CN114334145B - Dynamic identification method for atypical critical patients - Google Patents
Dynamic identification method for atypical critical patients Download PDFInfo
- Publication number
- CN114334145B CN114334145B CN202111481837.9A CN202111481837A CN114334145B CN 114334145 B CN114334145 B CN 114334145B CN 202111481837 A CN202111481837 A CN 202111481837A CN 114334145 B CN114334145 B CN 114334145B
- Authority
- CN
- China
- Prior art keywords
- dynamic
- data sequence
- sign data
- patient
- sequence set
- 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
Images
Classifications
-
- 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
- Measuring And Recording Apparatus For Diagnosis (AREA)
Abstract
The invention relates to a dynamic identification method of atypical critical patients, which comprises the steps of obtaining first K second dynamic sign data sequence sets with minimum approximation distances by calculating approximation distances between a first dynamic sign sequence set of a patient to be tested and second dynamic sign data sequence sets of historical patients to form candidate sets, calculating risk scores of the patient to be tested by the approximation distances between the first dynamic sign sequence set of the patient to be tested and each second dynamic sign sequence set of the candidate sets, and finally judging whether the patient to be tested is the critical patient by comparing the risk scores with a set threshold value. Because the first dynamic sign sequence set corresponds to at least two signs of the patient to be tested one by one, the first dynamic sign sequence set is a dynamic data sequence of the patient to be tested sign data in a time domain, and compared with the original method for manually judging whether the patient to be tested is a critical patient through static data, the method is more accurate and has higher efficiency.
Description
Technical Field
The invention relates to identification, in particular to a dynamic identification method for atypical critical patients.
Background
Aiming at the contradiction between the increase of emergency treatment amount and the limited medical resource allocation, the emergency treatment sub-treatment system can be adopted at present to arrange the diagnosis and treatment priority according to the urgency of the illness state of patients so as to reasonably allocate the medical resources. The triage system has the timeliness characteristic, and triage errors can cause emergency treatment congestion and diagnosis and treatment delay problems of different degrees, and accordingly cause problems of medical resource waste or increased mortality of patients.
The current use of China is a China 4-level diagnosis system determined in the "emergency patient condition grading guidance principle" issued by the guard and commission 2011, wherein endangered and critical patients with a diagnosis of 1-2 level directly enter a rescue room or monitor vital signs, and emergency and non-emergency patients with a diagnosis of 3-4 level need to wait for a diagnosis room to receive a diagnosis. The current triage system is accurate and effective in most cases, but some potentially critical patients are triaged by the electronic triage system to be 3-4 grade, but are sent to the rescue room in a short time. The process of waiting for such patients can be dangerous. The errors of the triage system are common in various triage systems, and potential critical patients are identified through reasonable means, so that the problem of diagnosis and treatment delay in emergency scenes is solved. If the patient can be dynamically evaluated during the waiting period of the patient, the change of the illness state of the atypical critical patient can be found in time, the identification of the atypical critical patient is more beneficial than the single evaluation during the room entering, the international peer also adopts the intermittent measurement mode of critical signs of the emergency room, the data are acquired for a plurality of hours at intervals and then re-rated, but the mode is still static evaluation, and the dynamic development data of the patient are not fully mined.
In summary, the existing triage system mainly has the following two problems:
1. the existing emergency classification usually carries out primary classification according to the data of the patients in the hospital, the data used for classification are vital signs and important complaints of the patients in the hospital, and the data are static evaluation modes, can only reflect the disease state of the patients in the hospital, can not find the change of the illness state of the patients in the waiting time, and also do not consider other factors related to critical occurrence.
2. For patients initially classified as urgent and not urgent, during their waiting, doctors can acquire key signs of the patients at different moments in a discontinuous measurement mode and then re-rate. However, the grading system under the rule system can only evaluate the current state of the patient and still belongs to a static evaluation mode.
Disclosure of Invention
In order to solve the technical problems, the invention aims to provide a dynamic identification method for atypical critical patients based on multivariate time sequence data mining by fully utilizing dynamic sign data of patients in a waiting period.
