Research Resume

Signal Processing and Classification

Recently there was high research interest in parameter extraction and classification of bio signal. Listed below are of ITB BME’s topics of interest:

  • ECG signal processing and analysis

The output of this research topic is a software that detects irregularities in ECG graph. The software is used as an additional tool for medical experts for further analysis/decision. The ECG data is from the PTB Diagnostic ECG Database including the attached manual diagnostic notes. The research outcomes result in similar diagnostic as the reference and, thus, can adequately describe our heart condition. Interpretation of 12 lead ECG signals follows normal procedure of the signal diagnosis by a medical expert:

  1. General impression
  2. Calibration
  3. Rhythm determination
  4. QRS assessment: R Wave Progression (RWP), QRS Axis, Bundle Branch Block, Fascicular Block
  5. Hypertrophy: Atrial Enlargement, Ventricular Hypertrophy
  6. Ischemia infarction
  • Pulse wave velocity calculation as an indicator of arterial stiffness

PWVPulse Wave Velocity (PWV) parameter is well known to quantify arterial stiffness. The PWV also has been an indicator of vascular damages. A study shows that people with hypertension experience an increasing PWV for about 5 m/s will have 50% increase of mortality risk. Due to its importance, algorithms to derive the PWV parameter from PPG signal are found based on derivative algorithm.

  • Parameter extraction of photoplethysmography signal

Various blood flow parameters can be extracted from photoplethysmography (PPG) signal to indicate the existence of endothelial dysfunction. An early detection of vascular diseases might be derived from the endothelial behavior, namely how much the blood vessels dilate and constrict when blood flows through. The extracted parameters that characterize the shape of the PPG signal are:

  1. Systolic amplitude
  2. Pulse width
  3. Inflection Point Area (IPA)
  4. Pulse interval
  5. Augmentation Index (AI)
  6. Stiffness Index (SI)
  7. Age index
  8. Reactive Hyperemia Index (RHI)
  • Heart Sound Visualization and Classification

HS VisualizationEnvelope detection is an important function to classify heart signal as to locate murmur components. Heart sound to be analyzed should be pre-processed to reduce noise, before the system determines location of S1 and S2 components. Hilbert filter is proposed to calculate the envelope of the heart sound; that is the magnitude of analytic heart signal.
HS ClassificationSome artificial neural network algorithms are studied to differentiate three types of heart sounds: normal, systole murmurs, and diastole murmurs. Comparison between Kohonen’s Self-Organizing Map (SOM), Incremental Self-Organizing Map (ISOM), and Multi-Layer Perception Back Propagation (MLP-BP) suggests that highest accuracy is attained by ISOM.

  • EMG signal processing for assistive technology application

Single channel surface electromyography (sEMG) signal on the biceps is processed to estimate the average of muscle’s force. The processing involves filter, rectification, and calculation of Mean Absolute Value (MAV). The measurement procedure asks users to lift their hand with different loads: empty, 2kg, 3kg, 4kg, and 5kg. The results show that full rectification, instead of half rectification, is better to estimate the force on the muscle as confirmed by linear function of the measured sEMG signals.

  • Wearable gait measurement system based on accelerometer and pressure sensor

Wearable gait measurement system based on accelerometer and pressure sensorGait analysis is developed by using gyroscopes and two-axis accelerometers to measure angle between patient’s leg and ground while the patient walks. The systemic study of human walking involves measurement of ground reaction forces by attaching flexible force sensors beneath insoles of the shoes. The system uses Kalman filter to make better estimation of the angle or gait. To communicate with monitoring system, this wearable gait measurement system involves low-rate wireless personal area networks. This study can be used to compare normal and abnormal gait. It allows measuring gait temporal features, such as stance, swing, single and double support during a walking.

  • Indonesian text-to-speech

Text to speechMore than 3.5 million native Indonesian have visual impairment such that they often find difficulty to read text. Thus, software text-to-speech dedicated to Indonesian language is needed. Indonesian language has unique characteristic, namely strong correlation between sounds and its written form. This enables creation of a syllable sound library with a one-to-one mapping to its written form. To mimic natural human sound, some major improvements have been explored including syllable sound database preprocessing, syllabification method, and playback algorithms to reach human reading speed.

  • Parameter extraction and classification of EEG signal

Image Processing and Classification

  • Cervical cancer early detection

Pap SmearTexture analysis is one of image classification methods to distinguish normal and abnormal pap smear nucleus cell images. The other method, image segmentation, can identify cytoplast area from pap smear cell images. The segmentation is used for classification of cervical cancer by comparing the cytoplast area to the nucleus area. Canny edge detection is also implemented to characterize the nucleus for the same purpose, as an early detection of cervical cancer.

  • Tuberculosis bacteria detection

Tuberculosis (TB) bacteria can be detected through Ziehl-Neelsen staining in sputum smear after color segmentation processing. The wide spectrum of red color from Ziehl-Neelsen staining might confuse the clinicians to locate tuberculosis bacteria. The research, then, is to distinguish between the segment of TB bacteria, non TB bacteria, and background. The proposed method of color segmentation is to deploy Bayesian method from some training data set.

