Vacancies

RetinaCheck: Blindness prevention and early diabetes detection by large-scale retina screening

Diabetes
Diabetes is reaching epidemic proportions worldwide, especially in Asia due to fast lifestyle changes and genetic factors. It is estimated that 11.6 % of the Chinese population had diabetes-2 in 2014, the most of any country. Diabetic retinopathy is the main cause of newly formed blindness in the working population, leading to high societal costs. Visual inspection through fundoscopy of the retina is easy, low cost and highly informative. Early detection is the key to prevention and successful treatment of these forms of blindness. Losing sight is an incredible reduction in quality of life. However, many cases still go unnoticed and are not treated in time, especially in rural areas.

Mission of the project:
He Shi Eye Care, one of the largest eye care providers in Northern China, covering 11 eye hospitals and a private university, is setting up a large-scale retina screening program. for diabetes and other diseases. TU/e, the University Eye Clinic Maastricht, and TU/e’s sister BMIE department at Northeastern University in Shenyang are developing a comprehensive automated retinal image analysis and computer-aided diagnosis suite, a one-stop-shop application, for large-scale screening for Diabetic Retinopathy (DR), Glaucoma, Age-related Macular Degeneration (AMD) and Macular Edema.

The project is based on a low-cost Dutch scanning-laser ophthalmoscope with topography and innovative software by i-Optics BV. Final goal is to screen the total population of the province of Liaoning, China, some 24 million people, in order to find as early as possible indications for these diseases, so treatment is feasible, effective, quality of life is maintained, and healthcare costs-reducing.

Funding
The project is funded by the European Foundation for the Study of Diabetes, the Chinese Scholarship Council, i-Optics BV, He Shi Eye Care, Maastricht University Eye Clinic, Northeastern University and TU/e. Funding request are pending with the NWO-He program.

Partners:
The partners are: TU/e Biomedical Engineering, University Eye Clinic Maastricht, i-Optics BV (www.i-optics.com), Northeastern University Dept. of Biomedical and Information Engineering (Shenyang, China), and He Shi Eye Care (Shenyang, China).
Partners collaborate in the Sino-Euro Vision and Brain Institute, see the VBI website and RetinaCheck.org for full details.

Partners have formed two RetinaCheck teams, one in Eindhoven, and one in Shenyang, China. The Eindhoven / Maastricht RetinaCheck team meets every last week of the month on Thursday 12:00 in room GEM-Z 1.03. The international teams communicate by skype.

Project work flow:
The total project consists of several steps:

A) Innovative retinal fundus camera development, low-cost (i-Optics)
B) Automated admissible image quality check (TU/e, BMIE)
C) Software development for computer-aided diagnosis (TU/e, BMIE)
D) Pattern recognition algorithms for combining features (TU/e, BMIE)
E) Data collection for validation (UECM, He Shi)
F) Validation study (TU/e, UECM, He Shi)
G) Screening preparation (He Shi, BMIE)
H) Screening (He Shi, assisted by all partners)
I) Clinical follow-up (He Shi)
J) Web-based CAD and IPR
K) Publish, share and exploit the results (all partners)

We need application builders.
This is a large and international project, and we need MSc students why like to build effective applications.

TU/e and BMIE are currently building efficient applications for automated retinal image analysis.
So far we have developed, in successful student projects:

– Vascular tracking with brain-inspired multi-orientation methods (Erik Bekkers, Remco Duits)
– Vessel curvature measures (May Wong, Erik Bekkers)
– Vessel enhancement (Jiong Zhang, Emre Baspinar, Remco Duits, Smaneh Abbasi)
– Optic disk detection (May Wong, Erik Bekkers)
– Arterio/Venous diameter ratio (Paul Bloembergen, Koen Eppenhof)
– Support Vector Machine combined feature classification (Fan Huang)
– Micro-aneurysms (Han van Triest, Li Bo)
– Fractal Dimension for vascularity patterns (Rick Philipsen, Fan Huang)
– Graphical User Interface (GUI) for the validation study (Frank Martin, Guilherme Moura, Behdad Dasht Bozorg)