The invention relates to a dynamic identification method for atypical critical patients, which comprises the following steps:
s1: acquiring a first dynamic sign data sequence set of a patient to be tested, wherein the first dynamic sign data sequence set comprises a plurality of first dynamic sign data sequences corresponding to at least two signs of the patient to be tested in a one-to-one mode;
s2: acquiring N second dynamic sign data sequence sets corresponding to N historical patients, wherein each second dynamic sign data sequence set comprises a plurality of second dynamic sign data sequences corresponding to the corresponding historical patients and signs one by one;
s3: respectively calculating the approximation distance D between the first dynamic sign data sequence set in the step S1 and each second dynamic sign data sequence set in the N second dynamic sign data sequence sets in the step S2;
s4: acquiring first K second dynamic sign data sequence sets with minimum approximation distances in the step S3, and forming candidate sets by the K second dynamic sign data sequence sets with minimum approximation distances, wherein K is smaller than N;
s5: obtaining a risk score of the patient to be tested by the following formula:
wherein D is i A proximity distance between the first dynamic vital sign data sequence set and an ith second dynamic vital sign data sequence set in the candidate set;
wherein c i The values of (2) are as follows:
c when the ith second dynamic vital sign data sequence set in the candidate set is a positive sample i ＝1；
C when the ith second dynamic vital sign data sequence set in the candidate set is a negative sample i ＝-1；
Wherein if the historical patient corresponding to the second dynamic vital sign data sequence set is a critical patient, the second dynamic vital sign data sequence set is a positive sample;
if the historical patient corresponding to the second dynamic vital sign data sequence set is not a critical patient, the second dynamic vital sign data sequence set is a negative sample;
s6: judging whether the patient to be tested is a critical patient or not according to the risk score obtained in the step S5 and the set threshold value:
the risk score is larger than a set threshold, and the patient to be tested is a critical patient;
and if the risk score is less than or equal to the set threshold, the patient to be tested is not a critical patient.
The invention relates to a dynamic identification method of atypical critical patients, which comprises the steps of obtaining first K second dynamic sign data sequence sets with minimum approximation distances by calculating approximation distances between a first dynamic sign sequence set of a patient to be tested and second dynamic sign data sequence sets of historical patients to form candidate sets, calculating risk scores of the patient to be tested by the approximation distances between the first dynamic sign sequence set of the patient to be tested and each second dynamic sign sequence set of the candidate sets, and finally judging whether the patient to be tested is the critical patient by comparing the risk scores with a set threshold value. Because the first dynamic sign sequence set corresponds to at least two signs of the patient to be tested one by one, the first dynamic sign sequence set is a dynamic data sequence of the patient to be tested sign data in a time domain, and compared with the original method for manually judging whether the patient to be tested is a critical patient through static data, the method is more accurate and has higher efficiency.
The approximation distance calculation method in step S3 is as follows:
s31: calculating the approximation distance d between each first dynamic sign data sequence in the first dynamic sign data sequence set and a second dynamic sign data sequence corresponding to the first dynamic sign data sequence in the second dynamic sign data sequence set;
the first dynamic sign data sequence and the second dynamic sign data sequence correspond to each other, and indicate that the types of sign data recorded by the first dynamic sign data sequence are the same as the types of sign data recorded by the second dynamic sign data sequence; the approximation distance d may be a Euclidean distance;
s32: obtaining the approximation distance between the first dynamic sign data sequence set and the second dynamic sign data sequence set by the following formula:
wherein n is the number of sign species, d i Is the approximation distance between the first dynamic sign data sequence corresponding to the ith sign and the second dynamic sign data sequence set;
wherein w is i Attribute weights that are proximity distances corresponding to the i-th sign data.
The method comprises the steps of calculating the approximation distances between each first dynamic sign data sequence in a first dynamic sign data sequence set and the corresponding second dynamic sign data sequence in a second dynamic sign data sequence set, and acquiring approximation values between the first dynamic sign data sequence set and the second dynamic sign data sequence set by combining attribute weights of the approximation distances, so that the calculation of approximation distances between multiple sequence sets is realized. The multiple sequence set values here are the set formed by the corresponding data sequences of the various signs. The attribute weight is set by an operator according to historical data experience or after autonomous learning by a deep learning system.
Furthermore, in the method for dynamically identifying atypical critically ill patients of the present invention, in step S31, dynamic time warping is performed on the two data sequences by DTW algorithm before calculating the approximation distance d between the first dynamic vital sign data sequence and the corresponding second dynamic vital sign data sequence.
And the DTW algorithm is adopted to carry out dynamic time warping on the data sequence, so that the calculated approximation distance is more accurate after the warping.
Furthermore, the method for dynamically identifying atypical critically ill patients further comprises the steps of data segmentation and feature extraction of the data sequence before dynamic time warping.
Aiming at the high-dimensional characteristics of the dynamic physical sign data of the patient, local feature extraction or global feature decomposition is required to be carried out on the data sequence before the classification and identification model is constructed so as to reduce the dimension and capture the related information between the trend change and different dimensions, thereby extracting the effective information of the patient condition change. The strategy and result of the feature representation directly relate to the performance of the subsequent classification recognition model.