  • Trabecular analysis for osteoporosis study

TrabecularMicro Computed Tomography (uCT) has famously been utilized to study structural properties of bone. Image segmentation is to separate the cortical and trabecular components of the bone. The latter is important source to analyze the density of the bone for osteoporosis study.

  • Segmentation and centerline extraction of coronary arteries

The challenge is to extract only centerline of coronary artery from heart Multi-Slice Computed Tomography (MSCT) scan. The importance of centerline extraction is to measure the diameter of coronary artery for Coronary Artery Disease detection. Morphological operation, Frangi filter, region growing, and multistencils fast marching methods are procedure to achieve the goal.

  • Denoising of MRI Phase Image

MRI PhaseIt has been shown that MRI phase images have important roles in cancer detection, heart motion measurement, brain analysis, and knee assessment. To achieve those roles, complex valued MRI images should be filtered to reduce noise that has significant effect on getting the MRI phase images. The amount of image denoising might be regulated by a threshold function before unwrapping the phase to get the MRI phase images.

  • Coronary angiogram image stabilization for QuBE values calculation

Coronary angiogram image stabilization for QuBE values calculationCalculation of Quantitative Blush Evaluator (QuBE) to calculate myocardial perfusion score are often affected by undesired patient motion during a coronary angiogram procedure. An algorithm to reduce unstable images based on SIFT method is developed by shifting each frame according to the best correlation to the previous frame. Correlation between two consecutive frames is obtained by searching scale-invariant feature by SIFT algorithm during the coronary angiogram. A modified QuBE is acquired by neglecting combinational polygon movement compensation. The stabilization algorithm using SIFT method decreased the deviation by 15% and also the execution time by 71%.

  • Curvature analysis based cerebral aneurysm detection method

Curvature analysis based cerebral aneurysm detection methodThe main purpose of the research is to design a method of cerebral aneurysm detection based on its geometric properties and classify different vascular geometry from normal arteries with different color. The method is developed based on curvature analysis of wall vasculatures model. The curvature shape-index derived from the main curvature of vascular model provides better vascular classification compared to Gaussian and Mean curvature. Moreover, scale-invariant feature of this shape-index easily simplifies scale system for geometric classification.

  • Role of pressure and wall shear stress in cerebral aneurysms

Role of pressure and wall shear stress in cerebral aneurysmsHemodynamic has significant role in the initiation and development of cerebral aneurysm. Wall shear stress and pressure were believed to be the hemodynamic factors that affect the initiation and development of the cerebral aneurysm. By computational fluid dynamic method, the distribution of wall shear stress and pressure around the aneurysm locations are analyzed. This hemodynamic study shows that wall shear stress may have a role in the initiation and development process of aneurysms, yet pressure, in contrast, may not contribute to the initiation of the aneurysms.

  • Diabetic retinopathy early detection

Diabetic retinopathy early detectionThe presence of micro aneurysms in the eye can be an early sign of diabetic retinopathy. Micro aneurysms filter algorithm based on vessel enhancement is developed to extract the structure of micro aneurysms in the retinal image. Gaussian filter is proposed to describe the intensity distribution of vessel profile. The same filter can also detect symptoms of diabetic retinopathy-hemorraghes. Non-Proleferative Diabetic Retinopathy (NPDR) classification, whether it is mild, moderate, or severe, is attained based on the numbers of micro aneurysms and hemorraghes in every quadrant of the retinal images.

  • Image reconstruction

Image ReconstructionProblems raise in 3D reconstruction include how to acquire position and angle of objects from image acquisitions, to transform the position and angle to matrix in 3D plot, to extract the object from background and noise. In our research, we use ultrasound imaging modalities with bone phantom as the objects.

E-health and Telemedicine

  • Smartcard e-health system design

Centralized medical record database is a solution for countries with well-established network infrastructure. However, for countries with high variance of network infrastructure quality, it causes some problems, e.g. what if a hospital is not able to access the centralized database due to internet connectivity problems. Smartcard e-health that stores patients’ medical records and identity becomes solution for tackling that limitation. The requirements of the smartcard e-health system are accurate, secure, and unique to a single individual so that patients are only asked to bring their smartcard when visiting different healthcare centres.

  • Secured e-health smart card

Paperless medical records have attracted some ITB BME’s researchers. Smart cards that are able to record data of patients require a secured system. In this research, application protocol data unit has been embedded to the smart card system.

  • Mobile health system development for childhood illness management

Mobile health system development for childhood illness managementIntegrated data management of childhood illness is important since it can assist policy makers to issue immediate response of real-time condition. In Indonesia, cellular industries have provided services to about 80 percent of the population for 15 years. Health workers need to be mobile to reach patients in remote and rural area. To instantly report the patient condition to health office, a mobile application is developed. The mobile application, then, can reduce childhood illness management time, promote health and childcare and improve the knowledge of mothers, and, eventually, provide ubiquitous health services almost everywhere at any time.

  • Compressive sampling for efficient compression of digital medical video signal

Compressive SamplingThe principal requirement of telemedicine applications is to gain high processing speed and low computing time without significantly degrading the medical video quality, i.e. we need to deploy advanced video compression techniques. The medical test videos are grayscale brain MRI and endoscopy video. Compressive sensing provides the solution as it could improve the conventional sensing results for measurement rate greater than 30%.

Comments are closed.