presentation

Detection of Mechanical Detection of Mechanical Failure in BSCC Artificial Failure in BSCC Artificial

Heart Valve using Heart Valve using EMAT TechniqueEMAT Technique

Materials Assesment Research Group, Department of Electrical and Computer Engineering, Michigan State University, East Lansing, MI 48824.

TeamTeam

Naveen V Nair

Sneha Teresa Thomas

Sridhar Ramakrishnan

Robert Ahn

Ihuma Ugenyi

Pradeep Ramuhalli

Dr Satish S UdpaDr Lalita Udpa

WHAT IS A BSCC HEART WHAT IS A BSCC HEART VALVE?VALVE?

A mechanical prosthetic heart valve that is famous not for an excellent design but for a history of failure .

Of the 86,000 patients who received these valves, four hundred died from a strut fracture, in the first year.

Two hundred additional patients survived similar strut fractures through open-heart surgery

BSCC HEART VALVE

The EMAT TechniqueThe EMAT Technique

The EMAT method is one of the possible ways of detecting SLS in BSCC heart-valves. The method involves use of an electro magnetically generated force to induce acoustic vibrations in the strut.

ProcedureProcedure

A large magnetic field is produced with the help of a permanent magnet. In this field a electromagnet is placed and made to carry an alternating current.

The patient placed in the field of the electromagnet such that the plane of the ring is perpendicular to the alternating magnetic field.

This induces a current in the strut. The permanent magnetic field and the induced current in the strut produce a force on the strut, which causes it to vibrate

AnalysisAnalysis

As for any resonating structure the heart valves also have their own resonant frequencies and the resonance frequencies of the BSCC valve with an SLS is found to be different from that of an IOS (Intact Outlet Strut).

Challenges involvedChallenges involved

The proper excitation is possible only when the ring is properly aligned to the magnetic fields of both the permanent magnetic field and also the electromagnet

It is difficult to produce a permanent magnetic field of high intensity to produce detectable vibrations in the strut

The need of sophisticated instruments to pick up signals generated from the strut without invasion

Contd..Contd..

The signals obtained do not represent signals from the strut alone. The signals from the strut are mixed with signals from the lungs, Gastro intestinal tracts and other parts of the body

EMATEMAT

Signal Processing Algorithms

ObjectivesObjectives

To detect signals from the heart valve at either of the two resonance frequencies (~7.5 kHz or ~2.3 kHz)

Separate these from the other acoustic interference signals. (Potential sources include lungs and GI tract).

Remove ambient noise in measurements

Why a sophisticated Why a sophisticated algorithm? algorithm?

Low Signal to Noise Ratio (SNR)– The acoustic signals picked up by the sensors

are corrupted by noise and other interfering signals.

Frequency overlap– Traditional frequency domain filtering

techniques would not work as the desired acoustic signal and noise have overlapping frequency spectra.

Algorithm - conceptsAlgorithm - concepts

Beam-forming: An algorithm that will enable us to selectively listen to only one source at a particular location and frequency.

Dimensions in cm

Algorithm - conceptsAlgorithm - concepts

Algorithm - conceptsAlgorithm - conceptsSensor Signals

Estimate Number of Sources

(Method: Minimum Description Length)

Estimate Locations and Frequencies of Sources

(Method: Wax Sub-optimal algorithm)

Estimate Desired Source Signal

(Method: Linearly Constrained Minimum Variance Beamforming)

Approximate Locations and Frequencies of Sources

Desired Source Signal

Sub-Optimal Technique - Sub-Optimal Technique - problem modelproblem model

The sources are assumed to be characterized by the following parameters– X coordinate– Y coordinate– Z coordinate– Center frequency

Beamforming conceptBeamforming concept

ARRAY OF SENSORS

Combine sensor signals

optimally

BEAMFORMER ESTIMATE OF THE DESIRED

SOURCE

NOISE / INTERFERING SOURCES

DESIRED SOURCE

Experiments (known location & Experiments (known location & frequency)frequency)

1. Simulated Data

2. Acoustic experiment – in air

3. Acoustic experiment - with human body as the transfer medium

Simulation - BeamformerSimulation - Beamformer

Dimensions in cm

Original Source Signals - timeOriginal Source Signals - time

Original Source Signals - freqOriginal Source Signals - freq

Sensor Signals - timeSensor Signals - time

Sensor Signals - freqSensor Signals - freq

Simulation Results - timeSimulation Results - time

Simulation Results -freqSimulation Results -freq

Using exact source positions

Simulation Results -freqSimulation Results -freq

Using source locations with error

Experiment 2. Acoustic Data –in airExperiment 2. Acoustic Data –in airTest set-up

