bp/QMUL_Final_Project
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Stuff

Browse Source
This commit is contained in:
Sam Perry
2017-08-07 15:18:52 +01:00
parent 7f673752df
commit e144f7079f
1 changed files with 57 additions and 109 deletions
Download Patch File Download Diff File Expand all files Collapse all files
Project_Writeup.tex
+57 -109
View File
@@ -143,24 +143,24 @@ section~\ref{performance}
\subsection{Signal Preprocessing}
There are a large number of factors that lead to variation in quality of PCG
recordings: stethoscope type, make and model, its microphone/sensors used for
recording of the data, the position used to record (i.e.\ lower left sternal
border, apex, pulmonic area, aortic area), built in filters/signal processing
used by the stethoscope (i.e.\ noise filters, anti-tremor filters), medication that
a patient may be taking, as well as many other factors that may influence the
recorded signal~\parencite[p.4]{Pavlopoulos2004}. This presents a significant
issue when attempting to analyse and compare a dataset of signals, as
variations in recordings and artefacts caused by factors other than heart
sounds will most likely interfere with analysis and comparison methods. To
account for this, pre-processing methods are widely used, aiming to standardize
a dataset. This is also used as a way to accentuate features of the data that
are expected to be relevant during classification.\\
recordings: stethoscope type, make and model, its microphone/sensors, the
position used to record (i.e.\ lower left sternal border, apex, pulmonic area,
aortic area), built in filters/signal processing used by the stethoscope (i.e.\
noise filters, anti-tremor filters), medication that a patient may be taking,
as well as many other factors that may influence the recorded
signal~\parencite[p.4]{Pavlopoulos2004}. This presents a significant issue when
attempting to analyse and compare a dataset of signals, as variations in
recordings and artefacts caused by factors other than heart sounds will most
likely interfere with analysis and comparison methods. To account for this,
pre-processing methods are widely used, aiming to standardize a dataset. This
is also used as a way to accentuate features of the data that are expected to
be relevant for classification.\\
A common method employed is the use of decimation and a static filter to remove
unwanted spectral content that is most likely noise~\parencite{Liang1997a,
Homsi2016, Springer2016, Gupta2007}. This helps reduce higher frequency noise
such as speech, microphone movement, breething and other interference caused
externally. Decimation tends to downsample to around 1--4KHz, with
externally. Signals are commonly downsampled to around 1--4KHz, with
anti-aliasing filter specifications varying across the literature. Generally,
highpass chebychev or butterworth filters are favoured with cutoff frequencies
ranging from 400--750Hz.\\
@@ -224,9 +224,9 @@ A variety of machine learning methods have been implemented with reasonable
success. Gupta et.\ al present a method that applies $k$-means clustering to
replace standard threshold based methods for determining peak classification in
a standard envelope based segmentation algorithm~\citeyearpar{Gupta2007}. This achieved a reported
accuracy of 90.29\%. Due to the standard envelope based method for feature
extraction, this method is still suceptible to noise and artefacts that occur
within the frequency bands of the heart sounds.\\
accuracy of 90.29\%. Due to the envelope based method for feature extraction,
this method is still suceptible to noise and artefacts that occur within the
frequency bands of the heart sounds.\\
Sepehri et.\ al propose a method that combines neural networks with Power
Spectral Density (PSD) estimates~\citeyearpar{Sepehri2010}. This method
@@ -236,7 +236,7 @@ other sounds and murmurs. This method achieves a reported 93.6\% accuracy on a
significantly larger database than other methods detailed.\\
Most significant success in segmentation algorithms has been observed through use
of probabilistic Models such as Hidden Markov Models (HMMs). Early research
of probabilistic models such as Hidden Markov Models (HMMs). Early research
using these models by Ricke et.\ al utilized embedded HMMs to model the 4
states of the PCG and their transitions~\citeyearpar{Ricke2005}. MFCCs and
Shannon Energy are used as feature vectors for the models. Results of
@@ -253,7 +253,7 @@ of a duration-dependent hidden Markov (DHMM)~\citeyearpar{Schmidt2015}. The
DHMM is a modified HMM that considers the duration of the current state when
calculating the probability of transition to another state. This modification
scored a reported sensitivity of 98.8\% and a positive predictivity of
98.8\%.\\
98.6\%.\\
Building on previous work using HMMs, Springer et.\ al presents a segmentation
algorithm by using hidden semi-markov models (HSMMs) in combination with
logistic regression~\citeyearpar{Springer2016}. Use of Hidden semi markov model
@@ -262,8 +262,8 @@ in probability calculation of the subsequent state. In this case, the knowlege
that there is an upper and lower limit on the duration of each component is
used in calculation of transition probabilities. A modified viterbi algorithm
is then used to calculate the most likely set of transitions based on observed
features. Logistic regression is then used to improve discrimination between
state features when compared to discriminatory methods used by previous work.
