Projects
Proposals
Deep Learning
Project Title:

Super-Resolution Fluorescence Microscopy using Deep Learning

Abstract:

Super resolution fluorescence microscopy techniques are an ensemble of light-microscopy techniques which achieve spatial resolution beyond the limitations imposed by the diffraction of light.
However, these techniques are currently limited by low temporal resolution and long acquisition times. In this project, we will investigate an exciting new direction which will combine Deep Learning of both low and high resolution images, to improve both its
temporal and spatial resolution.
The students will get a hands on experience with a research project, combining disciplines in fluorescence microscopy, machine learning and
optimization techniques.

Description:

PDF

Supervisor(s):
Requirements:
Signal and systems, Mavlas, Mavla. Environment:Matlab
Project Title:

Machine Learning methods For Pain Analysis from EEG Recordings

Abstract:

Electroencephalogram (EEG) is a typically non-invasive test used to measure the electrical activity of the brain (i.e. neuronal activations), by placing electrodes along the scalp. Traditional analysis of EEG signals generally focus on measuring the: (i) Magnitudes of neuronal activations within a specific frequency bands. (ii)Time of the response to a stimulus (from the averaged signal called ERP).
The goal of this project is to perform automatic analysis of such signals recorded from patients suffering from neuropathic pain and a control group. In order to do so, we will: (i) Use signal processing tools in order to extract from the raw EEG recordings valuable information (called features). (ii) Apply a machine learning methodologies on the extracted features to separate between the conditions and extract valuable information regarding the origin of the problem.

Description:

PDF

Supervisor(s):
Requirements:
Signal and systems,Mavlas Environment:MATLAB
Project Title:

A Deep Learning approach for image reconstruction from binary pixels

Abstract:

Driven mostly by the booming mobile market, digital photography has advanced staggeringly. However, due to constraints in form factor, image quality in poorly lighten environments is often unsatisfactory.

Recently, Eric Fossum (inventor of CMOS image sensor) proposed a novel concept of an image sensor with dense

sub-diffraction limit one-bit pixels (jots). Newly developed computational photography techniques show great potential for this sort of camera to produce high quality HDR images.

In this project, we will make an attempt to improve upon state-of-the-art reconstruction methods by using advanced deep learning tools. The student will implement the special noise model of these sensors, which will be used to train deep convolutional neural network architectures. Finally, real data collected from these camera will be used for testing the quality of reconstruction.




Description:

PDF

Supervisor(s):
Requirements:
Image Processing' Machine Learning Environment: Python (Tensorflow)
Biomedical
Project Title:

Deep Learning MRI

Abstract:

Deep leaning is a machine learning approach that has gained very much attention recently. It has been proven to be very successful in many real world problems, such as object detection and face recognition, and it is currently widely used in many search engines. Recentness, deep learning has been adopted also by the world of medical imaging. While Magnetic Resonance Imaging (MRI) is the method of choice for diagnosis, evaluation and follow-up of brain clinical pathologies, the bottleneck in many MRI applications is image reconstruction from undersampled data. The goal of the project is to use deep learning approaches for fast MRI reconstruction (unlike the common problems of deep learning, which are mostly classification problems). The project is expected to significantly improve many MRI applications, and speed-up the clinical workflow.



Description:

PDF

Supervisor(s):
Requirements:
Project Title:

Super-resolution ultrasound scans of moving microvasculature

Abstract:

A new method developed in the SAMPL lab has achieved sub-diffraction resolution in contrast enhanced ultrasound scans with a very fast acquisition rate. Such a method allows for real-time scanning of the microvasculature of moving organs such as the heart or kidney, which until recently was considered impossible.
In this project, we will investigate the application of the method on real clinical data to see if such a method can lead to new diagnosis of patients.
This project is performed as part of a collaboration with Ichilov hospital.


Description:

PDF

Supervisor(s):
Requirements:
Signal and systems, Mavlas Environment: Matlab
Project Title:

Tensor Based Reconstruction for Computed Tomography

Abstract:

Medical CT imaging is probably the most widely used imaging tool in-use today. CT images suffer from different artifacts and issues, and generally require radiating a patient with a big dose of X-Ray radiation in order to produce good images.