Sensors

Sources

Source Freq:1. 50002. 60003. 3000

Experiment 2. ResultsExperiment 2. Results

Experiment – 3Experiment – 3

0.5 kHz

7.5 kHz

Experiment 3 - Sensor Signals - freq

Experiment 3- Estimation of location and Experiment 3- Estimation of location and frequencyfrequency

Parameters Source 1 Source 2

Error in X

(in cm)0.05 0.12

Error in Y

(in cm)0.1 0.1

Error in Z

(in cm)0.1 0.12

Error in

Frequency

(in Hz)40 40

ExperimentExperiment 3 Results - freq 3 Results - freq

•Initial results show promise

Future Plans

•More extensive studies on practical data•Optimize algorithm parameters

Conclusions

Blind Source Separation of Acoustic Blind Source Separation of Acoustic Signals from BSCC Heart valveSignals from BSCC Heart valve

Techniques and Issues

Blind Source WHAT ????Blind Source WHAT ????

SeparationConcept of Spatial LocalisationLack of frequency clarity -overlaps in freq

domainUtilise location and frequency information Conduct a sine before the actual test is

performed to determine actual position of the sources

Location and FrequencyLocation and Frequency

Wax’s algorithm Cocktail party problem

– Selective listening

MDL principleOptimisation of the likelihood parameter

Details of the problemDetails of the problem

Consider a set of sources emitting sounds (or any signals) which are band limited and have a single center frequency.

The signal can be written as

– Where the parameters a, w and are slow moving functions of time.

This signal is generated by a number of sources and picked up by a number of sources.

))(*)(2sin()()( tttwtats

Details – cont’dDetails – cont’d

At the sensor what is obtained is a mixture of the signals. – The sensor signal looks like

– Where n is a noise parameter dependent on sensor characterestics only .

– is the time of flight– a is the transfer parameter relating the source signal to

the sensor. This is dependent only on the direction of arrival.

Likelihood parameterLikelihood parameter

The likelihood parameter is found to be dependent only on the first term and is found to be

– In this the l the eigen values of which is the largest eigen value and u are the corresponding eigen vectors

Maximizing this yeilds the optimal set of b() which give the optimal , or direction of arrival

)(.)(

.)()()( 2

bb

ublL

H

iH

is

Troubles and TribulationsTroubles and Tribulations

Heavily dependent on sensor locations How do we get this signal with almost –50dB

SNR out Lack of knowledge of source locations eg. the GI

tract is nowhere near a stationary source infact moves in an approximately gaussian path

Near zero information about the acoustic transfer parameters of the human torso…

Ways around stumbling blocksWays around stumbling blocksSensor location optimisation…

– Maximize a cost function made up by summing energy in the 1.5-3.5 and 6-8 KHz freq range

Require transfer function of body– Make one ????

• Nearly 200 different entities with diff parameters !!!

– Do somekind of ray tracing to arrive at optimum location

Again requires some idea of geometry…

Round and round we go ’round the mulberry bush

Current SolutionCurrent Solution

Use the same cost function but for data, use the Wax algo to determine the source location for a particular sensor location and then use the same algo inverted to find out the signal spectrum for various sensor locations…

Other issues (non signal processing)Other issues (non signal processing) Sensors… ? If u want high sensitivity (like stethoscopes) then

freq range is very low… For these freq ranges no sensors with high enough

sensitivity… Have to get things custom made.. How do we know what

sensitivity we need ?

We rented hydrophones for around $1700 then found out one day before delivery that we couldn’t use them… got lucky this time.

People (sridhar) still working on modifying stethoscopes and so on…

- We don’t

Synchronous data input Synchronous data input

We need to synchronise data received from signal for beamforming to work.– We used a software to split up the left and right channel

of the computer and used MATLAB to generate emulate two sources through each speaker.

– Used atomic clock to sync two machines to get 4 sources.

– Did some more vague stuff.. Split up a microphone using two wires and one way connectors… !!!!

Positions of sensorsPositions of sensors

Measure accurately positions of the sensors– Used Flock of Birds.. – Have to do better

Generating magnetic fieldGenerating magnetic field

Use a superconducting magnet pulled out of an MRI machine.

Incidentally it cupped all our computers

Magnetic noise…Flock of Birds refused to fly

Other methodsOther methods

GradiometerInvasiveUse a magnet to generate current in strut.

Measure magnetic field peturbationUse a catheter

•There must be no barriers to the freedom of inquiry. There is no place for dogma in science. The scientist is free and must be free to ask any questions, to doubt any assertion and to seek any evidence and thus to correct any errors.

- J Robert Oppenhiemer (1904-1967)