SlideShare a Scribd company logo
1 of 14
Anomaly Detection for
Seizure Prediction
(anomaly detection in waveforms)
Cynthia Alexander
22nd September, 2016
Anomaly Detection for Seizure Prediction Cynthia Alexander
iEEG = intracranial + electricity + brain + picture
(intracranial electro encephalogram)
Electrode assemblies
(implant)
Telemetry unit
(implant)
Personal advisory
device (hand-held)
Anomaly Detection for Seizure Prediction Cynthia Alexander
Implanted system
Seizure prediction – brief Hows and Whys
Motivation:
Easily enhance day-to-day life of individuals suffering from
epilepsy (~1% of adults over 18 years old*) by warning them
about events that might happen so they can take precautions.
Or, even better, aid seizure prevention when combined with new
treatments (i.e. deep-brain stimulation).
10m 10m 10m 10m 10m 10m 5m Seizure
1 hour (Pre-seizure)
Telemetry unit output – data over time for 16 channels
Anomaly Detection for Seizure Prediction Cynthia Alexander
Standby
For the model, each patient’s recordings over the 16 channels
are split in 10 minutes windows
The goal is to accurately call pre-seizure windows as they
happen, to then warn an individual facing a seizure
Anomaly Detection for Seizure Prediction Cynthia Alexander
Pipeline
Random Forests
for Statistical Metrics and
Signal Transformations
2 layers Neural Network
for spectrograms classification
using Tensor Flow tools
Sample selection
Formatting and structure
EDA, Data cleaning
Input: Raw data from iEEGs
Output: Structured clean data
Input: Descriptive metrics, Signal
transformations, Channel Correlations,
Spectrograms
Engineering
Feature Selection
Summaries
Testing models
Parameter tuning
Validation
Features – Statistical metrics
Anomaly Detection for Seizure Prediction Cynthia Alexander
mean channel 11
Normalized distribution
std channel 00
Normalized distribution
absum channel 02
Normalized distribution
Standby
Pre-seizure
Features – Statistical metrics
Anomaly Detection for Seizure Prediction Cynthia Alexander
mean channel 11
Normalized distribution
std channel 00
Normalized distribution
absum channel 02
Normalized distribution
Standby
Pre-seizure
Remarkable differences across patients – i.e. Values for the median across channels
Each column one sample (10 min x 16 channels).
Medianperchannel
andwindow
Patient A Patient B Patient C
Pre-seizure
Image from mriquestions.com
Anomaly Detection for Seizure Prediction Cynthia Alexander
Features – Correlation and Signal
TransformationsAverage correlation of channels
over all samples
Standby samples
vs. all samples
Pre-seizure samples
vs. all samples
Wave decomposition
Fourier Transform
Each readout can be decomposed into the frequencies that make it up:
Gamma
(32+ Hz)
Beta
(16 – 31 Hz)
Alpha
(8 – 15 Hz)
Theta
(4 -7 Hz)
Delta
(< 4 Hz)
Image from mriquestions.com
Anomaly Detection for Seizure Prediction Cynthia Alexander
Features – Correlation and Signal
TransformationsAverage correlation of channels
over all samples
Standby samples
vs. all samples
Pre-seizure samples
vs. all samples
Wave decomposition
Fourier Transform
Each readout can be decomposed into the frequencies that make it up:
Gamma
(32+ Hz)
Beta
(16 – 31 Hz)
Alpha
(8 – 15 Hz)
Theta
(4 -7 Hz)
Delta
(< 4 Hz)
Anomaly Detection for Seizure Prediction Cynthia Alexander
Features – Spectrograms
Standby Pre-Seizure
Time window
Frequency
Results
Personalized model (train and predict on self over time):
Trained on other patient’s data predict on (no self-prior knowledge):
Anomaly Detection for Seizure Prediction Cynthia Alexander
Anomaly Detection for Seizure Prediction Cynthia Alexander
Interactive D3 App – Red shaded areas depict windows in which the model called a pre-seizure sample
In development and in order to help
users:
• Concentrate on personalized models
• Feed and predict on-App using real time data
• In the long run, aim for preemptive medicine
where care is specially provided in high-risk
windows (i.e. deep brain stimulation)
Anomaly Detection for Seizure Prediction Cynthia Alexander
Questions? Reach out!
Cynthia Alexander
cynth.alexander@gmail.com
4chromat
cynthiaalexanderrascon
4chromat.github.io
Thank you!
Anomaly Detection for Seizure Prediction Cynthia Alexander
Anomaly Detection for Seizure Prediction Cynthia Alexander
Data cleaning – Signal drop-outs and artifacts!

