in equation (5) t ,s t |q t-1 ,s t-1 ,…,q0,s0), expanded by conditional probability:

[0070] P(qt,s t |q t-1 ,s t-1 ,…,q0,s0)=P(q t |s t ,q t-1 ,s t-1 ,…,q0,s0)·P(s t |qt -1 ,s t-1 ,…,q0,s0) (6)

[0071] Where: P(q) t |s t ,q t-1 ,s t-1,…,q0,s0), denoted as Equation ①, represents the position s selected in round t under the state of round t-1. t At that time, the observed anatomical structure sequence was q. t The probability of P(s) t |q t-1 ,s t-1 ,…,q0,s0), denoted as Equation ②, represents the position s chosen in round t under the state of round t-1. t The probability of.

[0072] C1: First consider equation (1), and consider the Markov property, that is, in round t, the body position is s t At that time, the observed anatomical structure was q. t The probability depends only on the body position s selected in round t-1. t-1 The observed anatomical structure sequence q t-1 If the time is related to t-2 or earlier, then equation (1) can be transformed into:

[0073] P(q t |s t ,q t-1 ,s t-1 ,…,q0,s0)=P(q t |s t ,q t-1 ,s t-1 (7)

[0074] In equation (7), P(q) t |s t ,q t-1 ,s t-1 ) indicates that the body position selected in round t-1 is s. t-1 The observed anatomical structure sequence is q t-1 In round t, the body position selected is s. t The observed anatomical structure sequence is q t The probability of.

[0075] C2: Next, consider equation (2), which is transformed into a random probability problem, that is, the body position selected each time is from the set of available body positions S. set Randomly selected from the middle, equation (2) can be further transformed into:

[0076]

[0077] Equation (8) is obviously a constant value, the size of which is determined by the set of possible body positions S. set The size is determined by.

[0078] B4: Continuing to consider equation (7), we again utilize the Markov property, that is, in round t, the body position is s tAt that time, the observed anatomical structure was q. t The probability depends only on the body position s selected in round t-1. t-1 and the observed anatomical structure sequence qt -1 The observed stable anatomical structures are related to the final anatomical structures observed during the process, but not to other anatomical structures photographed throughout the process, which can be further transformed into:

[0079] P(q t |s t ,q t-1 ,s t-1 )=P(q t |s t ,(q t-1 ) last ,s t-1 (9)

[0080] Where: P(q) t |s t ,(q t-1 ) last ,s t-1 ) indicates that the body position selected in round t-1 is s. t-1 The last stable anatomical structure in the observed anatomical sequence is (q t-1 ) last In round t, the body position selected is s. t The observed anatomical structure sequence is q t The probability is given, and this probability is defined as the state transition probability in round t.

[0081] B5: Integrated formulas (5)-(9) can be used to obtain the action combination sequence S N Below, record the observed anatomical structure sequence Q. N The probability P(Q) N |S N )as follows:

[0082]

[0083] Where: P(q) t |s t ,(q t-1 ) last ,s t-1 Let t be the state transition probability in round t.

[0084] A3: State transition probability modeling

[0085] Specifically, regarding the state transition probability P(q) mentioned in B5 of A2... t |s t ,(q t-1 ) last ,s t-1For modeling, this invention provides two methods: one is a statistical modeling method based on real data from capsule endoscopy, and the other is a probabilistic quantitative modeling method based on the physician's experience space.

[0086] C1: Statistical modeling method based on real data from capsule endoscopy

[0087] The statistical modeling method based on real capsule endoscopy data yields results entirely derived from actual data. The accuracy of the modeling increases with the accumulation of real data over time. The entire process is as follows: Figure 2 As shown.

[0088] 1) Target Data Truncation: Real patient test data typically includes the entire process of the capsule endoscope from insertion into the body to its removal, making the data redundant for modeling. Taking the stomach space as an example, the original data is truncated based on the start and end times of the desired target area.