For data segmentation, to extract local features, a time sequence data sequence is required to be segmented, and local information is extracted for the segmented sequence.
For feature extraction, the acquisition interval of time sequence data in the waiting period is larger, and in order to ensure the capture of important information such as time sequence morphological change, key points and the like, time sequence representation based on statistical features, symbolization representation in a two-dimensional space and other methods are adopted to process the features of time sequence data, and the change trend and morphological characteristics of the time sequence data are described. The time domain value comprises maximum value, minimum value, peak value, average value, mean square error and the like, and dimensionless characteristics such as peak value factors, pulse factors, waveform factors and the like can be selected.
The above steps of data segmentation and feature extraction are conventional methods and will not be described in detail here.
The foregoing description is merely an overview of the embodiments of the present invention, and is intended to provide a more clear understanding of the technical means of the present invention, as embodied in the present invention, by way of example only.
Drawings
FIG. 1 is a flow chart of the method of dynamic identification of atypical critically ill patients of the present invention;
fig. 2 is a schematic diagram of data segmentation of a sequence.
Detailed Description
The following describes in further detail the embodiments of the present invention with reference to the drawings and examples. The following examples are illustrative of the invention and are not intended to limit the scope of the invention.
Referring to fig. 1 and 2, the dynamic identification method of an atypical critical patient of the present embodiment is described with the heartbeat sign r (heart rate) and the blood pressure sign p (blood pressure) of the patient as the tested signs, and the identification method for a plurality of signs is analogized, and will not be described in detail.
Let a be the first dynamic sign data sequence set of the patient to be tested, b be the second dynamic sign data sequence set of the historical patient b e { b } 1 ，b 2 ，...b N Sequence sets a, b contain sequences corresponding to two signs r and p, respectively, as follows:
a＝{{Sr(1)，S p (1)}，...，{S r (t)，S p (t)}}
b＝{{T r (1)，T p (1)}，...，{T r (t)，T p (t)}}
wherein the sequence set a can be split into two first dynamic sign data sequences a r And a p ，a r ＝{S r (1)，...，S r (t1)}，a p ＝{S p (1)，...，S p (t 2) }, and similarly, splitting the sequence set b into two second dynamic vital data sequences b r And b p ，b r ＝{T r (1)，...，T r (f)}，b p ＝{T p (1)，...，T p (f) After the data self-adapting based segmentation method D and feature extraction F (including statistical features, symbolized representation, dimension and dimensionless features based on sliding window, etc.), a new dynamic sign data sequence is formed:
new first dynamic vital data sequence:
a′ r ＝D(F(a r ))＝{S′ r (1)，...，S′ r (n1)}
a′ p ＝D(F(a p ))＝{S′ p (1)，...，(n2)}
new second dynamic vital data sequence:
b′ r ＝D(F(a r ))＝{T′ r (1)，...，(m1)}
b′ p ＝D(F(a r ))＝{T′ P (1)，...，T′ P (m2)}
calculation of the sequence a' r And b' r The approximation distance of (2) is calculated as follows:
firstly, a curved path P with the smallest accumulated distance is searched by a DTW algorithm best ＝p 1 ，p 2 ...p U P represents a r ，b r A match between points. The method is an optimal path planning problem, and the model obtains an optimal path through continuous iterative optimization. The constraints are as follows:
(1) Boundary of path: p is p 1 ＝(1，1)，p U ＝(n1，m1)
(2) Path monotonicity: given p k ＝(i，j)，p k+1 = (i ', j') with i '. Gtoreq.i, j'. Gtoreq.j-
(3) Continuity of path: given p k ＝(i，j)，p k+1 = (i ', j') with i '. Ltoreq.i+1, j'. Ltoreq.j+1
The recurrence relation is expressed as follows:
DTW(i，j)＝∞ifi＝0，j≠0or i≠0，j＝0
DTW(i，j)＝0ifi＝j＝0
DTW＝γ(a ri ，b rj )+min{DTW(i-1，j)，DTW(i-1，j-1)，DTW(i，j-1)}
wherein gamma (a) ri ，b rj ) Representation a ri And b rj Where i.e.1, n1]，j∈[1，m1]：
γ(a ri ，b rj )＝||a ri -b rj || w
w is usually 2, where γ (a ri ，b rj ) I.e. the euclidean distance.
