Professor Chris Williams from the University of Edinburgh will discuss research aimed at improving ICU patient care using condition monitoring. This is often impeded by the presence of artifacts in the data; maintaining blood pressure in critically ill patients is a key management goal and yet it is the physiological variable most prone to error.
Using data from vital signs data collected from the Neuro ICU at the Southern General Hospital, Glasgow, Chris will describe work on using the the Factorial Switching Linear Dynamical System (FSLDS) and the Discriminative Switching Linear Dynamical System (DSLDS) for the detection, removal and cleaning of artifacts.
Chris will also present a non-linear dynamical system for modelling the effect of drug infusions on the vital signs of patients admitted in Intensive Care Units (ICUs). More specifically the work is interested in modelling the effect of a widely used anaesthetic drug called Propofolon a patient's monitored depth of anaesthesia and haemodynamics. The approach is compared with one from the Pharmacokinetics/Pharmacodynamics (PK/PD) literature.
Joint work with: Konstantinos Georgatzis, Chris Hawthorne, Partha Lal, Martin Shaw, Ian Piper.
Meetup was here: https://www.meetup.com/London-Bayesian-network-Meetup-Group/events/242965982/
The QRS changes during ischemia have historically been more difficult to parameterize
and have not come into clinical practice. This paper presented a new approach to analyze ischemia
by time parameter extraction of RS-Segment of the QRS complex. The proposed methodology
mainly focused on two prominent areas; first: detection of R and S points via Fast Fourier Transform
(FFT) based windowing & thresholding techniques with a sliding edge method. Second: calculating
the RS-Duration. The performances of the detection methods are validated and RS-Duration is
evaluated with the Fantasia database (Fantasia) for 20 healthy subjects & Long-Term ST Database
(LTSTDB) for 80 ischemic patients. The RS-Segment detection sensitivity (Se) and specificity (Sp)
are calculated 100% for Fantasia Database, whereas sensitivity (Se) is 91.6% and specificity (Sp) is
974% for LTSTDB.
Microvascular & Functional Ultrasound Imaging: Insights into Stroke and Neuro...InsideScientific
Professors Franck Lebrin and Denis Vivien discuss in vivo molecular and functional imaging, including ultrasound-based markers, and their application to the study and treatment of neurological disorders such as cerebral hereditary angiopathies and stroke.
Early vascular dysfunction is increasingly recognized as the underlying cause of many neurological diseases. The development of drugs targeting vascular damage at its earliest stages could therefore pave the road towards the treatment of neurological disorders. However, to be effective, this therapeutic approach will require the identification of early markers of vascular injury.
In the first portion of the webinar, Prof. Lebrin discusses his research on ultrasound markers of early vascular dysfunction in cerebral hereditary angiopathies (CHA) and the testing of novel therapeutic agents that restore vascular function. Albeit rare, blood vessels from CHA exhibit the salient features of other neurological diseases and as such provide tractable preclinical models for research and defined patient groups for trials. Prof. Lebrin expects to identify early cerebrovascular markers that could ultimately be translated to the clinic for monitoring of disease progression and drug action.
In the second portion of the webinar, Prof. Vivien presents current therapeutic approaches for ischemic stroke intended to restore cerebral blood flow (CBF) as quickly and efficiently as possible, including rtPA treatment (tissue type Plasminogen Activator) and endovascular thrombectomy (EVT). Prof. Vivien’s research aims to better understand the spatiotemporal evolution of specific functional and molecular events that occur during and following stroke, using a unique combination of in vivo high-resolution functional ultrafast ultrasound imaging (HR-fUS) and high-resolution molecular magnetic resonance imaging (HRmol-MRI). This project has led to the proposal of an innovative platform to test future therapeutics of ischemic stroke with a greater chance of successful translation to the clinic.
This talk begins by showing how accurately 4000 different diagnoses can be predicted in advance for any patient, from one month to twenty years before first occurrence in the patient, using high-throughput machine learning. Shortcomings of this approach motivate ways to turn prediction methods into algorithms for finding causal associations; the resulting algorithms attain high accuracy in tasks of drug repurposing and discovery of adverse drug events, but they do not come with provable guarantees of making correct causal inferences. We then introduce variants of probably-approximately correct (PAC) learning for finding causal associations, that can provide weaker but useful guarantees for such algorithms as these motivated by our experiences with EHR data.