[0089] 2) Trajectory serialization: The truncated data is processed into trajectory data composed of time series. Specifically, the format of the i-th trajectory data is as follows:

[0090] QS i =[(q N ,s N ),(q N-1 ,s N-1 ),…,(q0,s0)]

[0091] Each sequence consists of a pair (q) t ,s t The structure consists of q, corresponding to the state at round t, where q t s represents the sequence of anatomical structures observed in round t. t This indicates the body position in round t.

[0092] 3) State transition modeling: Define a state transition as an event that changes from the state of the previous round to the state of the current round. For example, define the i-th state transition as:

[0093] T i =(q now ,s now ) i →(q next ,s next ) i

[0094] The specific meaning is: in the detected anatomical structure sequence q now The body position is s now At that time, by changing the body position to s next This makes the observed anatomical structure sequence q next .

[0095] From the Markov property of the derivation process of formula (9), we know that we only need to focus on (q) now ) last That is, the last one in the currently observed anatomical structure sequence, so the above state transition can be modeled as:

[0096] T i =((q) now ) last ,s now ) i →(q next ,s next ) i

[0097] 4) Calculate the state transition probability: Calculate the probability of the i-th state transition T in the actual data. i frequency n i Given NUM, the total number of occurrences, we have:

[0098]

[0099] In formula (10), P(q) t |s t ,(q now ) last ,s t-1 The value of ) can be found by looking up the corresponding P(T) i Thus, the modeling of the first state transition probability statistical method is completed.

[0100] C2: A probabilistic quantization modeling method based on physician experience space

[0101] The probabilistic quantification modeling method based on physician experience space transforms the rich experience of endoscopists into specific probabilities when real data is scarce, thereby achieving the modeling of state transition probabilities. The entire process is as follows: Figure 3 As shown.

[0102] 1) Doctor's Experience Table: A table designed for professional doctors. Doctors mark the following based on their experience: under the current body position and the currently observed anatomical structure, the probability and probability level of reaching the next anatomical structure through the next body position movement. Among them: the probability of reaching (0: impossible, 1: possible) and the probability level (1-4 correspond to different levels).

[0103] Table 1: Examples of Empirical Representations

[0104] Current position Current anatomical structure Next position Next anatomical structure Possibility of arrival probability level Left side stomach fundus Hips high, head low, left side position cardia 1 3 …… …… …… …… …… ……

[0105] 2) Two empirical sequences: Based on the physician's experience and according to the probability level of 1-4, the annotations are transformed into two observable anatomical structure sequences: 1-2-1-2-1-2...1 indicates that the sequence goes back and forth between the anatomical structures labeled 1-2, and finally stabilizes at the anatomical structure labeled 1; 1-2-1-2-3-4-3-4...3 indicates that the sequence first goes back and forth between the anatomical structures labeled 1-2, then transitions to the anatomical structures labeled 3-4, and finally stabilizes at the anatomical structure labeled 3.

[0106] 3) Quantification of probability: Based on the doctor's experience, the different probability levels 1-4 are converted into specific probabilities. Specifically: 1 corresponds to a probability of 1.0, 2 corresponds to a probability of 0.95, 3 corresponds to a probability of 0.5, and 4 corresponds to a probability of 0.45. The quantified probabilities here can be dynamically optimized as the doctor's experience is improved.

[0107] 4) Based on the experience table marked by doctors, combined with empirical sequences and quantified probabilities, realize the state transition probability P(q) for any different state. next |s next ,(q now ) last ,s now Modeling and rapid computation of ).

[0108] A4: Postural Combination Assessment Modeling

[0109] Given any combination of body positions S can be obtained based on the sampling method i The following is a collection of all different anatomical structure sequences. Let G be the number of elements in set G. length , of which elements For any observable anatomical structure sequence.

[0110] The value assessment of a combination of body positions is defined as follows:

[0111]

[0112] in It can be calculated by combining formula (10) in A2 with the state transition probability in A3. This indicates the coverage of the anatomical structure sequence. This indicates normalization.

[0113] in, The larger the value, the better the combination of body positions, and the greater the possibility of covering a more comprehensive range of anatomical structures.