More Related Content

Viewers also liked

B ortiz aprendizaje autónomo
B ortiz aprendizaje autónomoB ortiz aprendizaje autónomo
B ortiz aprendizaje autónomoBryan Ortiz
 
Relief india trust ngos
Relief india trust ngosRelief india trust ngos
Relief india trust ngosvikasrelif
 
Role of entrepreneur
Role of entrepreneurRole of entrepreneur
Role of entrepreneurishwar kumar
 
TOKA-ΠΑΡΟΥΣΙΑΣΗ ΠΤΥΧΙΑΚΗΣ
TOKA-ΠΑΡΟΥΣΙΑΣΗ ΠΤΥΧΙΑΚΗΣTOKA-ΠΑΡΟΥΣΙΑΣΗ ΠΤΥΧΙΑΚΗΣ
TOKA-ΠΑΡΟΥΣΙΑΣΗ ΠΤΥΧΙΑΚΗΣElpida Toka
 
Relief india trust(everything)
Relief india trust(everything)Relief india trust(everything)
Relief india trust(everything)vikasrelif
 
Relief india trust an educational ngo in india
Relief india trust  an educational ngo in indiaRelief india trust  an educational ngo in india
Relief india trust an educational ngo in indiavikasrelif
 
Burning rocket in greenfoot
Burning rocket in greenfootBurning rocket in greenfoot
Burning rocket in greenfootsulthan zaky
 

Viewers also liked (14)

Preguntas
PreguntasPreguntas
Preguntas
 
B ortiz aprendizaje autónomo
B ortiz aprendizaje autónomoB ortiz aprendizaje autónomo
B ortiz aprendizaje autónomo
 
Aprendizaje autónomo
Aprendizaje autónomoAprendizaje autónomo
Aprendizaje autónomo
 
Nathaly acuña
Nathaly acuñaNathaly acuña
Nathaly acuña
 
La depresión
La depresiónLa depresión
La depresión
 
Relief india trust ngos
Relief india trust ngosRelief india trust ngos
Relief india trust ngos
 
Sirley y nathaly
Sirley y nathalySirley y nathaly
Sirley y nathaly
 
Role of entrepreneur
Role of entrepreneurRole of entrepreneur
Role of entrepreneur
 
TOKA-ΠΑΡΟΥΣΙΑΣΗ ΠΤΥΧΙΑΚΗΣ
TOKA-ΠΑΡΟΥΣΙΑΣΗ ΠΤΥΧΙΑΚΗΣTOKA-ΠΑΡΟΥΣΙΑΣΗ ΠΤΥΧΙΑΚΗΣ
TOKA-ΠΑΡΟΥΣΙΑΣΗ ΠΤΥΧΙΑΚΗΣ
 
Relief india trust(everything)
Relief india trust(everything)Relief india trust(everything)
Relief india trust(everything)
 
Relief india trust an educational ngo in india
Relief india trust  an educational ngo in indiaRelief india trust  an educational ngo in india
Relief india trust an educational ngo in india
 
Creatividad y-proceso-creativo
Creatividad y-proceso-creativoCreatividad y-proceso-creativo
Creatividad y-proceso-creativo
 
Ed in india
Ed in indiaEd in india
Ed in india
 
Burning rocket in greenfoot
Burning rocket in greenfootBurning rocket in greenfoot
Burning rocket in greenfoot
 

Similar to Anomaly Detection for Seizure Prediction

Acquiring Ecg Signals And Analysing For Different Heart Ailments
Acquiring Ecg Signals And Analysing For Different Heart AilmentsAcquiring Ecg Signals And Analysing For Different Heart Ailments
Acquiring Ecg Signals And Analysing For Different Heart AilmentsIJERA Editor
 
心臟植入性電子儀器(CIED )之歷史”CIED Overview “_20130914中區
心臟植入性電子儀器(CIED )之歷史”CIED Overview “_20130914中區心臟植入性電子儀器(CIED )之歷史”CIED Overview “_20130914中區
心臟植入性電子儀器(CIED )之歷史”CIED Overview “_20130914中區Taiwan Heart Rhythm Society
 
Sleep 2008 Electronicsv.3
Sleep 2008 Electronicsv.3Sleep 2008 Electronicsv.3
Sleep 2008 Electronicsv.3weckhardt
 
Electrophysiologic basics,part1(lecture)
Electrophysiologic basics,part1(lecture)Electrophysiologic basics,part1(lecture)
Electrophysiologic basics,part1(lecture)salah_atta
 
Electrophysiology basics,part1(lecture)
Electrophysiology basics,part1(lecture)Electrophysiology basics,part1(lecture)
Electrophysiology basics,part1(lecture)salah_atta
 