Control of Nonlinear Heartbeat Models under Time- Delay-Switched Feedback Usi...idescitation
In this paper, we adopt the Zeeman nonlinear heart model to discuss its stability
and control its operation using emotional learning control (ELC). We also demonstrate the
control of the heart model under threats of possible time delay introduced in the sensing
loop. We compare the robustness of the ELC with other control methods such as the
classical PID and the model predictive control (MPC) for the heart model under time delay
attack. We have showed that ELC is more robust than the classical PID and the MPC.
Elsevier Medical Graph – mit Machine Learning zu Precision MedicineRising Media Ltd.
Elsevier Health Analytics entwickelt den Medical Knowledge Graph, welcher Korrelationen zwischen Krankheiten und zwischen Krankheiten und Behandlungen darstellt. Auf einem Gesamtdatensatz von sechs Millionen anonymisierten Patienten, beobachtbar über sechs Jahre, haben wir über 2000 Modelle erstellt, welche die Entwicklung von Krankheiten prognostizieren. Jedes Modell ist adjustiert für mehr als 3000 Kovariablen. Dazu kam ein Boosting Algorithmus mit Variablenselektion zum Einsatz. Die Betas der selektierten Variablen wurden extrahiert, getestet hinsichtlich Kausalität und Signifikanz, und daraus wurde die erste Version des Medical Graphen mit über 2000 Krankheitsknoten und 25.000 Effekt-Kanten gebaut. Der Graph wird aktuell in der Praxis getestet, mit dem Ziel, dem Arzt eine patienten-individuelle Entscheidungsunterstützung für die Behandlung zu geben.
Heart Rate Variability (HRV) analysis is the
ability to assess overall cardiac health and the state of the
autonomic nervous system (ANS), responsible for regulating
cardiac activity. ST-change due to ischemia and their HRV
analysis have not been well discussed in the previous works.
The proposed simple and time efficient TBC algorithm has
been tested in four sets of standard databases with selected
patient’s data having ischemic conditions (i.e.MIT-BIH
Normal-Sinus Rhythm Database (NSRDB), European ST-T
Database (EDB), MIT-BIH ST Change Database (STDB) &
Long-Term ST Database (LTSTDB))for the detection of R-peak
& HRV analysis. The pre-processing is done by MAF and DWT
to remove the baseline drift and noise induced in the ECG
signal. The mean/average of HR is calculated for each set of
databases and in case of EDB it is of 57 BPM (subjected to
bradycardia). The Probability with normal distribution is
analyzed by comparing the NSRDB data with the ischemic data sets. The performance of this algorithm is found to be 98.5%.
Este manual es útil e indispensable para el uso del "Package TesSurvRec_1.2.1" de CRAN. Importante para estadístico, médicos, farmacéuticos, seguros, bancos, ingenieros, psicólogos, astrónomos, entre otras profesiones. Son pruebas estadísticas que se utilizan para medir diferencias entre funciones del análisis de supervivencias de grupos de poblaciones que manifiestan eventos recurrentes.
Essentials of Automations: The Art of Triggers and Actions in FMESafe Software
In this second installment of our Essentials of Automations webinar series, we’ll explore the landscape of triggers and actions, guiding you through the nuances of authoring and adapting workspaces for seamless automations. Gain an understanding of the full spectrum of triggers and actions available in FME, empowering you to enhance your workspaces for efficient automation.
We’ll kick things off by showcasing the most commonly used event-based triggers, introducing you to various automation workflows like manual triggers, schedules, directory watchers, and more. Plus, see how these elements play out in real scenarios.
Whether you’re tweaking your current setup or building from the ground up, this session will arm you with the tools and insights needed to transform your FME usage into a powerhouse of productivity. Join us to discover effective strategies that simplify complex processes, enhancing your productivity and transforming your data management practices with FME. Let’s turn complexity into clarity and make your workspaces work wonders!