Searching a matching point according to the DTW model to obtain a r ，b r Distance between
d r ＝DTW(a r ，b r )
D can be obtained in the same way as above P 。
Analyzing the contribution rate of heartbeat r and blood pressure p to the attribute of the patient according to the historical data, or obtaining the attribute weight w corresponding to r and p by calculating a correlation matrix r ，w p
The proximity distance of the first and second sets of dynamic vital data sequences a and b may be expressed as,
setting k candidates according to the k-means algorithm, the method can obtain
candidate＝k-means(a，B)＝{b 1 ，b 2 ，...，b k }
D (a, b) was calculated as described above 1 )，....D(a，b k )；
The risk score for calculating the first set of vital data sequences a is expressed as:
setting a threshold value thred according to actual conditions: score > wired, and early warning is performed.
The foregoing is merely a preferred embodiment of the present invention for assisting a person skilled in the art in realizing the corresponding technical solution, and is not intended to limit the scope of the present invention, which is defined by the appended claims. It should be noted that, on the basis of the technical solution of the present invention, several improvements and modifications equivalent thereto can be made by those skilled in the art, and these improvements and modifications should also be regarded as the protection scope of the present invention. Meanwhile, it should be understood that, although the present disclosure describes the above embodiments, not every embodiment contains only one independent technical solution, and the description is merely for clarity, and those skilled in the art should consider the disclosure as a whole, and the technical solutions of the embodiments may be combined appropriately to form other embodiments that can be understood by those skilled in the art.
Claims (3)
1. A method for dynamically identifying atypical critically ill patients comprising the steps of:
s1: acquiring a first dynamic sign data sequence set of a patient to be tested, wherein the first dynamic sign data sequence set comprises a plurality of first dynamic sign data sequences corresponding to at least two signs of the patient to be tested in a one-to-one mode;
s2: acquiring N second dynamic sign data sequence sets corresponding to N historical patients, wherein each second dynamic sign data sequence set comprises a plurality of second dynamic sign data sequences corresponding to the corresponding historical patients and signs one by one;
s3: respectively calculating the approximation distance D between the first dynamic sign data sequence set in the step S1 and each second dynamic sign data sequence set in the N second dynamic sign data sequence sets in the step S2;
s4: acquiring first K second dynamic sign data sequence sets with minimum approximation distances in the step S3, and forming candidate sets by the K second dynamic sign data sequence sets with minimum approximation distances, wherein K is smaller than N;
s5: obtaining a risk score of the patient to be tested by the following formula:
wherein D is i A proximity distance between the first dynamic vital sign data sequence set and an ith second dynamic vital sign data sequence set in the candidate set;
wherein c i The values of (2) are as follows:
c when the ith second dynamic vital sign data sequence set in the candidate set is a positive sample i ＝1；
C when the ith second dynamic vital sign data sequence set in the candidate set is a negative sample i ＝-1；
Wherein if the historical patient corresponding to the second dynamic vital sign data sequence set is a critical patient, the second dynamic vital sign data sequence set is a positive sample;
if the historical patient corresponding to the second dynamic vital sign data sequence set is not a critical patient, the second dynamic vital sign data sequence set is a negative sample;
s6: judging whether the patient to be tested is a critical patient or not according to the risk score obtained in the step S5 and the set threshold value:
the risk score is larger than a set threshold, and the patient to be tested is a critical patient;
the risk score is less than or equal to a set threshold, and the patient to be tested is not a critical patient;
the approximation distance calculation method in step S3 is as follows:
s31: calculating the approximation distance d between each first dynamic sign data sequence in the first dynamic sign data sequence set and a second dynamic sign data sequence corresponding to the first dynamic sign data sequence in the second dynamic sign data sequence set;
s32: obtaining the approximation distance between the first dynamic sign data sequence set and the second dynamic sign data sequence set by the following formula:
wherein n is the number of sign species, d i Is the approximation distance between the first dynamic sign data sequence corresponding to the ith sign and the second dynamic sign data sequence set;
wherein w is i Attribute weights that are proximity distances corresponding to the i-th sign data.
2. The method of claim 1, wherein the step S31 is performed with dynamic time warping of the first dynamic vital sign data sequence and the corresponding second dynamic vital sign data sequence by DTW algorithm before calculating the approximation distance d between the two data sequences.