BIO MEDICAL RECORDERS
BIO MEDICAL RECORDERS BIO MEDICAL RECORDERS
BIO MEDICAL RECORDERS gowri R
 
Lecturing 1
Lecturing 1Lecturing 1
Lecturing 1gowri R
 
TMS for beginners
TMS for beginnersTMS for beginners
TMS for beginnersDennis
 
EARLY STROKE IDENTIFICATION USING MICROWAVE HELMET
EARLY STROKE IDENTIFICATION USING MICROWAVE HELMETEARLY STROKE IDENTIFICATION USING MICROWAVE HELMET
EARLY STROKE IDENTIFICATION USING MICROWAVE HELMETCiju Varghese
 
(short presentation) A Wearable Monitoring System for Continuously Assessing ...
(short presentation) A Wearable Monitoring System for Continuously Assessing ...(short presentation) A Wearable Monitoring System for Continuously Assessing ...
(short presentation) A Wearable Monitoring System for Continuously Assessing ...Meghan Hegarty
 
A Wearable Monitoring System for Continuously Assessing the Health of the Per...
A Wearable Monitoring System for Continuously Assessing the Health of the Per...A Wearable Monitoring System for Continuously Assessing the Health of the Per...
A Wearable Monitoring System for Continuously Assessing the Health of the Per...Meghan Hegarty
 
Patients with pacemaker anaesthetic implications
Patients with pacemaker anaesthetic implicationsPatients with pacemaker anaesthetic implications
Patients with pacemaker anaesthetic implicationsGowri Shankar
 
植入性心臟電子儀器(CIED )的基本原理及設定_20130907北區
植入性心臟電子儀器(CIED )的基本原理及設定_20130907北區植入性心臟電子儀器(CIED )的基本原理及設定_20130907北區
植入性心臟電子儀器(CIED )的基本原理及設定_20130907北區Taiwan Heart Rhythm Society
 
DSP_2018_FOEHU - Lec 02 - Sampling of Continuous Time Signals
DSP_2018_FOEHU - Lec 02 - Sampling of Continuous Time SignalsDSP_2018_FOEHU - Lec 02 - Sampling of Continuous Time Signals
DSP_2018_FOEHU - Lec 02 - Sampling of Continuous Time SignalsAmr E. Mohamed
 

Similar to Anomaly Detection for Seizure Prediction (20)

Acquiring Ecg Signals And Analysing For Different Heart Ailments
Acquiring Ecg Signals And Analysing For Different Heart AilmentsAcquiring Ecg Signals And Analysing For Different Heart Ailments
Acquiring Ecg Signals And Analysing For Different Heart Ailments
 
Ecg
EcgEcg
Ecg
 
心臟植入性電子儀器(CIED )之歷史”CIED Overview “_20130914中區
心臟植入性電子儀器(CIED )之歷史”CIED Overview “_20130914中區心臟植入性電子儀器(CIED )之歷史”CIED Overview “_20130914中區
心臟植入性電子儀器(CIED )之歷史”CIED Overview “_20130914中區
 
Sleep 2008 Electronicsv.3
Sleep 2008 Electronicsv.3Sleep 2008 Electronicsv.3
Sleep 2008 Electronicsv.3
 
Genomic Data Analysis
Genomic Data AnalysisGenomic Data Analysis
Genomic Data Analysis
 
Electrophysiologic basics,part1(lecture)
Electrophysiologic basics,part1(lecture)Electrophysiologic basics,part1(lecture)
Electrophysiologic basics,part1(lecture)
 
Electrophysiology basics,part1(lecture)
Electrophysiology basics,part1(lecture)Electrophysiology basics,part1(lecture)
Electrophysiology basics,part1(lecture)
 
Ct basics 2
Ct basics 2Ct basics 2
Ct basics 2
 
Ct scan
Ct scanCt scan
Ct scan
 
BIO MEDICAL RECORDERS
BIO MEDICAL RECORDERS BIO MEDICAL RECORDERS
BIO MEDICAL RECORDERS
 
Lecturing 1
Lecturing 1Lecturing 1
Lecturing 1
 
TMS for beginners
TMS for beginnersTMS for beginners
TMS for beginners
 
EARLY STROKE IDENTIFICATION USING MICROWAVE HELMET
EARLY STROKE IDENTIFICATION USING MICROWAVE HELMETEARLY STROKE IDENTIFICATION USING MICROWAVE HELMET
EARLY STROKE IDENTIFICATION USING MICROWAVE HELMET
 
(short presentation) A Wearable Monitoring System for Continuously Assessing ...
(short presentation) A Wearable Monitoring System for Continuously Assessing ...(short presentation) A Wearable Monitoring System for Continuously Assessing ...
(short presentation) A Wearable Monitoring System for Continuously Assessing ...
 