The QRS changes during ischemia have historically been more difficult to parameterize
and have not come into clinical practice. This paper presented a new approach to analyze ischemia
by time parameter extraction of RS-Segment of the QRS complex. The proposed methodology
mainly focused on two prominent areas; first: detection of R and S points via Fast Fourier Transform
(FFT) based windowing & thresholding techniques with a sliding edge method. Second: calculating
the RS-Duration. The performances of the detection methods are validated and RS-Duration is
evaluated with the Fantasia database (Fantasia) for 20 healthy subjects & Long-Term ST Database
(LTSTDB) for 80 ischemic patients. The RS-Segment detection sensitivity (Se) and specificity (Sp)
are calculated 100% for Fantasia Database, whereas sensitivity (Se) is 91.6% and specificity (Sp) is
974% for LTSTDB.
Microvascular & Functional Ultrasound Imaging: Insights into Stroke and Neuro...InsideScientific
Professors Franck Lebrin and Denis Vivien discuss in vivo molecular and functional imaging, including ultrasound-based markers, and their application to the study and treatment of neurological disorders such as cerebral hereditary angiopathies and stroke.
Early vascular dysfunction is increasingly recognized as the underlying cause of many neurological diseases. The development of drugs targeting vascular damage at its earliest stages could therefore pave the road towards the treatment of neurological disorders. However, to be effective, this therapeutic approach will require the identification of early markers of vascular injury.
In the first portion of the webinar, Prof. Lebrin discusses his research on ultrasound markers of early vascular dysfunction in cerebral hereditary angiopathies (CHA) and the testing of novel therapeutic agents that restore vascular function. Albeit rare, blood vessels from CHA exhibit the salient features of other neurological diseases and as such provide tractable preclinical models for research and defined patient groups for trials. Prof. Lebrin expects to identify early cerebrovascular markers that could ultimately be translated to the clinic for monitoring of disease progression and drug action.
In the second portion of the webinar, Prof. Vivien presents current therapeutic approaches for ischemic stroke intended to restore cerebral blood flow (CBF) as quickly and efficiently as possible, including rtPA treatment (tissue type Plasminogen Activator) and endovascular thrombectomy (EVT). Prof. Vivien’s research aims to better understand the spatiotemporal evolution of specific functional and molecular events that occur during and following stroke, using a unique combination of in vivo high-resolution functional ultrafast ultrasound imaging (HR-fUS) and high-resolution molecular magnetic resonance imaging (HRmol-MRI). This project has led to the proposal of an innovative platform to test future therapeutics of ischemic stroke with a greater chance of successful translation to the clinic.
This talk begins by showing how accurately 4000 different diagnoses can be predicted in advance for any patient, from one month to twenty years before first occurrence in the patient, using high-throughput machine learning. Shortcomings of this approach motivate ways to turn prediction methods into algorithms for finding causal associations; the resulting algorithms attain high accuracy in tasks of drug repurposing and discovery of adverse drug events, but they do not come with provable guarantees of making correct causal inferences. We then introduce variants of probably-approximately correct (PAC) learning for finding causal associations, that can provide weaker but useful guarantees for such algorithms as these motivated by our experiences with EHR data.
Control of Nonlinear Heartbeat Models under Time- Delay-Switched Feedback Usi...idescitation
In this paper, we adopt the Zeeman nonlinear heart model to discuss its stability
and control its operation using emotional learning control (ELC). We also demonstrate the
control of the heart model under threats of possible time delay introduced in the sensing
loop. We compare the robustness of the ELC with other control methods such as the
classical PID and the model predictive control (MPC) for the heart model under time delay
attack. We have showed that ELC is more robust than the classical PID and the MPC.
Elsevier Medical Graph – mit Machine Learning zu Precision MedicineRising Media Ltd.
Elsevier Health Analytics entwickelt den Medical Knowledge Graph, welcher Korrelationen zwischen Krankheiten und zwischen Krankheiten und Behandlungen darstellt. Auf einem Gesamtdatensatz von sechs Millionen anonymisierten Patienten, beobachtbar über sechs Jahre, haben wir über 2000 Modelle erstellt, welche die Entwicklung von Krankheiten prognostizieren. Jedes Modell ist adjustiert für mehr als 3000 Kovariablen. Dazu kam ein Boosting Algorithmus mit Variablenselektion zum Einsatz. Die Betas der selektierten Variablen wurden extrahiert, getestet hinsichtlich Kausalität und Signifikanz, und daraus wurde die erste Version des Medical Graphen mit über 2000 Krankheitsknoten und 25.000 Effekt-Kanten gebaut. Der Graph wird aktuell in der Praxis getestet, mit dem Ziel, dem Arzt eine patienten-individuelle Entscheidungsunterstützung für die Behandlung zu geben.