In this project, we will learn on new tensor based algebraic techniques, and use them for solving difficult CT reconstructions, made with low radiation dosages. Our goal would be to lower the radiation dosage, while improving the CT images, using tensor based methods that can exploit the inner structure of the human body.

The students will first learn on tomographic imaging, and will get familiar with the required background in advanced signal processing and tensor decompositions. We will then try and formulate a specific tensor based solution for the CT problem. We will work with state-of-the-art toolboxes, useful for many other problems, from the world of machine learning.


Description:

PDF

Supervisor(s):
Requirements:
Signal and systems, Mavlas Environment: MATLAB
Project Title:

CT Super-Resolution by Non-Uniform Detector Arrays

Abstract:

Medical CT imaging is probably the most widely used imaging tool in-use today. The biomedical companies constantly work to improve and get better resolution for the CT images.

In this project, we will learn and use sophisticated tools from signal processing, such as sub-Nyquist sampling, array processing, and super-resolution algorithms. Using these tools, we will create enhanced CT images, with higher resolution than the native scanner’s resolution.

We will collaborate with General Electric Healthcare® to examine new ways to define the CT detector array in order to digitally create cleaner and sharper CT images. Doctors from Rambam hospital would also be happy to provide us with clinical feedback.

The students will first learn on tomographic imaging, and will get familiar with the required background in advanced signal processing. We will then design and build a simulator that can enhance the resolution in CT images.


Description:

PDF

Supervisor(s):
Requirements:
Signal and systems, Mavlas Environment: MATLAB
Project Title:

Reducing Metal Artifacts in CT

Abstract:

Metal artifacts in CT scans are a major problem for radiologists today. Whenever a patient has a metallic object in his body (dental fills, platinum bolts or knife wounds for instance), the CT image gets completely ruined, as seen in the image below.

We suggest to use advanced signal processing tools, such as sparse representations, dictionary learning and others, in order to exploit the structure of the CT scans to our advantage. By smart modelling of the metal artifacts, we hope to reduce or even remove them altogether from the output scans.

Doctors from both Rambam and Ichilov hospitals have shown tremendous interest in an algorithm that can reduce these artifacts, and we will be working in collaboration with them. Our goal is to ultimately include our solution on a real CT machine, within the hospital. Real experiments can be conducted in General Electric, CT department.

In this project the students will learn state of the art methods in signal processing and sampling, and will get familiar with principles of tomography.



Description:

PDF

Supervisor(s):
Requirements:
Signal and systems, Mavlas Environment: MATLAB
Project Title:

Limited Angle Reconstruction for Computed Tomography

Abstract:

Medical CT imaging is probably the most widely used imaging tool in-use today. In this project, we consider the problem of scanning from only a limited angle of projections (less than 180o).

Successfully reconstructing from a limited projection angle would allow us to install small CT scanners in small clinics, and maybe even on-board ambulances. However, limited angle reconstruction is a tough mathematical problem, and we will attempt to solve it using recently developed, signal processing tools.

In this project, we will learn and use sophisticated tools from signal processing, such as sub-Nyquist sampling, algebraic reconstruction techniques and new optimization based solvers.

The students will first learn on tomographic imaging, and will get familiar with the required background in advanced signal processing. We will then design and build a simulator that can reconstruct images from limited scanning angles, using the structure of the tomographic scans.


Description:

PDF

Supervisor(s):
Requirements:
Signal and systems, Mavlas Environment: MATLAB
Project Title:

Super Resolution of Ultrasound Doppler Imaging

Abstract:

Doppler ultrasound is a non-invasive and safe modality that is used for the estimation of blood velocities by transmitting high-frequency sound waves (ultrasound) and analyzing the signals reflected from circulating red blood cells. Doppler scans help diagnose many conditions, including: heart valve defects and congenital heart disease, artery occlusions and aneurysms.
In order to allow the medical doctor to navigate and choose the region in which the blood velocity is to be estimated, a B-mode image of the tissue is generated first. Only then, a velocity map of the blood is imaged on top ( See image above). Nowdays, 2 interleaved transmission sequences are used, one for Doopler and one for B-mode imaging, which results in degraded B-mode image and corrupted velocity estimation. In this project, we will perofrm the two imaging, B-mode and Doppler, simulatenously by using only Doppler transmission and applying super resolution on the received signal in order to yield a high resolution B-mode image. The project will be perfomred using Matlab and state of the art research oriented ultrasound system.