A Wearable Monitoring System for Continuously Assessing the Health of the Per...
A Wearable Monitoring System for Continuously Assessing the Health of the Per...A Wearable Monitoring System for Continuously Assessing the Health of the Per...
A Wearable Monitoring System for Continuously Assessing the Health of the Per...
 
Patients with pacemaker anaesthetic implications
Patients with pacemaker anaesthetic implicationsPatients with pacemaker anaesthetic implications
Patients with pacemaker anaesthetic implications
 
植入性心臟電子儀器(CIED )的基本原理及設定_20130907北區
植入性心臟電子儀器(CIED )的基本原理及設定_20130907北區植入性心臟電子儀器(CIED )的基本原理及設定_20130907北區
植入性心臟電子儀器(CIED )的基本原理及設定_20130907北區
 
DSP_2018_FOEHU - Lec 02 - Sampling of Continuous Time Signals
DSP_2018_FOEHU - Lec 02 - Sampling of Continuous Time SignalsDSP_2018_FOEHU - Lec 02 - Sampling of Continuous Time Signals
DSP_2018_FOEHU - Lec 02 - Sampling of Continuous Time Signals
 
NeuralTechPPT.ppt
NeuralTechPPT.pptNeuralTechPPT.ppt
NeuralTechPPT.ppt
 
Pacemaker
PacemakerPacemaker
Pacemaker
 

Recently uploaded

Saket, (-DELHI )+91-9654467111-(=)CHEAP Call Girls in Escorts Service Saket C...
Saket, (-DELHI )+91-9654467111-(=)CHEAP Call Girls in Escorts Service Saket C...Saket, (-DELHI )+91-9654467111-(=)CHEAP Call Girls in Escorts Service Saket C...
Saket, (-DELHI )+91-9654467111-(=)CHEAP Call Girls in Escorts Service Saket C...Sapana Sha
 
{Pooja: 9892124323 } Call Girl in Mumbai | Jas Kaur Rate 4500 Free Hotel Del...
{Pooja:  9892124323 } Call Girl in Mumbai | Jas Kaur Rate 4500 Free Hotel Del...{Pooja:  9892124323 } Call Girl in Mumbai | Jas Kaur Rate 4500 Free Hotel Del...
{Pooja: 9892124323 } Call Girl in Mumbai | Jas Kaur Rate 4500 Free Hotel Del...Pooja Nehwal
 
Aminabad Call Girl Agent 9548273370 , Call Girls Service Lucknow
Aminabad Call Girl Agent 9548273370 , Call Girls Service LucknowAminabad Call Girl Agent 9548273370 , Call Girls Service Lucknow
Aminabad Call Girl Agent 9548273370 , Call Girls Service Lucknowmakika9823
 
EMERCE - 2024 - AMSTERDAM - CROSS-PLATFORM TRACKING WITH GOOGLE ANALYTICS.pptx
EMERCE - 2024 - AMSTERDAM - CROSS-PLATFORM  TRACKING WITH GOOGLE ANALYTICS.pptxEMERCE - 2024 - AMSTERDAM - CROSS-PLATFORM  TRACKING WITH GOOGLE ANALYTICS.pptx
EMERCE - 2024 - AMSTERDAM - CROSS-PLATFORM TRACKING WITH GOOGLE ANALYTICS.pptxthyngster
 
Customer Service Analytics - Make Sense of All Your Data.pptx
Customer Service Analytics - Make Sense of All Your Data.pptxCustomer Service Analytics - Make Sense of All Your Data.pptx
Customer Service Analytics - Make Sense of All Your Data.pptxEmmanuel Dauda
 
Call Girls in Defence Colony Delhi 💯Call Us 🔝8264348440🔝
Call Girls in Defence Colony Delhi 💯Call Us 🔝8264348440🔝Call Girls in Defence Colony Delhi 💯Call Us 🔝8264348440🔝
Call Girls in Defence Colony Delhi 💯Call Us 🔝8264348440🔝soniya singh
 
B2 Creative Industry Response Evaluation.docx
B2 Creative Industry Response Evaluation.docxB2 Creative Industry Response Evaluation.docx
B2 Creative Industry Response Evaluation.docxStephen266013
 
High Class Call Girls Noida Sector 39 Aarushi 🔝8264348440🔝 Independent Escort...
High Class Call Girls Noida Sector 39 Aarushi 🔝8264348440🔝 Independent Escort...High Class Call Girls Noida Sector 39 Aarushi 🔝8264348440🔝 Independent Escort...
High Class Call Girls Noida Sector 39 Aarushi 🔝8264348440🔝 Independent Escort...soniya singh
 