Heart Rate Variability (HRV) analysis is the
ability to assess overall cardiac health and the state of the
autonomic nervous system (ANS), responsible for regulating
cardiac activity. ST-change due to ischemia and their HRV
analysis have not been well discussed in the previous works.
The proposed simple and time efficient TBC algorithm has
been tested in four sets of standard databases with selected
patient’s data having ischemic conditions (i.e.MIT-BIH
Normal-Sinus Rhythm Database (NSRDB), European ST-T
Database (EDB), MIT-BIH ST Change Database (STDB) &
Long-Term ST Database (LTSTDB))for the detection of R-peak
& HRV analysis. The pre-processing is done by MAF and DWT
to remove the baseline drift and noise induced in the ECG
signal. The mean/average of HR is calculated for each set of
databases and in case of EDB it is of 57 BPM (subjected to
bradycardia). The Probability with normal distribution is
analyzed by comparing the NSRDB data with the ischemic data sets. The performance of this algorithm is found to be 98.5%.
Este manual es útil e indispensable para el uso del "Package TesSurvRec_1.2.1" de CRAN. Importante para estadístico, médicos, farmacéuticos, seguros, bancos, ingenieros, psicólogos, astrónomos, entre otras profesiones. Son pruebas estadísticas que se utilizan para medir diferencias entre funciones del análisis de supervivencias de grupos de poblaciones que manifiestan eventos recurrentes.
Essentials of Automations: The Art of Triggers and Actions in FMESafe Software
In this second installment of our Essentials of Automations webinar series, we’ll explore the landscape of triggers and actions, guiding you through the nuances of authoring and adapting workspaces for seamless automations. Gain an understanding of the full spectrum of triggers and actions available in FME, empowering you to enhance your workspaces for efficient automation.
We’ll kick things off by showcasing the most commonly used event-based triggers, introducing you to various automation workflows like manual triggers, schedules, directory watchers, and more. Plus, see how these elements play out in real scenarios.
Whether you’re tweaking your current setup or building from the ground up, this session will arm you with the tools and insights needed to transform your FME usage into a powerhouse of productivity. Join us to discover effective strategies that simplify complex processes, enhancing your productivity and transforming your data management practices with FME. Let’s turn complexity into clarity and make your workspaces work wonders!
Mobile app Development Services | Drona InfotechDrona Infotech
Drona Infotech is one of the Best Mobile App Development Company In Noida Maintenance and ongoing support. mobile app development Services can help you maintain and support your app after it has been launched. This includes fixing bugs, adding new features, and keeping your app up-to-date with the latest
Visit Us For :
GraphSummit Paris - The art of the possible with Graph TechnologyNeo4j
Sudhir Hasbe, Chief Product Officer, Neo4j
Join us as we explore breakthrough innovations enabled by interconnected data and AI. Discover firsthand how organizations use relationships in data to uncover contextual insights and solve our most pressing challenges – from optimizing supply chains, detecting fraud, and improving customer experiences to accelerating drug discoveries.
Mobile App Development Company In Noida | Drona InfotechDrona Infotech
Looking for a reliable mobile app development company in Noida? Look no further than Drona Infotech. We specialize in creating customized apps for your business needs.
Visit Us For : https://www.dronainfotech.com/mobile-application-development/
OpenMetadata Community Meeting - 5th June 2024OpenMetadata
The OpenMetadata Community Meeting was held on June 5th, 2024. In this meeting, we discussed about the data quality capabilities that are integrated with the Incident Manager, providing a complete solution to handle your data observability needs. Watch the end-to-end demo of the data quality features.