Description:

PDF

Supervisor(s):
Requirements:
Matlab, Introduction to Digital Signal Processing (044198)
Project Title:

Coded Excitation for diverging wave Imaging

Abstract:

Medical ultrasound is used for tissue visualization by radiating it with acoustic energy transmitted by an array of elements. Novel imaging method based on insonification with diverging improves image quality and acquisition time, however, it is limited by data transfer rates and severe computational load. In addition, modern imaging systems use single-carrier short pulses for transducer excitation, while the usage of more sophisticated signals can be beneficial in terms of SNR, penetration depth and frame rate. Therefore, coded signals are extremely valuable in medical ultrasound imaging although their implementation is challenging due to imaging and not detection nature of ultrasound, its high dynamic range and frequency dependent attenuation. Recently, diverging waves imaging was implemented using Fourier domain beamforming developed in SAMPL, leading to significant reduction in both sampling and processing rates. In this project we aim to extend this work by developing Fourier-based coded ultrasound for diverging waves imaging. The combination of these novel techniques will pave a way to enhanced SNR, real-time ultrafast ultrasound system. Performance of the developed method will be tested on a Verasoincs (research-oriented) ultrasound system Matlab environment. Required background: Introduction to Digital Signal Processing (044198)



Description:

PDF

Supervisor(s):

Regev Cohen, Tanya Chernyakova

Requirements:
Introduction to Digital Signal Processing(044198)
Project Title:

Super Resolution of Ultrasound Doppler Imaging

Abstract:

Doppler ultrasound is a non-invasive and safe modality that is used for the estimation of blood velocities by transmitting high-frequency sound waves (ultrasound) and analyzing the signals reflected from circulating red blood cells. Doppler scans help diagnose many conditions, including: heart valve defects and congenital heart disease, artery occlusions and aneurysms.

In order to allow the medical doctor to navigate and choose the region in which the blood velocity is to be estimated, a B-mode image of the tissue is generated first. Only then, a velocity map of the blood is imaged on top ( See image above). Nowdays, 2 interleaved transmission sequences are used, one for Doopler and one for B-mode imaging, which results in degraded B-mode image and corrupted velocity estimation. In this project, we will perofrm the two imaging, B-mode and Doppler, simulatenously by using only Doppler transmission and applying super resolution on the received signal in order to yield a high resolution B-mode image. The project will be perfomred using Matlab and state of the art research oriented ultrasound system.


Project description:

דופלר אולטרסאונד היא שיטת הדמיה לא פולשנית ובטוחה שמשמשת למדידה של מהירות הדם באמצעות משלוח של גלי קול בתדרים גבוהים וניתוח של ההחזר המתקבל מתאי הדם הנעים בזרם הדם. סריקות דופלר מאפשרות אבחון של מגוון מחלות כגון פגמים בשסתומי הלב, מחלות לב מולדות חסימות של כלי דם ומפרצות.

במטרה לאפשר לרופא לבחור את האזור בו יתבצע שערוך המהירות, יש לבצע ראשית שידור להדמיה בגווני אפור של הרקמה ורק לאחר מכן מתבצע שידור לשכבה נוספת של הדמיית המהירות (כפי שמופיע בתמונה למטה). כיום שני השידורים השונים מתבצעים לסירוגין, דבר הפוגע הן באיכות התמונה והן באיכות שערוך המהירות.

בפרויקט זה נשלב בין ההדמיות השונות בו-זמנית על די שידור להדמיית מהירות בלבד וביצוע סופר-רזולוציה על האות המתקבל לקבלת תמונת גווני אפור באיכות גבוהה. הפרויקט יתבצע ב-Matlabוכן ישתמש בסורק אולטרסאונד מחקרי חדיש.

Description:

PDF

Supervisor(s):
Requirements:
Matlab, Introduction to Digital Signal Processing (044198)
Project Title:

Functional MRI via temporal mixing

Abstract:

Functional Magnetic Resonance Imaging (fMRI) is the method of choice for evaluation and analysis of brain activity by detecting changes associated with blood flow. However, current acquisition of a routine brain fMRI suffers from low spatial and temporal resolution. As such, it causes many difficulties, such as low-quality maps of brain activity regions and brain networks.