[0114] A5: Optimal Body Position Combination Optimization

[0115] refer to Figure 4 The optimal body position combination optimization method proposed in this invention has the following specific implementation steps:

[0116] S1: State transition probability modeling, specifically constructed based on the two modeling methods described in A3, combining the probabilities obtained from the two methods according to the set weights. The weights here can be adjusted according to the magnitude of the real data.

[0117] S2: Randomly sample based on the state transition probability to generate M randomly sampled trajectories, each trajectory corresponding to (q N ,s N ,q N-1 ,s N-1 The sequence of ,…,q0,s0) is stored in the random sampling trajectory library.

[0118] S3: Calculate the coverage rate of each trajectory in the random trajectory library, and store the corresponding body position combinations of trajectories whose coverage rate meets the condition (such as coverage rate reaching 100%) into the deduplicated candidate body position combination library.

[0119] S4: Any combination of body positions from the pool of selected body position combinations By traversing the state transitions with non-zero probabilities, all unique trajectory sequences are obtained. Then, using the sequence probability modeling method described in A2, the probability value of each trajectory sequence is calculated. And store it in the Body Position - All Trajectory Library;

[0120] S5: By calculating any combination of body positions Evaluation value And according to The different body position combinations are sorted in descending order by size, and the results are stored in the body position combination assessment library;

[0121] S6: Select the top-ranked optimal body position combination from the body position combination assessment library and provide it to the endoscopist, providing the doctor with the optimal and universally applicable body position combination path.

[0122] Specifically, after obtaining the optimal combination of patient positions, doctors can use their experience or practical verification to feed the verified data back into the statistical modeling based on real capsule endoscopy data, and feed the updated experience back into the probabilistic quantification modeling based on the doctor's experience space. This optimizes the modeling process of iterative state transition probabilities from two aspects. The entire process is dynamically updatable.

[0123] In particular, many processes described above are naturally suited to be processed in a parallel or distributed framework, such as the sampling process in S2, the coverage calculation in S3, the trajectory probability calculation in S4, and the position combination evaluation calculation in S5. This method can achieve fast computation in a parallelized environment.

[0124] Specific embodiments of the present invention are as follows:

[0125] The following example illustrates the effectiveness of the optimal body position combination method. All experiments were implemented in a single-machine environment using Python, with 128GB of available memory, 24 cores, and 4TB of available storage. Given a sequence length N=13, the algorithm automatically combines and ranks body position combinations (only the two optimal combinations are shown here), comparing the results with the empirical body position combinations currently used by doctors. Generally speaking: for a body position combination, the smaller the number of possible trajectories, the higher the certainty of possible internal paths, and the better the combination; the higher the average trajectory coverage, the better the combination; and the higher the value assessment value, the better the combination.

[0126] Table 2: Numerical Comparison Results

[0127]

[0128]

[0129] As can be seen from the results, the two body position combinations identified and displayed by the algorithm are superior to the body position combinations used by doctors based on experience in all three evaluation indicators. The body position combinations found are more universal and more suitable for the large-scale intelligent diagnosis and treatment mode of capsule endoscopy in the future.

[0130] Optionally, the present invention also provides a computer device, including a processor and a memory, wherein the memory is used to store a computer executable program, the processor reads part or all of the computer executable program from the memory and executes it, and the processor can realize the capsule endoscopy optimal position combination optimization method described in the present invention when executing part or all of the executable program.

[0131] On the other hand, the present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, can realize the capsule endoscopy optimal position combination optimization method described in the present invention.

[0132] A program that can be written in a computer programming language to perform the methods described in this application can be used. The computer program can be in the form of source code, object code, executable file or some intermediate form. The computer programming language can be C++, Java, Fortran, C# or Python.

[0133] The computer device may be a laptop, tablet, desktop computer, mobile phone, or workstation.

[0134] The processor can be a central processing unit (CPU), a digital signal processor (DSP), an application-specific integrated circuit (ASIC), or an off-the-shelf programmable gate array (FPGA).

[0135] The memory described in this invention can be an internal storage unit of a laptop, tablet, desktop computer, mobile phone, or workstation, such as memory or hard disk; or it can be an external storage unit, such as a portable hard disk or flash memory card.