* How to run your own data quality framework
* What is the performance impact of running data quality frameworks
* How to run the test cases in your own ETL pipelines
* How the Incident Manager is integrated
* Get notified with alerts when test cases fail
Watch the meeting recording here - https://www.youtube.com/watch?v=UbNOje0kf6E
Code reviews are vital for ensuring good code quality. They serve as one of our last lines of defense against bugs and subpar code reaching production.
Yet, they often turn into annoying tasks riddled with frustration, hostility, unclear feedback and lack of standards. How can we improve this crucial process?
In this session we will cover:
- The Art of Effective Code Reviews
- Streamlining the Review Process
- Elevating Reviews with Automated Tools
By the end of this presentation, you'll have the knowledge on how to organize and improve your code review proces
When deliberating between CodeIgniter vs CakePHP for web development, consider their respective strengths and your project requirements. CodeIgniter, known for its simplicity and speed, offers a lightweight framework ideal for rapid development of small to medium-sized projects. It's praised for its straightforward configuration and extensive documentation, making it beginner-friendly. Conversely, CakePHP provides a more structured approach with built-in features like scaffolding, authentication, and ORM. It suits larger projects requiring robust security and scalability. Ultimately, the choice hinges on your project's scale, complexity, and your team's familiarity with the frameworks.
Launch Your Streaming Platforms in MinutesRoshan Dwivedi
The claim of launching a streaming platform in minutes might be a bit of an exaggeration, but there are services that can significantly streamline the process. Here's a breakdown:
Pros of Speedy Streaming Platform Launch Services:
No coding required: These services often use drag-and-drop interfaces or pre-built templates, eliminating the need for programming knowledge.
Faster setup: Compared to building from scratch, these platforms can get you up and running much quicker.
All-in-one solutions: Many services offer features like content management systems (CMS), video players, and monetization tools, reducing the need for multiple integrations.
Things to Consider:
Limited customization: These platforms may offer less flexibility in design and functionality compared to custom-built solutions.
Scalability: As your audience grows, you might need to upgrade to a more robust platform or encounter limitations with the "quick launch" option.
Features: Carefully evaluate which features are included and if they meet your specific needs (e.g., live streaming, subscription options).
Examples of Services for Launching Streaming Platforms:
Muvi [muvi com]
Uscreen [usencreen tv]
Alternatives to Consider:
Existing Streaming platforms: Platforms like YouTube or Twitch might be suitable for basic streaming needs, though monetization options might be limited.
Custom Development: While more time-consuming, custom development offers the most control and flexibility for your platform.
Overall, launching a streaming platform in minutes might not be entirely realistic, but these services can significantly speed up the process compared to building from scratch. Carefully consider your needs and budget when choosing the best option for you.
Do you want Software for your Business? Visit Deuglo
Deuglo has top Software Developers in India. They are experts in software development and help design and create custom Software solutions.
Deuglo follows seven steps methods for delivering their services to their customers. They called it the Software development life cycle process (SDLC).
Requirement — Collecting the Requirements is the first Phase in the SSLC process.
Feasibility Study — after completing the requirement process they move to the design phase.
Design — in this phase, they start designing the software.
Coding — when designing is completed, the developers start coding for the software.
Testing — in this phase when the coding of the software is done the testing team will start testing.
Installation — after completion of testing, the application opens to the live server and launches!
Maintenance — after completing the software development, customers start using the software.
Graspan: A Big Data System for Big Code AnalysisAftab Hussain
We built a disk-based parallel graph system, Graspan, that uses a novel edge-pair centric computation model to compute dynamic transitive closures on very large program graphs.
We implement context-sensitive pointer/alias and dataflow analyses on Graspan. An evaluation of these analyses on large codebases such as Linux shows that their Graspan implementations scale to millions of lines of code and are much simpler than their original implementations.
These analyses were used to augment the existing checkers; these augmented checkers found 132 new NULL pointer bugs and 1308 unnecessary NULL tests in Linux 4.4.0-rc5, PostgreSQL 8.3.9, and Apache httpd 2.2.18.
- Accepted in ASPLOS ‘17, Xi’an, China.
- Featured in the tutorial, Systemized Program Analyses: A Big Data Perspective on Static Analysis Scalability, ASPLOS ‘17.
- Invited for presentation at SoCal PLS ‘16.