VIP Call Girls Service Miyapur Hyderabad Call +91-8250192130
VIP Call Girls Service Miyapur Hyderabad Call +91-8250192130VIP Call Girls Service Miyapur Hyderabad Call +91-8250192130
VIP Call Girls Service Miyapur Hyderabad Call +91-8250192130Suhani Kapoor
 
Schema on read is obsolete. Welcome metaprogramming..pdf
Schema on read is obsolete. Welcome metaprogramming..pdfSchema on read is obsolete. Welcome metaprogramming..pdf
Schema on read is obsolete. Welcome metaprogramming..pdfLars Albertsson
 
Call Girls In Mahipalpur O9654467111 Escorts Service
Call Girls In Mahipalpur O9654467111  Escorts ServiceCall Girls In Mahipalpur O9654467111  Escorts Service
Call Girls In Mahipalpur O9654467111 Escorts ServiceSapana Sha
 
VIP High Class Call Girls Bikaner Anushka 8250192130 Independent Escort Servi...
VIP High Class Call Girls Bikaner Anushka 8250192130 Independent Escort Servi...VIP High Class Call Girls Bikaner Anushka 8250192130 Independent Escort Servi...
VIP High Class Call Girls Bikaner Anushka 8250192130 Independent Escort Servi...Suhani Kapoor
 
dokumen.tips_chapter-4-transient-heat-conduction-mehmet-kanoglu.ppt
dokumen.tips_chapter-4-transient-heat-conduction-mehmet-kanoglu.pptdokumen.tips_chapter-4-transient-heat-conduction-mehmet-kanoglu.ppt
dokumen.tips_chapter-4-transient-heat-conduction-mehmet-kanoglu.pptSonatrach
 
Brighton SEO | April 2024 | Data Storytelling
Brighton SEO | April 2024 | Data StorytellingBrighton SEO | April 2024 | Data Storytelling
Brighton SEO | April 2024 | Data StorytellingNeil Barnes
 
Log Analysis using OSSEC sasoasasasas.pptx
Log Analysis using OSSEC sasoasasasas.pptxLog Analysis using OSSEC sasoasasasas.pptx
Log Analysis using OSSEC sasoasasasas.pptxJohnnyPlasten
 
VIP High Class Call Girls Jamshedpur Anushka 8250192130 Independent Escort Se...
VIP High Class Call Girls Jamshedpur Anushka 8250192130 Independent Escort Se...VIP High Class Call Girls Jamshedpur Anushka 8250192130 Independent Escort Se...
VIP High Class Call Girls Jamshedpur Anushka 8250192130 Independent Escort Se...Suhani Kapoor
 
Building on a FAIRly Strong Foundation to Connect Academic Research to Transl...
Building on a FAIRly Strong Foundation to Connect Academic Research to Transl...Building on a FAIRly Strong Foundation to Connect Academic Research to Transl...
Building on a FAIRly Strong Foundation to Connect Academic Research to Transl...Jack DiGiovanna
 
04242024_CCC TUG_Joins and Relationships
04242024_CCC TUG_Joins and Relationships04242024_CCC TUG_Joins and Relationships
04242024_CCC TUG_Joins and Relationshipsccctableauusergroup
 

Recently uploaded (20)

Saket, (-DELHI )+91-9654467111-(=)CHEAP Call Girls in Escorts Service Saket C...
Saket, (-DELHI )+91-9654467111-(=)CHEAP Call Girls in Escorts Service Saket C...Saket, (-DELHI )+91-9654467111-(=)CHEAP Call Girls in Escorts Service Saket C...
Saket, (-DELHI )+91-9654467111-(=)CHEAP Call Girls in Escorts Service Saket C...
 
{Pooja: 9892124323 } Call Girl in Mumbai | Jas Kaur Rate 4500 Free Hotel Del...
{Pooja:  9892124323 } Call Girl in Mumbai | Jas Kaur Rate 4500 Free Hotel Del...{Pooja:  9892124323 } Call Girl in Mumbai | Jas Kaur Rate 4500 Free Hotel Del...
{Pooja: 9892124323 } Call Girl in Mumbai | Jas Kaur Rate 4500 Free Hotel Del...
 