Project description:

Functional Magnetic Resonance Imaging (fMRI) is the method of choice for evaluation and analysis of brain activity by detecting changes associated with blood flow. However, current acquisition of a routine brain fMRI suffers from low spatial and temporal resolution. As such, it causes many difficulties, such as low-quality maps of brain activity regions and brain networks.

Description:

PDF

Supervisor(s):
Requirements:
Project Title:

Sub-Nyquist Vector Tissue Doppler Ultrasound System - Part 2

Abstract:

Tissue Doppler ultrasound imaging (TDI) enables the estimation of cardiac function by transmitting streams of pulses in a certain direction and estimating the velocity of the tissue from the phase shifts of the returning echoes. Recently, Xampling and Compressed Sensing methods developed at SAMPL were applied to TDI, allowing the reconstruction of signals sampled at a sub-Nyquist rate with reduced number of pulses per velocity estimation, using priors on the sparsity of the signal. This reduced sampling was performed without compromising the temporal and spatial resolutions.

Currently, TDI imaging enables the estimation of the component of cardiac velocity parallel to the transmitted ultrasonic beam. Therefore, TDI measurements depends on the scan angle and thus suffer from high variability. Several methods have been developed in the last few years for the estimation of the full flow field including the transverse component, collectively known as vector Doppler imaging methods.

In the first part of this project, estimators for the full 2d velocity field were invastigated and implemented. In this project, sub-Nyquist TDI system incorporating vector Doppler imaging will be implemented using new programmable research ultrasound platform. The system developed in this project will pave the way for future high-end portable ultrasound systems enabling reliable quantification of cardiac function.

In this project you will become familiar with Xampling and Compressed Sensing theory and novel ultrasonic signal processing methods. In addition, you will gain experience in implementing signal processing algorithms on real ultrasound systems.

Description:

PDF

Supervisor(s):
Requirements:
Intro to digital signal processing (044198)
Radar
Project Title:

Sub-Nyquist Space-Time Adaptive Processing (STAP)

Abstract:

A radar mounted on an airborne platform transmits short pulses and collects the returned echo from the targets on the ground. The received signal contains the returns from the targets-of-interest as well as the illuminated ground surface. The latter is undesirable interference commonly termed as clutter. For a ground-based radar, the clutter is stationary and occupies the low-frequency part of the signal spectrum. It can, therefore, be easily filtered using a notch filter at low frequencies. However, this filtering method cannot be extended to airborne radars as the relative motion of the aircraft spread the clutter in all directions and Doppler frequencies.

Project description:

Space-Time Adaptive Processing (STAP) techniques have been proposed to filter the radar signal in both the angular and Doppler domains. The goal of this project is to formulate and devise STAP algorithms for sub-Nyquist radars. We will make use of the existing theoretical advances to integrate STAP in the sub-Nyquist processing framework.


Description:

PDF

Supervisor(s):
Requirements:
Signal and systems (essential), Mavlas (essential), Random signals (desirable). MATLAB.
Project Title:

Cognitive Foliage Penetration Radar (FPR)

Abstract:

The objective of this project is to improve detection of low Doppler targets in the ground-based Foliage Penetration Radar. The native radar system transmits a train of pulses and suffers from interference in the UHF band. We propose cognitive transmission to mitigate the radio-frequency interference and sub-Nyquist sampling to enhance the Doppler resolution to detect slowly moving targets. We focus on the transmission of sparse stepped-frequency continuous waveform (s-SFW) to enable the cognition and then develop sub-Nyquist processing for such a waveform.

Project description:

A foliage penetration (FOPEN) radar detects, tracks and images targets such as human intruders in a dense forest environment. The radar transmission should be able to penetrate the foliage without significant signal attenuation and the received signal should have sufficient backscatter from targets (vehicles and humans) to achieve good detection performance. In order to meet these criteria, often VHF and UHF frequencies are considered suitable candidates.


Description:

PDF

Supervisor(s):
Requirements:
Signal and systems (essential), Mavlas (essential), Random signals (desirable). MATLAB.
Project Title:

Spectrum Sharing Radar Prototype

Abstract:

With the advent of mobile communications, the microwave spectrum is becoming increasingly dense. Several surveillance radar systems have traditionally operated in L-, S-, and C-bands for decades. However, their operation now faces challenges due to spectrum scarcity and RF interference from communication networks. Recently, there has been considerable interest in designing radar systems that have the ability to share spectrum with communication networks. Programs such as DARPA SSPARC, NSF EARS, CommRad, and RC3S are also actively working on equipping existing radar systems with spectrum-sharing transmission and reception. This design methodology has been termed as ‘coexistence’ spectrum sharing.