- Invited for poster presentation at PLDI SRC ‘16.
AI Pilot Review: The World’s First Virtual Assistant Marketing SuiteGoogle
AI Pilot Review: The World’s First Virtual Assistant Marketing Suite
👉👉 Click Here To Get More Info 👇👇
https://sumonreview.com/ai-pilot-review/
AI Pilot Review: Key Features
✅Deploy AI expert bots in Any Niche With Just A Click
✅With one keyword, generate complete funnels, websites, landing pages, and more.
✅More than 85 AI features are included in the AI pilot.
✅No setup or configuration; use your voice (like Siri) to do whatever you want.
✅You Can Use AI Pilot To Create your version of AI Pilot And Charge People For It…
✅ZERO Manual Work With AI Pilot. Never write, Design, Or Code Again.
✅ZERO Limits On Features Or Usages
✅Use Our AI-powered Traffic To Get Hundreds Of Customers
✅No Complicated Setup: Get Up And Running In 2 Minutes
✅99.99% Up-Time Guaranteed
✅30 Days Money-Back Guarantee
✅ZERO Upfront Cost
See My Other Reviews Article:
(1) TubeTrivia AI Review: https://sumonreview.com/tubetrivia-ai-review
(2) SocioWave Review: https://sumonreview.com/sociowave-review
(3) AI Partner & Profit Review: https://sumonreview.com/ai-partner-profit-review
(4) AI Ebook Suite Review: https://sumonreview.com/ai-ebook-suite-review
Introducing Crescat - Event Management Software for Venues, Festivals and Eve...Crescat
Crescat is industry-trusted event management software, built by event professionals for event professionals. Founded in 2017, we have three key products tailored for the live event industry.
Crescat Event for concert promoters and event agencies. Crescat Venue for music venues, conference centers, wedding venues, concert halls and more. And Crescat Festival for festivals, conferences and complex events.
With a wide range of popular features such as event scheduling, shift management, volunteer and crew coordination, artist booking and much more, Crescat is designed for customisation and ease-of-use.
Over 125,000 events have been planned in Crescat and with hundreds of customers of all shapes and sizes, from boutique event agencies through to international concert promoters, Crescat is rigged for success. What's more, we highly value feedback from our users and we are constantly improving our software with updates, new features and improvements.
If you plan events, run a venue or produce festivals and you're looking for ways to make your life easier, then we have a solution for you. Try our software for free or schedule a no-obligation demo with one of our product specialists today at crescat.io
Professor Chris Williams et al - Healthcare condition monitoring using ICU data
1. Healthcare condition monitoring using ICU
data
Chris Williams
joint work with Yvonne Freer, Konstantinos Georgatzis, Chris
Hawthorne, Partha Lal, Neil McIntosh, Ian Piper, John Quinn,
Martin Shaw, Ioan Stanculescu
School of Informatics, University of Edinburgh,
and Alan Turing Institute, London
November 2017
1 / 32
2. My main research interests:
Time series understanding
Computer vision, especially object recognition, shape and
texture modelling
Semi-automation of data cleaning and preparation
Unsupervised learning
Gaussian processes
2 / 32
3. Time Series Understanding
Explain the multivariate time series in terms of an
underlying set of discrete factors
Make inferences for underlying variables when
observations are corrupted by artifact
We will address such problems with various switching
linear dynamical systems (SLDS) models
BS
Time (s)
BR
0 200 400 600 800
0
100
200
HR(bpm)
20
40
60
80
Sys.BP(mmHg)
20
40
60
Dia.BP(mmHg)
3 / 32
4. ICU Condition Monitoring
Population: patients receiving intensive care
Data: physiological vital signs recordings
Problems: artifact corruption, false alarms, amount of data
Goal: Determine the state of health of the patient,
uncorrupted vital signs
Image source: Wikipedia Intensive Care Unit page
4 / 32
6. Factors Affecting Measurements
The physiological observations are affected by different
factors.
Factors can be artifactual or physiological.