Student projects:

Summer 2015 we will start the following projects, and look for enthusiastic and motivated BME students, for an MSc project (9-12 months), internship (3 months) or (international) Erasmus project (3-6 months):

A) An efficient GUI for the Validation Study.
To know whether the detected features in the fundus images (such as number of micro-aneurysms, vessel curvature, A/V ratio etc.) are descriptive for the disease, we need to validate each feature for the several classes in which we can discriminate for the disease stages (no disease, early stage, etc.).
Another reason is to give to the end-user a homogeneous user-interface where he can access the different applications in a easy to learn way, in order to be able to process large amounts of images efficiently.
The GUI will be extended from earlier work. The GUI is developed in Mathematica 10, and will be tested by medical experts from Maastricht and Shenyang.

B) Pattern Recognition methods to optimally combine features
The many features combine into a so-called feature-vector for each image. These vectors can be viewed as points in a high-dimensional feature space. It is expected that clusters in this feature space indicate the different states of the disease. The challenge is to optimally separate these clusters, with optimal state-of-the-art pattern recognition techniques.
This work will extent earlier work on Support Vector Machines, and will be expanded to Learning Vector Quantization (LVQ).
A nice aspect of this large-scale screening project is that we will eventually obtain large-scale data, which may substantially assist the classification process. Google Translate works due to the huge database of spoken words, entered in the learning process. We envision such a recognition scheme here as well in the future.

C) Improving the speed of the CAD algorithms by innovative implementation on massively parallel hardware
The current applications are developed in Mathematica, which is a superior design language, but relatively slow for processing larger size and larger quantities of images. Such hardware can be GPUs (Graphical Procession Units), or FPGA (Field Programmable Gate Array) systems. For each strategy we have support in the group.

D) Image Quality Check
Image quality consists of two parts:

a) Physical performance of the camera, expressed in objective measures as PSF (Point Spread Function), MTF (Modulation Transfer Function), contrast-detail curve, SNR (signal to noise ratio) etc.
Techniques should be developed for both regular digital camera systems, as SLOs (Scanning Laser Ophthalmoscopes).

b) Subjective image quality: The images in the field are made by skilled operators, but may still be not suitable for automated processing. The required anatomy may not be in the field of view, the patient may have inclusions, floaters or cataract, leading to hazy recordings, or other flaws in the image acquisition.

E) Junction detection
Morphological and functional changes in the cardiovascular pattern can be observed through minute changes in the vascular layout, often difficult to see by the human observer. In this application we develop a robust vessel bifurcation- and vessel crossing detector, for different sized contributing vessels, for dim and highly curved vessels, and returning the bifurcation angle.

F) Efficient Arterio / Venous Ratio detection
A current implementation exists, based on the color histogram of the oxygen-rich and oxygen-poor vessels in color fundus images. However, we think that more vessel characteristics, such as specularity (arteries and veins reflect light differently), and vessel wall characteristics may improve the performance of the current application.

G) Design a vessel phantom to validate the measured vessel characteristics
It is extremely important that the developed CAD algorithms are calibrated, and give the true values.
The task in this project is to develop physiologically realistic phantoms to calibrate for retinal vessel characteristics: curvature, width, bifurcation angle, segment length, aneurysm width, stenotic index.

H) Inventory of retinal image analysis algorithms
There is a plethora of quantitative retinal image analysis algorithms in the literature. This shorter project is a literature study to evaluate the scope of features measured, and their effectiveness, and to give recommendations for the Retina Team to develop new members of the feature detecting algorithm tree in Mathematica.
The study must include other imaging techniques of the fundus, such as 3D OCT (Optical Coherence Tomography), and the new AO (Adaptive Optics) technology for extreme resolution.

I) Micro-aneurysms
Small aneurysms (vessel widenings) and small bleedings are among the earliest signs of diabetic retinopathy. However, they are often overlooked, similar to other features as vessel crossings, or difficult to quantify in a longitudinal study (study the increase or decrease over time). The application built should have an optimal balance between sensitivity and specificity (good detection rate, but not too many false positives).