[0136] Computer-readable storage media can include computer storage media and communication media. Computer storage media include volatile and non-volatile, removable and non-removable media implemented using any method or technology for storing information such as computer-readable instructions, data structures, program modules, or other data. Computer-readable storage media can include: read-only memory (ROM), random access memory (RAM), solid-state drives (SSDs), or optical discs, etc. Random access memory can include resistive random access memory (ReRAM).

[0137] The specific examples described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific examples of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for finding the optimal combination of patient positions for capsule endoscopy, characterized in that, It includes the following steps: Original problem serialization: Describing body position combinations and anatomical structures using serialization parameters; Trajectory sequence probabilistic modeling: given a combination of body positions The anatomical structure sequence observed under these circumstances is By applying Markov properties and the product of probabilities, the probability of the trajectory sequence is modeled as the cumulative product of state transition probabilities. Get in the action combination sequence Below, record the observed anatomical structure sequence. probability as follows: in: for Round state transition probability, The size of the optional body position combination, State transition probability modeling: State transition probability models were obtained through statistical modeling methods based on real capsule endoscopy data and probabilistic quantitative modeling methods based on physician experience space, respectively. That is, the last anatomical structure in the observed sequence of anatomical structures is The body position is At that time, by changing body position This makes the observed anatomical structure sequence as The state transition probability; Probabilistic quantification modeling methods based on physician experience space include: transforming the rich experience of endoscopists into specific probabilities, and modeling state transition probabilities. 1) Doctor's Experience Table: A table designed for professional doctors. Doctors mark the following based on their experience: the probability and probability level of reaching the next anatomical structure through the next body position and the currently observed anatomical structure. Among them, the probability of reaching the next anatomical structure is 0, which means impossible and 1 means possible. Regarding the probability level, 1-4 correspond to different levels. 2) Two empirical sequences: Based on the physician's experience and according to the probability level of 1-4, the annotations are transformed into two observable anatomical structure sequences: 1-2-1-2-1-2...1 indicates that the sequence goes back and forth between the anatomical structures labeled 1-2, and finally stabilizes at the anatomical structure labeled 1; 1-2-1-2-3-4-3-4...3 indicates that the sequence first goes back and forth between the anatomical structures labeled 1-2, then transitions to the anatomical structures labeled 3-4, and finally stabilizes at the anatomical structure labeled 3. 3) Quantification of probability: Based on the doctor's experience, the different probability levels 1-4 are converted into specific probabilities. Specifically: 1 corresponds to a probability of 1.0, 2 corresponds to a probability of 0.95, 3 corresponds to a probability of 0.5, and 4 corresponds to a probability of 0.

45. The quantified probabilities are dynamically optimized based on the improvement of the doctor's experience. 4) Based on the experience table marked by doctors, combined with empirical sequences and quantified probabilities, the probability of state transitions for any different states can be determined. Modeling and computation; Posture combination assessment: By sampling, a set of anatomical structure trajectory sequences that can be sampled under a given posture combination is obtained, and the state transition probability and coverage of a certain anatomical structure sequence are combined to obtain an assessment index that can reflect universality. Body position combination optimization: Different body position combinations are quickly obtained through random sampling, and the optimal body position combination sequence is output by calculating the evaluation index.

2. The method for finding the optimal combination of patient positions for capsule endoscopy according to claim 1, characterized in that, The original problem is serialized, and the body position combinations and anatomical structures are described using serialization parameters as follows: Define the set of selectable body positions, ={Left lateral view, Hip-high-head-low left lateral view, ...}; Defines the set of target anatomical structures. ={esophagus, fundus, cardia…}; Defines the sequence of anatomical structures. , Indicates the first The sequence of anatomical structures captured in each round is defined as follows: ,in This indicates the number of images captured in round t. An anatomical structure, and Define a sequence of body position combinations. ), Indicates the first The positions selected in each round, among which The anatomical structure coverage rate is defined as the ratio of the number of observed non-repeating anatomical structures to the total number of anatomical structures. = ,in: Indicate extraction The non-repeating anatomical structure in the set, where length represents the number of elements to retrieve.

