Completed Projects

Research projects our undergraduate students have completed.

Home Research Completed Projects
Image
Project 1

MRI Visualization Tool
for Crohn’s Disease Assessment

The goal of this project was to design and implement an open-source based software tool that would enable the presentation of the MR data in MPR views, thus providing a better description of the disease.

By: Yael Zaffrani

Under the supervision of: Dr. Moti Freiman Technion's Computational MRI Laboratory

Find Out More

Image
Project 2

MRI Brain Tumor
Segmentation

Our goal was to overcome the need for large datasets by implementing regularization methods to increase the accuracy of automatic segmentation and find the ideal learning method for each contrast.

By: Shany Biton, Zohar Avinoam, & Michaella Ayoun

Under the supervision of: Dr. Moti Freiman Technion's Computational MRI Laboratory

Find Out More

Image
Project 3

Quantitative DW-MRI
Analysis Algorithms

The goal of our project was to create a Python library that simulates DW-MRI images, analyzes DW-MRI images by different algorithms, and evaluates the different results between them.

By: Marina Khizgilov and Judit Ben Ami

Under the supervision of: Dr. Moti Freiman Technion's Computational MRI Laboratory

Find Out More




MRI Visualization Tool for Crohn’s Disease Assessment


Crohn’s disease is a chronic inflammatory bowel disease (IBD) which often presents with discontinuous transmural inflammation of the distal ileum and proximal colon. Individuals with this disease often suffer from diarrhea, abdominal pain, fever, and/or weight loss.
Currently most individuals are diagnosed using ileocolonoscopies, which is an invasive procedure. Magnetic resonance enterography (MRE) is a non-invasive procedure that has the ability to become the standard diagnostic tool for Crohn’s disease. A challenge of using an MRE for diagnosis is it requires scrolling back and forth to understand the anatomy of the bowel from 2D images making it difficult to fully appreciate the disease extent.
The goal of this project was to design and implement an open-source based software tool that would enable the presentation of the MR data in MPR views, thus providing a better description of the disease. These improvements have the potential to lead to better assessments of medical therapies and decision-making for surgical treatments.
Using software programming in python we created open source software to provide a depiction of the disease state on a single image. This enables the visualization of extraction of Crohn’s disease quantitative imaging markers.
Next steps would include starting a clinical trial to compare measurements between conventional approaches and our MPR software tool. Endoscopic measurements should be used as the reference standard. Advanced algorithms should be implemented for centerline extraction. Finally, our software should be integrated with the MR system, with the long-term goal of replacing the conventional approach.

This project was awarded 2nd place in the Technion's Underegrad Project's Day Competition on August 20, 2020! Check out the announcement (in Hebrew) on the Technion's webpage here.
See Yael's presentation on her work (in Hebrew) here.

By: Yael Zaffrani
Acknowledgements to: Dr. Moti Freiman Director of Technion's Computational MRI Laboratory; Dr. Anat Grinfeld Head of Research unit, Radiology Department at Rambam; Dr. Anat Ilivitzki Director of the Pediatric Radiology Unit at Ruth Rappaport Children's Hospital


MRI Brain Tumor Segmentation


img
Glioma is a type of tumor that starts in the glial cells of the brain or the spinal cord. Some symptoms may include headaches, seizures, irritability, vomiting, visual difficulties, and weakness or numbness of the extremities.
Most cases are diagnosed by individual radiologists reading magnetic resonance imaging (MRI) or computed tomography (CT or CAT scan). These scans are difficult to read and each diagnosis is dependent on the skills and biases of the radiologist. Specifically, brain tumor segmentation is one of the most important and difficult tasks in many medical-imaging applications because it involves a huge amount of data.
Automatic segmentation algorithms can overcome many of the difficulties inherent in reading MRI images. Our goal was to overcome the need for large datasets by implementing regularization methods to increase the accuracy of automatic segmentation and finding the ideal learning method for each contrast. Computer-aided diagnostic systems have the potential to have meaningful impacts on medical treatments by reducing the workload of doctors and giving more accurate results. Better medical imaging also leads to the improved planning and accuracy of surgical procedures.
We implemented deep learning algorithms using Unet. To overcome the difficulty of having a small dataset, we used Autoencoder regularization (AER) and Mixed Structure Regularization (MSR) for optimization. These algorithms enabled us to create images with tumor regions segmentation that display the whole tumor, tumor core, and active tumor.
Next steps for our research would include analysis and assessment of the segmented data. We need to run comparisons of various segmenting techniques and draw conclusions from the results.

By: Shany Biton, Zohar Avinoam and Michaella Ayoun
Acknowledgements to: Dr. Moti Freiman Director of Technion's Computational MRI Laboratory & Dr. Anat Grinfeld Head of Research unit, Radiology Department at Rambam


Quantitative DW-MRI Analysis Algorithms


Crohn’s disease is a chronic inflammatory bowel disease (IBD). Individuals with this disease often suffer from diarrhea, abdominal pain, fever, and/or weight loss.
Diffusion-weighted MRI (DW-MRI) is very sensitive to tissue architecture, such as cell density and microcirculation, and can therefore be harnessed to get a detailed description of disease characteristics. Two major challenges to using DW-MRI are the negative impact noise has on the data and the computational time needed to get clear results.
Our goal was to create a Python library that simulates DW-MRI images, analyzes DW-MRI images by different algorithms, and evaluates the different results between them. Using least squares fitting (SEGb, SEG, and LSQ), and Bayesian shrinkage prior (BSP) modeling ..
When produced using multiprocessing, the images were produced nearly twice as fast as without multiprocessing and they were able to withstand noise. These improvements to the efficiency and accuracy of DW-MRI open up many potential clinical uses.

By: Marina Khizgilov and Judit Ben Ami
Acknowledgements: Dr. Moti Freiman Director of Technion's Computational MRI Laboratory; Dr. Anat Grinfeld Head of Research unit, Radiology Department at Rambam; Elad Rotman MSc Candidate with Technion's Computational MRI Laboratory