Supervisors:
Students will be supervised by the following team members, depending on the topic:

– prof.dr.ir. Bart M. ter Haar Romenij, project leader
B.M.terHaarRomeny@tue.nl, tel. 040-2475537
– dr. Behdad Dasht Bozorg, postdoc
B.Dasht.Bozorg@tue.nl, tel. 040-2475537
– dr. Tos Berendschot, head Clinical Physics Maastricht University Eye Clinic
Tos.Berendschot@maastrichtuniversity.nl, tel. 043-3877345
– ir. Erik Bekkers, PhD student TU/e and research scientist at i-Optics BV, Den Haag
E.J.Bekkers@tue.nl, tel. 040-2473037
– dr.ir. Remco Duits, TU/e, staff member BMIA and Dept. of Mathematics
R.Duits@tue.nl, Tel. 040-2473037 / 2472859
– drs. Jiong Zhang, PhD student TU/e
J.Zhang1@tue.nl, tel. 040-2475571
– ir. Han van Triest, PhD student and lecturer at BMIE, Shenyang, China
J.W.v.Triest@gmail.com, tel. +86-13504986945
– prof. Cao Yue, Head Ophthalmology Research at He Shi Eye Care, Shenyang, China
cy791120@163.com, tel. +86-24-86549318

projects can start between now and September 2015, please contact us.

Status: Open
Category: Master project
Duration: 3-12 months
Start date: 9/1/2015
End date: 09/01/2016
Student: BMT, ME, N, EE, WSK
Research group: Multivalued Image Analysis
Supervisor: Bart ter Haar Romeny
Sec. Supervisor: Remco Duits
Collaboration: Maastricht University Eye Clinic, i-Optics BV, Northeastern University, BMIE School Shenyang, China
External supervisor: dr. Tos Berendschot, ir. Han van Triest
Location: Eindhoven, Maastricht, Shenyang China

 

Shenyang, China: Validation of retinal imaging biomarkers in different diabetic stages

PROJECT DESCRIPTION AND GOAL

This is a 3-months externship in Shengjing Hospital in Shenyang, China. The  project is in close collaboration with our TU/e sister Biomedical Engineering department in Shenyang (BMIE).

[read more=”Read more” less=”Read less”]

This project is ideal for two colleague students.  See Project description.

Introduction:
Diabetes is a systemic disease affecting all parts of the body. Diabetic retinopathy (DR) is the damage done to the eye due to diabetes, and a major source of blindness.

The RetinaCheck project (www.retinacheck.org) focuses on the fighting of the diabetes epidemic in China. Its mission is to build up computer-aided diagnosis (CAD) software to analyze retinal fundus images for early detection of diabetic retinopathy (DR). A large-scale screening in Liaoning province, China for DR is set up. Teams at TU/e and Northeastern University develop a suite of software packages.

To validate this software for its predictive and classification power, a study has been set up with simultaneous recording of retinal images and extensive diabetic metadata for a large number of Chinese diabetes patients: the “Shengjing Study”.

Diabetic retinopathy can cause observable changes to the retina, such as change in bifurcation angles, retinal arteriolar-to-venular-width ratio (AVR), fractal dimension, vessel tortuosity, length to diameter ratio, which may result in micro-aneurysms, stenoses, angiogenesis, drusen and micro-bleeds vessel leakage. Therefore, analysis of the micro-vascular pattern can provide us with rich features of the retinal vascular tree.

Goal of this externship:
To understand how diabetes progresses into DR, we need to correlate the biomarkers extracted from the retinal fundus images with the diabetes metadata in this clinical validation study. Many combinations will be explored. The externship will lead to a publication.
In this externship you will discover Chinese healthcare and culture in depth.

REQUIREMENTS
This project is suitable for BMT master students who, after their Biomedical Engineering study, wish to continue their clinical study based on medical image processing techniques. Since this student will carry out multidisciplinary research in a fascinating cultural environment in China, he/she needs to communicate well with researchers, diabetes experts, ophthalmologists, students and patients. A positive attitude and good communication skills are essential.