Project description:

The goal of this project is to develop a hardware prototype of a coexistence spectrum-sharing radar. We would leverage on existing hardware prototypes and designs of cognitive radio and cognitive radar in the design process. The cognitive transmission allows the system to only transmit in those frequency bands that are sampled by the receiver.

Description:

PDF

Supervisor(s):
Requirements:
Signal and systems (essential), Mavlas (essential), Random signals (desirable). MATLAB.
Communication
Project Title:

Compressed DOA Estimation for Millimeter Wave MIMO system

Abstract:

In the race toward increasing data rates at cellular networks, mmWave MIMO (Multiple Input Multiple Output) systems are considered a leading candidate for 5G standard, the next generation of cellular technology.

Using mmWave offers multiple advantages, such as channel bandwidths far greater than previously available and larger antenna arrays, but also arouse new difficulties such as expensive RF chains, and massive amount of data need to be processed digitally.

This difficulties raise the need to come up with a new solution that will consider both the analog and digital domains.

To reduce the necessary amount of expensive RF chains we aim to exploit the sparse mmWave channel model.

The goal of the project is to develop a new sensing scheme for the mmWave model along with an algorithm for the DOA estimation.

The project will include research next to matlab implementation


Description:

PDF

Supervisor(s):
Requirements:
Introduction to Digital Signal Processing (044198)
Project Title:

Smart Sensing for Off-Grid Error Reduction in Compressed Sensing Algorithms

Abstract:

Compressed Sensing is a novel family of algorithms that make use of a sparse structure of the signal to perform an effective recovery. This kind of algorithms make use of a "sensing matrix" to relate the measurements to the original signal. In many cases this sensing matrix is a grid matrix that take a continuous parameter on which the signal depends, such as frequency or angle, and discretizes it. However, in this case, a problem arise when the true parameter from which the signal was generated is "off the grid".

The project goal is to design a smart sensing matrix that reduce the possible implications of the off grid problem.

The project will include research next to matlab and will contain algorithmic aspect.


Description:

PDF

Supervisor(s):
Requirements:
Introduction to Digital Signal Processing (044198)
Project Title:

Pilot Contamination Mitigation Using Graph Theory

Abstract:

The first step in every communication scheme is to estimate the channel between the user and the base station that serves him. This can be done by letting the user send a known sequence to the base station, called "pilot sequence", and then let the base station estimate the channel from the received signal. Pilot contamination is a phenomenon in multi-user multi-cell wireless communication systems, where users from different cells interfere with each other channel estimation. The result is that the estimated channel is contaminated by interfering channels.

We wish to use tools from graph theory to solve the pilot contamination problem. The pilot contamination problem can be made similar to a clustering problem on graphs.

The goal of the project is to develop define the relevant graph for the problem and try different algorithms to solve it.

The project will include research next to matlab implementation

Description:

PDF

Supervisor(s):
Requirements:
Introduction to Digital Signal Processing (044198), Design and Analysis of Algorithms (046002)
Project Title:

Pilot Contamination Mitigation for Multi Antenna Users

Abstract:

The first step in every communication scheme is to estimate the channel between the user and the base station that serves him. This can be done by letting the user send a known sequence to the base station, called "pilot sequence", and then let the base station estimate the channel from the received signal. Pilot contamination is a phenomenon in multi-user multi-cell wireless communication systems, where users from different cells interfere with each other channel estimation. The result is that the estimated channel is contaminated by interfering channels. To mitigate this problem, one can design the pilot sequences of the different users to minimize the MMSE of the channel estimation. Prior work solved this problem for a single antenna users. It is now of interest to generalize this method for multi antenna users.

The goal of the project is to develop the theoretical solution for the above problem in the multi antenna case, and back the results with numerical simulations.

The project will include research next to Matlab implementation

The student will acquire advanced tools in signal processing and be exposed to new application in communication wireless networks.

