Call for Application: NAAMII Internship & Training 2023 [ CLOSED ]

• Post by: Naamii
• June 26, 2023
• No Comment

Are you interested in gaining practical experience in cutting-edge research and technology? Do you have a passion for solving complex problems in the fields of deep learning, medical imaging, or computational genomics? The Nepal Applied Mathematics and Informatics Institute for research (NAAMII) is pleased to announce its Internship & Training Program for the year 2023. This opportunity offers an exciting opportunity for talented individuals to work on impactful projects under the guidance of experienced supervisors and researchers.

Project 1: Geometric Deep Learning for EEG Epilepsy Classification
Supervisor: Bishesh Khanal

Summary:
The objective of the study is to design a geometric deep learning approach to identify whether a person is suffering from epilepsy or not using Electroencephalogram (EEG) data. EEG test is comparatively less expensive than another test like MRI and fMRI. Thus, the aim of this project is to make neurological disorder treatment affordable and accessible even for remote areas. The thesis student will require going through Graph Neural Networks (GNNs) and use it with publicly available EEG data for building prediction models. Collaboration with a medical expert in EEG and Epilepsy is possible once a reasonable model is built with public data.

• The student will explore whether the state-of the art geometric deep learning model give better result on EEG dataset for classifying Epilepsy?
• Will explore novel methods for better results using GNN on temporal EEG data

References 
[1] https://arxiv.org/abs/2106.09135

Project 2: 2D X-ray Images to 3D Reconstruction
Supervisors: Bishesh Khanal, Mahesh Shakya

Summary:
X-ray radiographs give a 2D image while CT scans give 3D images. CT scan is expensive, has high radiation and is not readily available in many towns or rural areas. In contrast, X-ray is relatively accessible, low-cost, and has less radiation. However, surgeons (e.g., spine surgery in scoliosis, knee/hip joint replacement etc.) benefit from 3D scans which give a better picture of the 3D anatomy for surgery planning. In this work, we will work towards developing novel and state- of-the-art methods for 3D reconstruction from two X-ray views: AP view and lateral view.

References 
[1] https://openreview.net/pdf?id=RlCa0pFZsR9

 

Project 3: Medical Image Segmentation
Supervisors: Bishesh Khanal, NAAMII RAs at TOGAI research group

Summary:
Medical Image Segmentation is an important component of various clinical and surgical workflows including quantitative imaging, personalized implant design, patient progress tracking and surgery planning. Unfortunately, manual segmentation of anatomical structures can be time- consuming and tedious which is only going to increase as various imaging modalities become more common and accessible.

As the workload for radiologists increases, Deep Learning has the potential to assist by performing automated segmentation of organ-of-interest. Automated Medical Image Segmentation poses various challenges due to the variability of patient anatomy, irregularity of diseased regions such as tumours, polyps etc. and effort required for data annotation, to name a few.

Various potential avenues have been proposed in the literature such as Robust Architecture Design, Task-specific Augmentation, Self-supervised Learning etc to improve the state-of-the-art. We will explore various methods to build robust Deep Learning-based Medical Image Segmentation to tackle the aforementioned problem.

References 
[1] FixMatchSeg: https://arxiv.org/pdf/2208.00400 
[2] TGANet: Text-guided attention for improved polyp segmentation https://doi.org/10.1007/978-3-031-16437-8_15

Project 4: Systematic Benchmarking of Machine Learning Models to Predict Drug-Resistant Tuberculosis using DNA Sequencing
Supervisors: Raunak Shrestha (NAAMII & University of California, San Francisco), Bishesh Khanal (NAAMII)

Summary:
Tuberculosis (TB) is a major infectious disease leading to over a million deaths worldwide. Nepal ranks among the prominent nations facing a significant TB burden and associated mortality. TB is caused by an airborne bacterium, Mycobacterium tuberculosis (Mtb). Active TB infections are commonly treated using different combinations of antibiotic drugs. Although initially highly effective, many patients develop resistance to these drugs, the state is often termed drug-resistant TB (DR-TB). DR-TB is rapidly increasing and poses a significant threat to global public health. Early diagnosis of DR-TB is crucial for successful disease management and represents a significant clinical challenge that has yet to be adequately addressed.

The emergence of drug resistance in TB has been linked to specific gene mutations in the Mtb genome. Despite the identification of many mutations that contribute to DR-TB, they do not account for all cases of drug resistance, highlighting the necessity to discover novel gene mutations that are associated with DR-TB. The rapidly accumulating Mtb whole genome sequencing (WGS) data offer a valuable resource to explore and detect new drug-resistant mutations. Thus, efficient machine learning (ML) models are required to mine these voluminous genomics data and effectively predict DR TB.

The Computational Genomics Laboratory (CGL) at NAAMII has been actively working on developing ML models to predict DR-TB. CGL has a collection of over 5,000 Mtb WGS acquired from clinical TB cases across the globe. Here, we propose to benchmark the performance of different ML models in classifying drug-resistance TB cases published by research groups around the globe. Systematic assessment of ML models is necessary to identify the advantages and limitations of different ML models ultimately increasing the transparency of algorithmic performance and closing gaps between method developers, genomic scientists, and the end users of the tools.

Prerequisite

• Proficiency in Python programming
• Experience with Deep Learning frameworks such as PyTorch or TensorFlow
• Strong mathematical knowledge
• Keen interest in graphs and the mathematics of Geometric Deep Learning
• Experience with Graph Neural Networks (GNNs) is a plus
• BSc in Computer Engineering (Final Year students are also eligible) with strong interest in AI/ML, Computational Mathematics, Computational Biology, Bioinformatics, Biomedical Informatics, or Biomedical Engineering.
• Applicants from other similar educational backgrounds can also be considered based upon the qualifications, strong interest and proven capacities in above domains.

Guidance and Opportunity for Paper Writing
Throughout the internship, you will receive close supervision, guidance, training, and monitoring of your project. You will have regular meetings with supervisors and can interact with other researchers at NAAMII for assistance. There will be an opportunity to write a scientific paper under the guidance of supervisors if the work done during the internship is deemed suitable. Additionally, based on mutual interest, there may be an opportunity to continue as a research assistant in the future.

Application Method
To apply for the NAAMII Internship & Training Program, please fill out the application form via this link https://tinyurl.com/bdhyvxcw and submit a single PDF document containing a cover letter expressing which of the listed projects you want to work on and a detailed CV in above link. Your CV should include any relevant accomplishments or certifications demonstrating your experience in Machine Learning and Computer Vision. Please clearly mention your interest project in the Cover letter.

Plagiarism and Writing Guidelines
Plagiarism is strictly prohibited, and any instances of plagiarism will result in disqualification from the program. When writing your application materials or any other documents, ensure that all content is original and properly cited if quoting from other sources. While we understand that scientific writing skills may vary, we value your ideas and thoughts more than the quality of English writing. Focus on expressing your ideas clearly and authentically, and we will provide guidance to help you improve your scientific writing skills.

Duration: 9 months

Minimum number of hrs per week required: 12

Application Deadline: 3rd July, 2023.