WORKPLACE, SUPERVISION, LIVING AND TIME PERIOD
The student will work in the Diabetes Department of Shengjing Hospital, Shenyang for 3 months. Supervision will be done by prof. Bart ter Haar Romeny and prof. Han Ping. Shenyang is an attractive large city, and has a lot to offer. It is a modern city, with low costs of living.

PROJECT DETAILS

Several aspects can be addressed during the externship, to be discussed:
• Statistical analysis of the database in order to reveal the potential/possibilities of the database. E.g., if the distribution of meta-data parameters is known, more specific research questions can be defined.
• Investigate the reproducibility of automated image analysis software for the extraction of retinal (vascular) features.
• Analyze/Improve the data infrastructure (from image- and metadata acquisition to data-storage, to (automated) data analysis, to feedback reports).
• Perform a clinical study with the available data and image processing tools (e.g. how does vessel tortuosity relate to different stages of DR, etc.).

CONTACT
For project details, please contact prof. Bart ter Haar Romeny.
Email: B.M.terhaarRomeny@tue.nl
Status: Open
Category: Externship
Duration: 3 months
Start date: Any time in 2017 (preferably Spring 2017)
End date: 3 months later.
Student: BME /ME
Research group: Biomedical Image Analysis & Interpretation
Supervisor: Prof. Bart ter Haar Romeny
Collaboration: Shengjing Hospital, Shenyang, China
External supervisor: Prof. Han Ping
Location: Shenyang, China
[/read]

Two full-time 2-year positions in Shenyang, Liaoning Province, China (both require a PhD).

postdocADV

  • Project leader clinical screening for diabetic retinopathy
  • Postdoc image analysis for diabetic retinopathy

The RetinaCheck project (www.retinacheck.org) is a large screening and early warning project in North-East China to find early signs of diabetes by automatic analysis of high-resolution optical images of the retina. Diabetes is epidemic in China, with an alarming incidence today of 11.6% of the population, and diabetic retinopathy is a leading cause of blindness. Aim is to screen 24 million people in the province of Liaoning, and to establish a one-stop-shop screening program for diabetic retinopathy, glaucoma and age-related macula degeneration.

The project is a Sino-Dutch collaboration between He Eye Care Systems and Northeastern University in Shenyang, China, and Eindhoven University of Technology and i-Optics Inc. in the Netherlands.

Advanced brain-inspired algorithms have been developed for the fully automatic computer-aided diagnosis of the retinal images by teams both at Eindhoven University of Technology in the Netherlands, and Northeastern University in China. Storage and networking will be based on a tele-radiology system developed at He Eye Hospital, and on a cloud-based Picture Archiving and Communication (PACS) system supplied by Neusoft Medical Systems, China’s largest medical imaging manufacturer. The clinical image acquisitions take place at the conglomerate of He Eye Hospitals in Shenyang and Liaoning.

The project consists of 4 stages: algorithm development, validation of the algorithms (against diabetic and other metadata), screening, and commercialization.

The two positions are for a period of two years, which may possibly be extended. The positions focus on the establishment of the clinical workflow at He Eye Hospitals, in close collaboration with all international members of the RetinaCheck project. Salary will be determined by the background and experience of the candidate. The main locations of work are at He Eye Hospital and the BMIE School of Northeastern University in Shenyang, but this is flexible given the conditions and progress of the project.

Tasks:

– Prepare, optimize, maintain and supervise the image acquisition workflow at He Eye Hospital in Shenyang.

– Implement the various developed software tools in a clinical setting.

– Integrate image acquisition with the diagnostic clinical metadata of the patient record.

– Set op validation studies with clinical experts, involving sufficient scale acquisitions and annotations.

– Supervise and coach students involved in partial projects of the RetinaCheck project.

– Publish scientific papers in the international journals and conferences in the field.

Requirements:

– A PhD in Biomedical Engineering, Computer Science, Physics, or Electrical Engineering.