30
40
50
60
Sys.BP(mmHg)
0 200 400 600 800 1000
0
20
40
60
Dia.BP(mmHg)
Time (s)
0 20 40 60 80 100
40
60
80
100
120
140
160
180
HR(bpm) Time (s)
Arterial blood sample Bradycardia
6 / 32
7. Factorial Switching Linear Dynamical System
Artifactual state
Physiological state
Observations
Physiological factors
Artifactual factors
7 / 32
8. FSLDS notation
st is the switch variable, which indexes factor settings, e.g.
‘blood sample occurring and first stage of TCP
recalibration’.
xt is the hidden continuous state at time t. This contains
information on the true physiology of the baby, and on the
levels of artifactual processes.
y1:t are the observations.
8 / 32
9. Kalman filtering
Continuous hidden state affects some observations:
xt ∼ N(Axt−1, Q)
yt ∼ N(Cxt , R)
Kalman filter equations can be used to work compute
p(x1:t |y1:t )
Done iteratively by predicting and updating
9 / 32
10. Switching dynamics
The switch variable st selects the dynamics for a particular
combination of factor settings:
xt ∼ N(A(st )
xt−1, Q(st )
)
yt ∼ N(C(st )
xt , R(st )
)
For each setting of st , the Kalman filter equations give a
predictive distribution for xt .
10 / 32
12. Related work
Switching linear dynamical models have been studied by
many authors, e.g. Alspach and Sorenson (1972),
Ghahramani and Hinton (1996).
Applications include fault detection in mobile robots (de
Freitas et al., 2004), speech recognition (Droppo and
Acero, 2004), industrial monitoring (Morales-Menedez et
al., 2002).
A two-factor FSLDS was used for speech recognition by
Ma and Deng (2004). Factorised SLDS also used for
musical transcription (Cemgil et al., 2006).
There has been previous work on condition monitoring in
the ICU, though we are unaware of any previous studies
that use a FSLDS.
12 / 32
13. Inference and Learning
For this application, we are interested in filtering, inferring
p(st , xt |y1:t )
Exact inference is intractable (Lerner and Parr, 2001)
We use the Gaussian sum approximation (e.g. Murphy,
1998)
Learning uses labelled data for different regimes, and
overwriting order of factors
13 / 32
14. Example inference results
Can examine variance of estimates of true physiology ˆxt ,
e.g. for blood sample (left) and temperature probe
disconnection (right):
Time (s)
BS
0 50 100 150 200 250
Sys.BP(mmHg)
35
40
45
50
55
Dia.BP(mmHg)
20
30
40
50
Time (s)
TD
0 500 1000
Coretemp.(°C)
35
35.5
36
36.5
37
37.5
38
14 / 32
15. Models: FSLDS, DSLDS
DSLDS (Georgatzis and Williams, UAI 2015)
st is predicted with a classifier
Inference for xt is similar to FSLDS
α-mixture combines FSLDS and DSLDS
15 / 32
16. FSLDS and DSLDS: pros and cons
+ Knowledge engineering tells us how the factors interact
generatively
+ There is not very much labelled data
+ Normality varies per patient (multi-task learning)
- In the DSLDS discrete state distributions are predicted
directly, rather than inferred. Can encode knowledge with
informative features.
- Some events (esp. artifactual) might be easier to identify
with a discriminative approach. Harder to come up with a
generative model.
16 / 32
17. Novel Regimes
There are many other factors influencing the data: drugs,
sepsis, neurological problems...
50
100
150
200
Heart rate
40
50
60
70
Dia. BP
0 200 400 600 800 1000 1200
0
50
100
SpO2
?
17 / 32
20. X-factor for static 1-D data
For static data, we can use a model M∗ representing
‘abnormal’ data points.
y
p(y|s)
The high-variance model wins when the data is not well
explained by the original model
20 / 32
21. X-factor with known factors
The X-factor can be applied to the static data in
conjunction with known factors (green):
y
p(y|s)
21 / 32
22. X-factor for dynamic data
xt ∼ N(Axt−1, Q)
yt ∼ N(Cxt , R)
Can construct an ‘abnormal’ dynamic regime analogously:
Normal dynamics: {A, Q, C, R}
X-factor dynamics: {A,ξQ, C, R}, ξ > 1.