3. The method for dynamic identification of atypical critically ill patients according to claim 2, wherein: the method further comprises the steps of data segmentation and feature extraction of the data sequence before dynamic time warping.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111481837.9A CN114334145B (en) | 2021-12-06 | 2021-12-06 | Dynamic identification method for atypical critical patients |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111481837.9A CN114334145B (en) | 2021-12-06 | 2021-12-06 | Dynamic identification method for atypical critical patients |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114334145A CN114334145A (en) | 2022-04-12 |
| CN114334145B true CN114334145B (en) | 2023-06-30 |
Family
ID=81049582
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111481837.9A Active CN114334145B (en) | 2021-12-06 | 2021-12-06 | Dynamic identification method for atypical critical patients |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114334145B (en) |
Family Cites Families (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20160135706A1 (en) * | 2014-11-14 | 2016-05-19 | Zoll Medical Corporation | Medical Premonitory Event Estimation |
| CN104915561B (en) * | 2015-06-11 | 2018-10-16 | 万达信息股份有限公司 | Genius morbi intelligent Matching method |
| GB201619902D0 (en) * | 2016-11-24 | 2017-01-11 | Univ Oxford Innovation Ltd | Patient status monitor and method of monitoring patient status |
| CN108962389A (en) * | 2018-06-21 | 2018-12-07 | 上海掌门科技有限公司 | Method and system for indicating risk |
| US20210350933A1 (en) * | 2018-09-18 | 2021-11-11 | Koninklijke Philips N.V. | General and personal patient risk prediction |
| US11443853B2 (en) * | 2018-12-14 | 2022-09-13 | Prescient Healthcare Consulting, LLC | Dynamic rolling seventy of illness score for a critically ill patient |
| CN112071425B (en) * | 2020-09-04 | 2022-10-21 | 平安科技（深圳）有限公司 | Data processing method and device, computer equipment and storage medium |
| CN112992350A (en) * | 2021-02-23 | 2021-06-18 | 联仁健康医疗大数据科技股份有限公司 | Physical sign data monitoring method and device, electronic terminal and storage medium |
| CN113327680A (en) * | 2021-07-15 | 2021-08-31 | 厦门市易联众易惠科技有限公司 | Disease risk early warning method, device, equipment and storage medium |
-
2021
- 2021-12-06 CN CN202111481837.9A patent/CN114334145B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN114334145A (en) | 2022-04-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110491465B (en) | Disease classification coding method, system, device and medium based on deep learning | |
| JP6941123B2 (en) | Cell annotation method and annotation system using adaptive additional learning | |
| CN113033622B (en) | Training method, device, equipment and storage medium for cross-modal retrieval model | |
| EP3905126A2 (en) | Image clustering method and apparatus | |
| US20120173576A1 (en) | Patient identification | |
| CN112204671A (en) | Personalized device recommendation for active health monitoring and management | |
| CN110222560B (en) | Text person searching method embedded with similarity loss function | |
| JP7106742B2 (en) | Face recognition method, device, electronic device and computer non-volatile readable storage medium | |
| US20190258935A1 (en) | Computer-readable recording medium, learning method, and learning apparatus | |
| CN110826525A (en) | Face recognition method and system | |
| Muhathir et al. | Analysis K-Nearest Neighbors (KNN) in Identifying Tuberculosis Disease (Tb) By Utilizing Hog Feature Extraction | |
| CN102521534B (en) | Intrusion detection method based on crude entropy property reduction | |
| WO2019167784A1 (en) | Position specifying device, position specifying method, and computer program | |
| EP3435264B1 (en) | Method and system for identification and classification of operational taxonomic units in a metagenomic sample | |
| CN113827240B (en) | Emotion classification method, training device and training equipment for emotion classification model | |
| Karnes et al. | Adaptive Few-Shot Learning PoC Ultrasound COVID-19 Diagnostic System | |
| CN114334145B (en) | Dynamic identification method for atypical critical patients | |
| CN113742348A (en) | Patient data matching method in CDR system, main index establishing method and device | |
| CN107085655B (en) | Traditional Chinese medicine data processing method and system based on attribute constraint concept lattice | |
| CN117422074A (en) | Method, device, equipment and medium for standardizing clinical information text | |
| Fournier et al. | Acute hypotension episode prediction using information divergence for feature selection, and non-parametric methods for classification | |
| CN116306909A (en) | Method for realizing model training, computer storage medium and terminal | |
| CN112861962B (en) | Sample processing method, device, electronic equipment and storage medium | |
| EP4220495A1 (en) | Task learning system and method, and related device | |
| CN112732690B (en) | Stabilizing system and method for chronic disease detection and risk assessment |
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: 20230714 Address after: 100005 Wangfujing, Beijing Dongcheng District Wangfujing Shuai Fu Garden No. 1 Patentee after: PEKING UNION MEDICAL COLLEGE Hospital Patentee after: TSINGHUA University Address before: 100005 Wangfujing, Beijing Dongcheng District Wangfujing Shuai Fu Garden No. 1 Patentee before: PEKING UNION MEDICAL COLLEGE Hospital |
|
| TR01 | Transfer of patent right |