– A background in and proven affinity for big data science, statistical analysis and/or modern computer vision techniques. A background in retinal image analysis is recommended.

– Management skills and drive, to efficiently establish an infrastructure and routine workflow of large-scale clinical image acquisition in Chinese hospitals.

– Fluent in Chinese and English.

– Capable to communicate both with clinical experts, researchers and hospital management.

For all information:

Prof. Bart ter Haar Romeny, project manager.
Northeastern University / He University / Eindhoven University of Technology
Email: bart.romeny@outlook.com.

Bifurcations and crossings of the vessels and nerves in the eye as disease predictor and classifier

Systemic diseases such as diabetes, hypertension, and arteriosclerosis become more common due to the growing wealth and aging. Therefore, much research has been done in the developments of automated computer aided systems for screening and diagnosis of different diseases. Two non-invasive, fast, and inexpensive image modalities for these applications are retinal fundus photography and corneal confocal laser microscopy. Both modalities can provide high-resolution information about systemic and brain diseases.

[read more=”Read more” less=”Read less”]
Pathological changes to the vasculature can be examined with retinal photography, while nerve degeneration can be shown by corneal microscopy . Automatic methods should be developed to quantify these pathological alterations for screening applications. Recently, we have developed an automatic method for the detection of vascular and nerve bifurcations and crossings. We are looking for a highly motivated student who can develop a method to quantify pathological changes related to bifurcations and crossings.
One of the main technical challenges in this project is the automatic modeling of blood vessels at junctions via tracking and/or segmentation algorithms. Additionally, it is of great importance to demonstrate the clinical relevance of the (to be) developed method. To this end, the student will work in close collaboration with our clinical partners (University Eye Hospital Maastricht, and Meditta), who have made available three clinical datasets:
1. Color fundus images are available from 100 stroke patients and 100 controls. It should be investigated whether features such as arteriovenous-nicking, arteriovenous ratio, and curvature of the crossing vessels are related to stroke.

2. Fluorescein angiographies of 35 patients are available containing images before and after developing retinal vein occlusion. It should be investigated, whether features of crossing vessels can predict the development of retinal vein occlusion.
3. Corneal nerve images are available for diabetes patients and controls. Manual annotation of the bifurcations has shown that the bifurcation density is decreased for diabetes patients. However, such a significant decrease in density was not found using the automatic method for bifurcation detection due to low bifurcation detection sensitivity, since this method was mainly designed for retinal images. It should be investigated how this method can be optimized for the corneal nerve images.
Other clinical datasets could be set-up in conjunction with the clinical partners in this project.
This project is part of the larger RetinaCheck project (see www.retinacheck,org). You will be participating in all meetings, and have frequent visits to the Maastricht University Eye Clinic. All costs are reimbursed.

 

Starting date: 1 July 2015.

Duration: 9 months (BME or ME or MWT)

Supervision: mrs. Samaneh Abbasi (S.Abbasi@tue.nl), mr. Erik Bekkers (E.J.Bekkers@tue.nl), and mr. Behdad Dasht Bozorg (B.Dasht.Bozorg@tue.nl.

Status: Open
Category: Master project
Duration: 9 months
Start date: 7/1/2015
End date: 04/01/2016
Student: BME – ME- MWT
Research group: Computer-aided Diagnosis
Supervisor: Bart ter Haar Romeny
Collaboration: Maastricht University Eye Clinic
External supervisor: dr. Tos Berendschot (UMCM)
Location: Eindhoven + Maastricht
[/read]

RetinaCheck Graphical User Interface Development

In the recent years, the needs of Graphical User Interfaces (GUI) for scientific visualization in general and biomedical applications in GuiSampleImageparticular are increased.The analysis of biomedical datasets is an iterative process where the user experiments with different methods and parameters. The correct selection of methods and parameters is fundamental to show semantically significant information of the data. The design of graphical user interfaces that ease the interactive selection of parameters is key to take profit of the last improvements of analysis and visualization techniques.