22 / 32
23. Spectral view of the X-factor
f
S
y
(f)
0 1/2
Plot shows the spectrum of a hidden AR(5) process, and
accompanying X-factor
More power at every frequency
Dynamical analogue of the static 1-D case
23 / 32
24. Data
27 patients from Neuro ICU in the Southern General
Hospital, Glasgow (15 TBI, 12 SAH)
Channels:
arterial blood pressure (ABP)
electrocardiogram (ECG)
pulse oximetry
intracranial pressure (ICP)
end tidal CO2 (EtCO2)
respiratory signal (Resp)
Downsampled to 1 Hz
24 / 32
25. Annotation
46 event-types labelled, including blood sample, damped
trace, patient turning and suctioning
Damped trace events have a mean duration of over 8
hours per patient
Other significant events: blood sample, patient turning and
suctioning, noisy channels, preparation for or return from
transfer
25 / 32
26. Processing pipeline
Extraction from
ICU database
Preprocessing FSLDS
Stability
detection
Made to work all together on ICU server
System operates at ∼ 10× realtime
Stability detection: need to estimate AR/ARMA parameters
for every patient individually for the stability regime
This is done by predicting intervals that are stable vs
non-stable, and using these to learn the stability regime
Software available at https:
//datashare.is.ed.ac.uk/handle/10283/855
26 / 32
28. AUC BS DT SC X
DSLDS 0.94 0.78 0.64 0.56
FSLDS 0.86 0.77 0.60 0.60
α-mixture 0.95(0.9) 0.79(0.9) 0.64(−∞) 0.61(1.4)
Blood sample performance is very good, and is potentially
useful for silencing false alarms
Damped trace is particularly interesting as it has significant
duration and is not an event caused by nursing
interventions; it is therefore particularly helpful to flag up
Suction events are complex and have a variable time
course. Also suction and position change events can have
similar effects on the patient. Position change was not
modelled with a factor in our experiments, thus it may not
be surprising if these two event types are confused
28 / 32
29. Damped Trace Example
True X
True SC
True BS
True DT
00:13:00 00:13:45 00:14:30 00:15:15 00:16:00 00:16:45 00:17:30 00:18:15 00:19:00 00:19:44 00:20:29 00:21:14 00:21:59 00:22:44 00:23:29 00:24:14 00:24:59
0
50
100
150
200
250
ABP
(mmHg)
Patient damped_trace_demo
Dia.
Mean
Sys.
X −− DSLDS
X −− FSLDS
X −− alpha
SC −− DSLDS
SC −− FSLDS
SC −− alpha
BS −− DSLDS
BS −− FSLDS
BS −− alpha
DT −− DSLDS
DT −− FSLDS
DT −− alpha
0.2
0.4
0.6
0.8
1
Note imputed x-state
Our clinicians believe that showing imputed state and
flagging up artifact would be helpful
29 / 32
31. Summary
Quantification of the amount of artifact in this dataset,
importance of damped trace events
AUC scores are very high for blood samples (0.95), good
for damped trace (0.79), and poor for suction (0.64) and
X-factor (0.61) events
Successful implementation of a real-time system carrying
out FSLDS analysis on the raw data coming from the ICU
FSLDS/DSLDS models can be applied to other ICU
monitoring tasks (e.g. identifying sepsis) and more
generally
We are also developing models for the effect of
interventions (e.g. drug administration)
Funding: Chief Scientist Office (Scotland) CHZ/4/801
31 / 32
32. References
Factorial Switching Linear Dynamical Systems applied to
Physiological Condition Monitoring.
John A. Quinn, Christopher K.I. Williams, Neil McIntosh. IEEE
Trans. on Pattern Analysis and Machine Intelligence 31(9) pp
1537-1551 (2009).
Discriminative Switching Linear Dynamical Systems applied to
Physiological Condition Monitoring. Konstantinos Georgatzis,
Christopher K. I. Williams, Proc UAI 2015.
Detecting Artifactual Events in Vital Signs Monitoring Data.
Partha Lal, Christopher K. I. Williams, Konstantinos Georgatzis,
Christopher Hawthorne, Paul McMonagle, Ian Piper, Martin
Shaw. Tech report, September 2015.
Available from http://homepages.inf.ed.ac.uk/ckiw/
32 / 32