Description:

PDF

Supervisor(s):
Requirements:
Introduction to Digital Signal Processing (044198)
Project Title:

Sub-Nyquist MWC Miniaturization

Abstract:

In light of the ever-increasing demand for new spectral bands and the underutilization of those already allocated, the new concept of Cognitive Radio (CR) has emerged. Opportunistic users could exploit temporarily vacant bands after detecting the absence of activity of their owners. A CR deals with wideband signals, with high Nyquist rate. To overcome the sampling rate bottleneck, a sub-Nyquist sampling and reconstruction prototype has been developed in the SAMPL lab, the modulated wideband converter (MWC).

One of the major challenges today is the adaptation of the MWC research platform to real world devices, such as laptops, cellular phones and other personal communication devices. The goal of this project is to use cutting edge technologies in order to implement the MWC system on a small scale factor, effectively miniaturizing the system by first designing and then implementing unique engineering solutions.

The students will learn and use advanced signal processing tools from the world of sub-sampling, and will gain a lot of hands-on experience on both software and hardware development.

We are confident the tools acquired will be of great use to the students in the future, in whichever field they would like to specialize.

Description:

PDF

Supervisor(s):
Requirements:
Signal and systems, Mavlas (can take during the project), Logic Design and/or VHDL background
Optics
Project Title:

Super-Resolution Light-sheet Fluorescence Microscopy using dictionary learning

Abstract:

Super resolution fluorescence microscopy techniques are an ensemble of light-microscopy techniques which achieve spatial resolution beyond the limitations imposed by the diffraction of light. However, these techniques are currently limited by low temporal resolution and long acquisition times. Light-sheet microscopy enables fast 3D volumetric imaging of living specimens, but with degraded spatial resolution, when using large field of views (FOVs). On the other hand, high resolution images can be acquired by considering a smaller FOV.



Description:

PDF

Supervisor(s):
Requirements:
Signal and systems, Mavlas, Mavla. Matlab.
Project Title:

Faster selective plane illumination microscopy (SPIM)

Abstract:

Super resolution fluorescence microscopy techniques are an ensemble of light-microscopy techniques which achieve spatial resolution beyond the limitations imposed by the diffraction of light. However, these techniques are currently limited by low temporal resolution and long acquisition times. Light-sheet microscopy enables fast 3D volumetric imaging of living specimens, but with degraded spatial resolution. Selective plane illumination microscopy allows for faster acquisition rates by changing the measurement process from Cartesian scanning to cylindrical scanning.




Description:

PDF

Supervisor(s):
Requirements:
Signal and systems, Mavlas, Mavla. Environment:Matlab
Project Title:

Super-Resolution Light-sheet Fluorescence Microscopy

Abstract:

Super resolution fluorescence microscopy techniques are an ensemble of light‐microscopy techniques which achieve spatial resolution beyond the limitations imposed by the diffraction of light. However, these techniques are currently limited by low temporal resolution and long acquisition times. Super‐resolution optical fluctuation imaging (SOFI) is a fluorescence microscopy technique enabling sub‐diffraction limit imaging with high temporal resolution by calculating high order statistics of the fluctuating optical signal. On the other hand, Light‐sheet microscopy enables fast 3D volumetric imaging of living specimens, but with degraded spatial resolution.

Project description:

In this project, we will investigate an exciting new direction which will combine the SOFI technique for sub‐diffraction imaging and light‐sheet microscopy, which potentially may lead to a new type of fast super‐resolution microscope for living specimen. The students will get a hands on experience with a research project, combining disciplines in fluorescence microscopy, sparse representations and optimization techniques.


Description:

PDF

Supervisor(s):
Requirements:
Signal and systems, Mavlas, Mavla.
Medical Imaging
Project Title:

Low resolution patch based denoising of high resolution images

Abstract:

Denoising is a fundamental problem in image processing. Many algorithms have been suggested for image denoising, the vast majority are not application specific, but are rather generic.
Often, it is possible to acquire two datasets of images – one is a low resolution image with high SNR, and the other is a high resolution image.
In this project, we propose to sue the high SNR, low resolution images to denoise the
high resolution, low SNR images, and to achieve better performance over contemporary denoising algorithms.
The project can have immediate applications, such as denoisisng of biomedical images.



Description:

PDF

Supervisor(s):
Requirements:
Signal and systems, Mavlas Environment: MATLAB
INTERIA Web Design & Development