[read more=”Read more” less=”Read less”]

In the framework of the important RetinaCheck project, a collaboration of TU/e and Northeastern University in Shenyang, China, for the large screening in China for early diabetes, we defined a new task for the development of a unique Retinal Computer-Aided Diagnosis GUI.

The high resolution fundus camera images generate a wealth of information, and due to the large size and huge numbers, specialized software is developed to do a fully automatic analysis and diagnosis. The integration of the many individual applications in one integrated system, both for the researcher and clinical user, is goal of this project.

Project goal

This project is focused on the development of Graphical User Interface and the implementation of all of our recently developed retinal image analysis methods in one application. Several requirements must be considered in the design of this software. The interface should facilitate the automatic retinal image analysis by establishing a common repeatable procedure, and also increase the performance and reliability of the entire analysis. The software should be capable of image storage and management, allowing the collaboration between experts in different locations for the different studies. The GUI must allow the integration of new image analysis modules and new pipeline design and it should provide several features and tools to increase the interactivity and usability.

Student profile

  • Enthusiastic Master student in electrical engineering, biomedical engineering, computer science, or a related field.
  • Highly skilled in Java GUI programming and familiarity with Java GUI frameworks such as Swing, SWT, and AWT
  • Proficient understanding of Mathematica and MATLAB programming languages (Python and C++ are optional but can be helpful).
  • Knowledge of parallel programming.
  • Familiarity with the concept of Adaptive User Interface.
  • Basic understanding of server-based applications.
  • Understanding fundamental design principles behind a scalable application.
  • Knowledge of image analysis is advantageous.
  • A good team player with excellent communication skills.
  • A creative solution-finder.

This project is part of the larger RetinaCheck project (see www.retinacheck.org).

CONTACT
For project details, please contact mr. Behdad Dasht Bozorg, PhD. Email: B.Dasht.Bozorg@tue.nl.

[/read]

 

Pathology Detection in Retinal Images for Large-Scale Screening of Diabetic Retinopathy

Diabetic Retinopathy (DR) is the main cause of blindness among the MBDetectionThumbmiddle-aged population in the world. The progresses of diabetic retinopathy can be divided into four different stages and normally in the first stage the disease is silent, which brings difficulties to the early diagnosis of DR. The first signs of DR are: capillary microaneurysms, dot and blot hemorrhages, hard and soft exudates.

Microaneurysms (MAs) are among the early signs of DR and are small swellings which are caused by a weakening of the vessel wall and located at the side of tiny blood vessels. In digital color fundus images, MAs appear as tiny, reddish isolated dots. The MAs detection and analysis have been considered as one of the most important strategies for the early diagnosis of DR. It can significantly improve the efficiency and reduce the costs in a large-scale DR screening setting.

[read more=”Read more” less=”Read less”]

Another main clinical sign of the presence of DR is exudates, which appear as white/yellow structures in color fundus retinal images. The sizes of exudates have very large variations, from the similar size as MAs to even the size as large as the optic disk region.

The classical approaches for detecting MAs and exudates usually start with a preprocessing step and a subsequent candidate extraction step. Finally, a selection- or a supervised classification procedure is applied based on features from candidates to find only MAs or exudates.

Project goal

This project is mainly focused on developing an algorithm for accurate Microaneurysms detection. These quantitative measurements of MAs will be used in the design of an automatic screening system for early findings of DR. Several public datasets will be used for the evaluation (validation with the ground truth). We are also preparing two datasets for MAs detection, a color fundus retinal dataset based on the high resolution DSR camera and a SLO dataset based on the EasyScan cameras.

Student profile

  • Enthusiastic Master student in electrical engineering, biomedical engineering, computer science, or a related field.
  • Able to program in Mathematica and/or Matlab (C++ is optional but can be helpful).
  • Knowledge of image analysis.
  • A good team player with excellent communication skills.
  • A creative solution-finder.

This project is part of the larger RetinaCheck project (see www.retinacheck.org).

CONTACT:

Mr. Jiong Zhang (J.Zhang1@tue.nl), Dr. Behdad Dasht Bozorg (B.Dasht.Bozorg@tue.nl)

[/read]