Mayank Golhar
About Me
Hello! नमस्कार! 您好! Bonjour! Hallo!
I'm a Ph.D. student at JHU BME. Previously, I completed my Masters in ECE from Johns Hopkins University. I am fortunate to be working with Dr. Nicholas Durr for my thesis.
My research interests lie at the intersection of fields of computer vision, machine learning, and optics, particularly their applications to the biomedical domain. My research involves improving the performance of deep learning models for datasets with small sizes or poor annotations. I'm also exploring the joint optimization of optical systems and AI algorithms for improved analysis of medical images.
Previously I worked as a Senior Software Engineer in the Medical Imaging Research Group, Healthcare & Medical Equipment division at Samsung Research Institute - Bangalore. I graduated from the Indian Institute of Technology (IIT) Guwahati in 2017 with a Bachelors in Electronics and Communications Engineering.
If I'm not tinkering with medical imaging algorithms, you can catch me playing volleyball or running on Baltimore streets.
As with life, this site is always a work in progress!
Research Interests
Computer Vision
Machine Learning
Deep learning
Biomedical Optics
Medical Image Analysis
Endoscopy
Education
Ph.D. Student, Johns Hopkins School of Medicine, 2021-Present
Biomedical Engineering
M.S.E., Johns Hopkins University, 2019-20
Electrical & Computer Engineering
B.Tech., Indian Institute of Technology Guwahati, 2013-17
Major : Electronics and Communication Engineering
Minor : Computer Science and Engineering
Research Experience
Research Assistant
Durr Lab, Johns Hopkins University, Baltimore, USA
Sept 19 - Present
Senior S/W Engineer
Medical Imaging Research Group, Samsung Research Institute, Bangalore, India
June 17 - Aug 19
Bachelor's Thesis
Image Processing and Computer Vision (IPCV) Laboratory, IIT Guwahati, India
July 16 - May 17
Research Intern
Computer Vision Laboratory (CVL), Chubu University, Japan
May 16 - July 16
Research Intern
Lab for Video and Image Analysis (LFOVIA), IIT Hyderabad, India
May 15 - July 15
Publications
Mayank Golhar, Taylor L. Bobrow, Mirmilad Pourmousavi Khoshknab, Simran Jit, Saowanee Ngamruengphong, and Nicholas J. Durr. "Improving colonoscopy lesion classification using semi-supervised deep learning." IEEE Access (2020).
Mayank Golhar, Yuji Iwahori, M. K. Bhuyan, Kenji Funahashi, Kunio Kasugai, "Blood Vessel Delineation in Endoscopic Images with Deep Learning Based Scene Classification", Springer Lecture Notes in Computer Science, Pattern Recognition Applications and Methods, pp. 147-168, June 2018.
Mayank Golhar, Yuji Iwahori, M. K. Bhuyan, Kenji Funahashi, Kunio Kasugai, "A Robust Method for Blood Vessel Extraction in Endoscopic Images with SVM-based Scene Classification", International Conference on Pattern Recognition Applications and Methods (ICPRAM) 2017, pp.148-156. 2017.
Selected Works
Improving colonoscopy lesion classification using semi-supervised deep learning
MSE ThesisDr. Nicholas Durr, BME, Johns Hopkins UniversityDec 19 - Aug 20
Explored jigsaw puzzle solving based semi-supervised learning for polyp classification to improve performance by up to 9.8% using unlabeled data. Additionally, investigated semi-supervised learning’s advantages of domain adaptation, and out-of-distribution detection over purely supervised methods.
Addressing Computer Vision Challenges in Endoscopy Videos
B.Tech Thesis & Research Internship Project, Prof. M. K. Bhuyan, ECE, IIT Guwahati & Prof. Yuji Iwahori, CS, Chubu UniversityMay 16 - May 17
Blood vessel delineation in endoscopy videos : Improved the performance of vessel detection with Frangi Vesselness method by 8% using custom symmetry detection filtering and background removal.
Endoscopic Scene classification : Classification was done into 4 classes based blood vessel informationand dye content. A support vector machine was trained on features based on edge, colour and texture information for classification. Further, the accuracy was improved to 98.5% by using a ResNet inspired CNN.
3D reconstruction of polyp : Used Structure from Motion & 3D Recurrent Reconstruction Neural Network for 3D shape recovery of polyp.
[Thesis]
Medical Image Analysis Algorithms for Cardiac & Musculoskeletal Ultrasound Images
Medical Imaging Research Group, Samsung Research Institute BangaloreJune-17 - Aug 19Worked on the development of following algorithms:
Semi-automatic cardiac valve segmentation in 3D Ultrasound (US) Transesophageal echocardiography images.
Enhancement and optimization of Panoramic image stitching algorithm specific to Musculoskeletal (MSK) US images;
Motion Detection in US images.
Please have a look at my project page for other interesting projects & details!
Technical Skills
Programming Languages: C, C++, Python, C#
Libraries/Other Softwares: MATLAB, PyTorch, OpenCV