features. Logistic regression is used to improve discrimination between state
features when compared to discriminatory methods used by previous work.
Performance was evaluated on a significantly larger database than previous
methods and achieved a reported accuracy of $95.63\% \pm 0.85\%$. Due to it's
rigorous evaluation and high accuracy, this method is currently considered the
@@ -306,100 +306,48 @@ Gupta et.\ al \citeyearpar{Gupta2007} & Homomorphic filtering, $k$-means clus
\subsection{Classification Models}
A wide variety of methods exist for the extraction of statistical
features from PCG data. These features are used for the creation of
robust, meaningful representations of the data.\\
The use of spectral representations for PCG data are prominent in the
literature. The ability to separate activity across the frequency
spectrum reveals patterns that may not be attainable by analysing the
time domain signal alone.\\
Due to the need for low frequency analysis and the high noise levels
found in PCG signals, it has been found that the traditional FFT
method for extracting spectral information may not be
suitable~\parencite{Akay1990}. For this reason, parametric methods for
spectral estimation have been a popular choice for extraction of such information.
Methods such as AR, ARMA, AR-HOS and MUSIC have been shown to provide spectral
representations suitable for analysis and classification of heart
sound~\parencite{Ergen2001, Schmidt2015}.\\
Other methods such as Wavelet Decomposition and MFCCs have also been
successfully employed for extracting spectral data for purposes such
as heart valve disease identification and heart murmur
detection~\parencite{Quiceno-Manrique2010a, Maglogiannis2009}.\\
A wide variety of methods exist for the extraction of statistical features and
classification of PCG data. Most notably, the recent Physionet/Computing in
Cardiology Challenge 2016 has prompted the development of a range of methods
that have improved the quality of abnormality classification in noisy signals.
The challenge was assembled to provide researchers with a large database of PCG
signals of varying quality. This enabled the development of algorithms that
could be evaluated on a significant database, in order to determine performance
across a range of conditions/signal qualities~\parencite{Clifford2016}. This
section first details significant work produced prior to the challenge, and
then highlights key works produced for the challenge to outline the breadth of
methods for robust heart sound analysis.
In addition to direct analysis, the ability to segment and extract RR values
from the signal allows for their statistical analysis, both in the time and
frequency domain, for use as features.\\
- Basic physionet challenge features
Dash et al.\ use a number of time-based statistical analysis on the RR
time series for the detection of atrial fibrillation. Statistical
analyses such as RMSSD, Shannon Entropy and Turning-point Ratio are
used as feature vectors for classification of
signals~\citeyearpar{Dash2009}. A similar approach is used by Yaghouby
et al.\ for the generalized classification of heart abnormality. Here,
a selection of linear and non-linear features are used for
classification with promising results~\citeyearpar{Yaghouby2009}.\\
Frequency domain analysis of RR values are also used by calculating the
PSD of the RR values via approaches such as VFCDM.\ This form of
approach allows for higher resolution time-frequency representations of
the RR data than approaches such as the FFT or wavelet transform~\parencite{Wang2006}.
From a spectral representations such as this, Yaghouby et al.\
demonstrate the use of such descriptors for the discrimination between
sympathetic and parasympathetic contents of the signal, not directly
detectable through time domain analysis~\citeyearpar{Yaghouby2009}.\\
Further in-depth analysis of statistical features for HRV can be found
in~\parencite{Electrophysiology1996}
\subsubsection{Work prior to the Physionet challenge}
Work prior to the Physionet challenge was conducted predominantly with the aim
of classifying specific heart conditions. Until recently, little research had
been produced with regards to general abnormality detection, with many projects
choosing to focus on specific conditions such as murmurs, atrial fibrillation
and flutter, and heart valve disease. This section outlines some key research
into these areas, alongside initial research into general abnormality
detection.