Aminabad Call Girl Agent 9548273370 , Call Girls Service Lucknow
Aminabad Call Girl Agent 9548273370 , Call Girls Service LucknowAminabad Call Girl Agent 9548273370 , Call Girls Service Lucknow
Aminabad Call Girl Agent 9548273370 , Call Girls Service Lucknow
 
EMERCE - 2024 - AMSTERDAM - CROSS-PLATFORM TRACKING WITH GOOGLE ANALYTICS.pptx
EMERCE - 2024 - AMSTERDAM - CROSS-PLATFORM  TRACKING WITH GOOGLE ANALYTICS.pptxEMERCE - 2024 - AMSTERDAM - CROSS-PLATFORM  TRACKING WITH GOOGLE ANALYTICS.pptx
EMERCE - 2024 - AMSTERDAM - CROSS-PLATFORM TRACKING WITH GOOGLE ANALYTICS.pptx
 
Customer Service Analytics - Make Sense of All Your Data.pptx
Customer Service Analytics - Make Sense of All Your Data.pptxCustomer Service Analytics - Make Sense of All Your Data.pptx
Customer Service Analytics - Make Sense of All Your Data.pptx
 
Call Girls in Defence Colony Delhi 💯Call Us 🔝8264348440🔝
Call Girls in Defence Colony Delhi 💯Call Us 🔝8264348440🔝Call Girls in Defence Colony Delhi 💯Call Us 🔝8264348440🔝
Call Girls in Defence Colony Delhi 💯Call Us 🔝8264348440🔝
 
B2 Creative Industry Response Evaluation.docx
B2 Creative Industry Response Evaluation.docxB2 Creative Industry Response Evaluation.docx
B2 Creative Industry Response Evaluation.docx
 
High Class Call Girls Noida Sector 39 Aarushi 🔝8264348440🔝 Independent Escort...
High Class Call Girls Noida Sector 39 Aarushi 🔝8264348440🔝 Independent Escort...High Class Call Girls Noida Sector 39 Aarushi 🔝8264348440🔝 Independent Escort...
High Class Call Girls Noida Sector 39 Aarushi 🔝8264348440🔝 Independent Escort...
 
VIP Call Girls Service Miyapur Hyderabad Call +91-8250192130
VIP Call Girls Service Miyapur Hyderabad Call +91-8250192130VIP Call Girls Service Miyapur Hyderabad Call +91-8250192130
VIP Call Girls Service Miyapur Hyderabad Call +91-8250192130
 
Schema on read is obsolete. Welcome metaprogramming..pdf
Schema on read is obsolete. Welcome metaprogramming..pdfSchema on read is obsolete. Welcome metaprogramming..pdf
Schema on read is obsolete. Welcome metaprogramming..pdf
 
Call Girls In Mahipalpur O9654467111 Escorts Service
Call Girls In Mahipalpur O9654467111  Escorts ServiceCall Girls In Mahipalpur O9654467111  Escorts Service
Call Girls In Mahipalpur O9654467111 Escorts Service
 
VIP High Class Call Girls Bikaner Anushka 8250192130 Independent Escort Servi...
VIP High Class Call Girls Bikaner Anushka 8250192130 Independent Escort Servi...VIP High Class Call Girls Bikaner Anushka 8250192130 Independent Escort Servi...
VIP High Class Call Girls Bikaner Anushka 8250192130 Independent Escort Servi...
 
dokumen.tips_chapter-4-transient-heat-conduction-mehmet-kanoglu.ppt
dokumen.tips_chapter-4-transient-heat-conduction-mehmet-kanoglu.pptdokumen.tips_chapter-4-transient-heat-conduction-mehmet-kanoglu.ppt
dokumen.tips_chapter-4-transient-heat-conduction-mehmet-kanoglu.ppt
 
Brighton SEO | April 2024 | Data Storytelling
Brighton SEO | April 2024 | Data StorytellingBrighton SEO | April 2024 | Data Storytelling
Brighton SEO | April 2024 | Data Storytelling
 
E-Commerce Order PredictionShraddha Kamble.pptx
E-Commerce Order PredictionShraddha Kamble.pptxE-Commerce Order PredictionShraddha Kamble.pptx
E-Commerce Order PredictionShraddha Kamble.pptx
 
Log Analysis using OSSEC sasoasasasas.pptx
Log Analysis using OSSEC sasoasasasas.pptxLog Analysis using OSSEC sasoasasasas.pptx
Log Analysis using OSSEC sasoasasasas.pptx
 
VIP High Class Call Girls Jamshedpur Anushka 8250192130 Independent Escort Se...
VIP High Class Call Girls Jamshedpur Anushka 8250192130 Independent Escort Se...VIP High Class Call Girls Jamshedpur Anushka 8250192130 Independent Escort Se...
VIP High Class Call Girls Jamshedpur Anushka 8250192130 Independent Escort Se...
 