3. The method for finding the optimal combination of patient positions for capsule endoscopy according to claim 1, characterized in that, When modeling the state transition probability, the last anatomical structure in the observed anatomical structure sequence is... The body position is At that time, by changing body position This makes the observed anatomical structure sequence as The state transition probability; A statistical modeling method based on real capsule endoscopy data is used to obtain the state transition probability by truncating the data, serializing the trajectory, and statistically analyzing the state transition frequency. The specific calculation formula is as follows: in, For the data State transition Frequency of occurrence This represents the frequency of all state transitions occurring in the data.

4. The method for finding the optimal combination of patient positions for capsule endoscopy according to claim 1, characterized in that, The assessment of the body position combination yields evaluation indicators that can reflect universality, specifically obtained through sampling of a given body position combination. The set of anatomical structure trajectory sequences that can be sampled and obtained. Combined with a certain anatomical structure sequence probability and coverage The specific formula is as follows: in The size of the set of all trajectory sequences. Indicates normalization, It can be calculated by combining the trajectory sequence probability formula with the state transition probability. The larger the value, the better the combination of body positions, and the greater the possibility of covering a more comprehensive range of anatomical structures.

5. The method for finding the optimal combination of patient positions for capsule endoscopy according to claim 1, characterized in that, The specific steps for optimizing the body position combination are as follows: S1: State transition probability modeling. The state transition probability is obtained by combining statistical modeling methods based on real capsule endoscopy data and probability quantification methods based on the doctor's experience space according to certain weights. The weights here are adjusted according to the magnitude of the real data. S2: Perform random sampling based on the state transition probability and store the generated random trajectory data into the random sampling trajectory library; S3: Calculate the coverage rate of each trajectory in the random sampling trajectory library, and store the body position combinations corresponding to the random sampling trajectories with a coverage rate of 100% into the deduplicated candidate body position combination library. S4: For any combination of body positions in the pool of candidate body position combinations By traversing the state transitions with non-zero probabilities, all unique trajectory sequences are obtained. Then, using a trajectory sequence probabilistic modeling method, the probability value of each trajectory sequence is calculated. And store it in the Body Position - All Trajectory Library; S5: By calculating any combination of body positions Evaluation value And based on the evaluation value The different body position combinations are sorted in descending order by size, and the sorting results are stored in the body position combination evaluation library. S6: Select the top-ranked optimal body position combination from the body position combination assessment library and provide it to the endoscopist, providing the doctor with the optimal and universally applicable body position combination path.

6. The method for finding the optimal combination of patient positions for capsule endoscopy according to claim 5, characterized in that, The entire process is handled within a parallel or distributed framework.

7. A capsule endoscopy optimal positioning combination optimization system based on state transition probability modeling, characterized in that, It includes a module for serializing the original problem, a module for probabilistic modeling of the trajectory sequence, a module for probabilistic modeling of the state transition, a module for evaluating body position combinations, and a module for optimizing body position combinations. The original problem serialization module describes the body position combination and anatomical structure using serialization parameters; The trajectory sequence probability modeling module provides a combination of body positions. The anatomical structure sequence observed under these circumstances is By applying Markov properties and the product of probabilities, the probability of the trajectory sequence is modeled as the cumulative product of state transition probabilities. The state transition probability modeling module obtains the state transition probability model through statistical modeling based on real capsule endoscopy data and probabilistic quantification modeling based on physician experience space, respectively. The body position combination assessment module obtains a set of anatomical structure trajectory sequences that can be sampled under a given body position combination by sampling, and combines the state transition probability and coverage of a certain anatomical structure sequence to obtain an assessment index that can reflect universality. The body position combination optimization module quickly obtains different body position combinations through random sampling, and outputs the optimal body position combination sequence by calculating the evaluation index.

8. A computer device, characterized in that, It includes a processor and a memory, the memory being used to store a computer-executable program, the processor reading the computer-executable program from the memory and executing it, and the processor executing the executable program is able to implement the capsule endoscope optimal position combination optimization method according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, A computer-readable storage medium stores a computer program that, when executed by a processor, enables the optimization method for optimal combination of capsule endoscopy positions as described in any one of claims 1 to 6.

Images