% TODO: Revise to include physionet entries
% TODO: Add section for parameter optimization/feature selection methods
Classification of signals for diagnostic purposes. The aim being to
distinguish healthy signals from those with certain heart
conditions/abnormality. This is most commonly achieved by extracting
sets of features vectors from PCG signals, followed by their
classification, most commonly using machine learning algorithms for
automatic classification. The features extracted and classification
algorithms applied vary across the literature based on factors such as
the diagnostic aims of the classification and computing performance
requirements.\\
- SVM classifier for heart valve diseas~\parencite{Maglogiannis2009}
- Threshold classifier for atrial fibrillation and flutter~\parencite{Dash2009}
- k-NN Classifier for murmur detection~\parencite{Quiceno-Manrique2010a}
- Feature analysis specifically for coronary artery
diseas~\parencite{Schmidt2015}
- GDA and MLP Neural-net classification of general abnormalities~\parencite{Yaghouby2009}
- SVM, k-NN and Bayesian classifier of general abnormalities~\parencite{Lubaib2016}
Artificial neural networks and support vector machines have proven to
be popular choices for classification. Much success has been seen in
employing these machine learning techniques for classification across
both PCG and ECG data for conditions such as chronic heart failure,
atrial fibrillation and flutter, diastolic murmurs, and for general
pathology detection~\parencite{Cathers1995, Wu1995, Bung2000,
Lubaib2016, Maji2014, Ari2010, Maglogiannis2009}. Results do vary based
on the combination of features and exact classification methods used.
However, encouraging results are presented with highly accurate
classifications for general abnormality detection and for more specific
pathological condition detection.\\
However, there is a lack of research into other machine learning
techniques such as bayesian classification~\parencite{Lubaib2016},
$k$-Nearest Neighbour~\parencite{Quiceno-Manrique2010a, Lubaib2016} and
Linear Regression~\parencite{Orhan2013}. Studies that utilize these
methods for classification have generated promising results. There is
therefore the potential for further research into exploiting the
benefits of these techniques for heart abnormality detection.\\
The selection of features used for classification also depends
predominantly on the aims for the classification. For general
abnormality classification, spectral representations such as wavelet
transformations, VFCMD, FFTs and MFCCs are a popular
choice~\parencite{Bung2000, Wu1995, Yaghouby2009, Dash2009}. Their
multi-dimensional representation of the data reveals details in the
signal that cannot be seen through a 1 dimensional time series alone,
allowing for more accurate classification. Higher-level statistical
methods are also widely used for both time and spectral
representations~\parencite{Bung2000, Quiceno-Manrique2010a,
Schmidt2015, Dash2009, Yaghouby2009}. These allow for the
classification based on more specific statistical properties of the
data. It is highlighted by Orhan that Higher level statistical methods
may add considerable complexity to computations, and so care should be
taken, particularly when considering systems in a real-time
context~\citeyearpar{Orhan2013}.
\subsubsection{Physionet challenge entries}
scoring method
- Benchmark classifier~\parencite{Liu2016}
- 100+ features and nested ensemble classifiers~\parencite{Homsi2016}
- Rnage of features using Adaboost classifier~\parencite{Potes2016}
- Ensemble of NNs, bootstrapping, range of features~\parencite{Zabihi2016}
- Classification through probability based methods~\parencite{Plesinger2017}
- Wavelet, MFCC and inter-beat neural network classifier~\parencite{Kay2017}
- Large number of features, tensor based feature reduction and
K-NN~\parencite{Bobillo2016}
- Convolutional neural networks, MFCCs~\parencite{Rubin2016}
\subsection{System Performance}\label{performance}
\newgeometry{margin=1cm} % modify this if you need even more space
\begin{table}[htbp]
\captionof{table}{Summary of research prior to the Physionet Challenge 2016} \label{PriorWorkTable}