꧁❤ Aerocity Call Girls Service Aerocity Delhi ❤꧂ 9999965857 ☎️ Hard And Sexy ...
꧁❤ Aerocity Call Girls Service Aerocity Delhi ❤꧂ 9999965857 ☎️ Hard And Sexy ...꧁❤ Aerocity Call Girls Service Aerocity Delhi ❤꧂ 9999965857 ☎️ Hard And Sexy ...
꧁❤ Aerocity Call Girls Service Aerocity Delhi ❤꧂ 9999965857 ☎️ Hard And Sexy ...
 
Building on a FAIRly Strong Foundation to Connect Academic Research to Transl...
Building on a FAIRly Strong Foundation to Connect Academic Research to Transl...Building on a FAIRly Strong Foundation to Connect Academic Research to Transl...
Building on a FAIRly Strong Foundation to Connect Academic Research to Transl...
 
04242024_CCC TUG_Joins and Relationships
04242024_CCC TUG_Joins and Relationships04242024_CCC TUG_Joins and Relationships
04242024_CCC TUG_Joins and Relationships
 

Anomaly Detection for Seizure Prediction

  • 1. Anomaly Detection for Seizure Prediction (anomaly detection in waveforms) Cynthia Alexander 22nd September, 2016 Anomaly Detection for Seizure Prediction Cynthia Alexander
  • 2. iEEG = intracranial + electricity + brain + picture (intracranial electro encephalogram) Electrode assemblies (implant) Telemetry unit (implant) Personal advisory device (hand-held) Anomaly Detection for Seizure Prediction Cynthia Alexander Implanted system Seizure prediction – brief Hows and Whys Motivation: Easily enhance day-to-day life of individuals suffering from epilepsy (~1% of adults over 18 years old*) by warning them about events that might happen so they can take precautions. Or, even better, aid seizure prevention when combined with new treatments (i.e. deep-brain stimulation).
  • 3. 10m 10m 10m 10m 10m 10m 5m Seizure 1 hour (Pre-seizure) Telemetry unit output – data over time for 16 channels Anomaly Detection for Seizure Prediction Cynthia Alexander Standby For the model, each patient’s recordings over the 16 channels are split in 10 minutes windows The goal is to accurately call pre-seizure windows as they happen, to then warn an individual facing a seizure
  • 4. Anomaly Detection for Seizure Prediction Cynthia Alexander Pipeline Random Forests for Statistical Metrics and Signal Transformations 2 layers Neural Network for spectrograms classification using Tensor Flow tools Sample selection Formatting and structure EDA, Data cleaning Input: Raw data from iEEGs Output: Structured clean data Input: Descriptive metrics, Signal transformations, Channel Correlations, Spectrograms Engineering Feature Selection Summaries Testing models Parameter tuning Validation
  • 5. Features – Statistical metrics Anomaly Detection for Seizure Prediction Cynthia Alexander mean channel 11 Normalized distribution std channel 00 Normalized distribution absum channel 02 Normalized distribution Standby Pre-seizure
  • 6. Features – Statistical metrics Anomaly Detection for Seizure Prediction Cynthia Alexander mean channel 11 Normalized distribution std channel 00 Normalized distribution absum channel 02 Normalized distribution Standby Pre-seizure Remarkable differences across patients – i.e. Values for the median across channels Each column one sample (10 min x 16 channels). Medianperchannel andwindow Patient A Patient B Patient C Pre-seizure
  • 7. Image from mriquestions.com Anomaly Detection for Seizure Prediction Cynthia Alexander Features – Correlation and Signal TransformationsAverage correlation of channels over all samples Standby samples vs. all samples Pre-seizure samples vs. all samples Wave decomposition Fourier Transform Each readout can be decomposed into the frequencies that make it up: Gamma (32+ Hz) Beta (16 – 31 Hz) Alpha (8 – 15 Hz) Theta (4 -7 Hz) Delta (< 4 Hz)
  • 8. Image from mriquestions.com Anomaly Detection for Seizure Prediction Cynthia Alexander Features – Correlation and Signal TransformationsAverage correlation of channels over all samples Standby samples vs. all samples Pre-seizure samples vs. all samples Wave decomposition Fourier Transform Each readout can be decomposed into the frequencies that make it up: Gamma (32+ Hz) Beta (16 – 31 Hz) Alpha (8 – 15 Hz) Theta (4 -7 Hz) Delta (< 4 Hz)
  • 9. Anomaly Detection for Seizure Prediction Cynthia Alexander Features – Spectrograms Standby Pre-Seizure Time window Frequency
  • 10. Results Personalized model (train and predict on self over time): Trained on other patient’s data predict on (no self-prior knowledge): Anomaly Detection for Seizure Prediction Cynthia Alexander
  • 11. Anomaly Detection for Seizure Prediction Cynthia Alexander Interactive D3 App – Red shaded areas depict windows in which the model called a pre-seizure sample
  • 12. In development and in order to help users: • Concentrate on personalized models • Feed and predict on-App using real time data • In the long run, aim for preemptive medicine where care is specially provided in high-risk windows (i.e. deep brain stimulation) Anomaly Detection for Seizure Prediction Cynthia Alexander
  • 13. Questions? Reach out! Cynthia Alexander cynth.alexander@gmail.com 4chromat cynthiaalexanderrascon 4chromat.github.io Thank you! Anomaly Detection for Seizure Prediction Cynthia Alexander
  • 14. Anomaly Detection for Seizure Prediction Cynthia Alexander Data cleaning – Signal drop-outs and artifacts!

Editor's Notes

  1. Hello, I am Cynthia and want to talk to you about a model for anomaly detection for seizure prediction 9s
  2. Something really quick on how this works: brain waves are obtained by placing electordes directly on your brain. The motivation to pursue this project is simple, about 1% of adults in the US suffer from epilepsy, and because of it they have seizures on which they might have no control on. Current warning systems rely on self-awareness, so patients could definitely benefit from a prediction model 45’
  3. In my dataset there are recordings over months for a couple of patients, each of this with 16 electrode groups. For the model recordings were split in two categories, standby and pre-seizure (encompassing 1 hour preceding a seizure), and further broke down to 10 minute windows. The goal here is call pre-seizure windows as they happen to warn people to take some precautions. 1:00
  4. Here I show overview on how this model was built: I started from raw brain recordings, and there was a fair amount of data cleaning, and exploratory data analysis. Which eased feature selection and engineering. Then the output of this features was tested on different models, and once the models where selected, there was a bit or parameter tuning. All these allowed me to put together an app showing real data and model predictions. 1:40
  5. About the features, the unit of prediction is a 10 min continuous signal varying in frequency and amplitude. So to come up with features we need to aggregate the signal across electrodes or break it in smaller windows. First I used plain statistical metrics, and it was amusing to see how basic metrics already showed different distribution among classes. 2:10
  6. And a really important take away is that when comparing patients side to side, their recordings are largely different. So it became a clear objective to work on a personalized model, or in the normalization across patients
  7. Another step was to consider the correlation among electrode readouts since it is likely there is not one that points to a seizure but the combination of them In this images each square represents the correlation among two electrodes coded by color, here the contrast of standby and pre-seizure samples to the bulk showed that pre-seizure samples have distinc correlations among electrodes.
  8. A necessary feature to add where signal transformations, specially the fourier transform which I performed over window slices on the samples. This would decompose the readout for each electrode into the frequencies that make it up, which are further classified into delta waves (those present in sleep), alpha (during relaxing times), beta (on a range of acivities) and so on. Higher frequencies would be more comon in pre-seizure samples.
  9. Another feature are spectograms, which describe the frequency and and amplitud over time windows, helping to differentiate the two classes
  10. The model produced decent results for a personalized model, where it was trained and tested on the same person over time. Here Random Forest got top performance, with a precision/recall curve that shows that with a bit over 60% precision it gets over 80% recall, which is great since we care more about true positives than false positives here However for the generalized model, when the model is trained and tested in different persons and even with normalized data, the predictive power drops. So this reflects the contrast about brain waves and possibly also faults in my model.
  11. To better visualize this, I made an app, where the red shaded areas depict windows called as high risk by the personalized model. In this case this first call was a false positive. However down the line there is a true signal, for which a patient could have got about 20 minute warning before a seizure.
  12. Applications of this can be made for any other data that could be encoded in wave form and personalized records, but more importantly to directly benefit users I need to concentrate on personalized models, and evaliate real time data.
  13. Logos d3 tensor fllw
  14. because in comparison to MEG/EEG studies, iEEG tends to be significantly less affected by artifacts, Although EEG is designed to record cerebral activity it also records electrical activitiesarising from sites other than the brain, usually just considered artifacts that may or may not arise from the patients bodyf Mechanical, external, artifacts, could be identified by a fixed form and a repetitive pattern. As well as signals that propagatte across electrodes that are not in proximity. Artifacts caused by muscles, would be of high amplitud and fast activity, and has a distribution that reflects the locations of muscles generating it. Typica muslces artifacts begin and end abruptly. Chewing, for example would make appear as a high amplitud wave across many electode readouts. The large space of artifacts that could exist in a readout add a layer of complexity in the identification of true positive signals, that as in other wave frequency problems is alsways benefitial to identify the source. Remove 50Hz noise