I graduated with a Masters degree in the ECE department of University of Florida. Lost a lot of blood, broke several several bones and went through tremendous physical and mental pain during this time. I am so proud that I did not quit.

The title says it all! Her name is Trina, the love of my life.

I started working under the supervision of Professor Navid Asadi in the FICS Lab in the department of ECE, University of Florida. I worked on projects regarding physical assurance for AI and package security.

AI Samurai Japan Limited is a sister concern of Chowagiken Company Limited Japan, a startup born from Hokkaido University laboratories with the mission of "Putting research into practical use and making it useful to society". They are focused on employing cutting-edge AI tools into practical uses. I worked in this company as a Machine Learning Engineer.

I completed my graduation from the Electrical and Electronic Engineering (EEE) department of Bangladesh University of Engineering and Technology (BUET) in February 2017.

The objective of this competition was to identify melanoma in images of skin lesions. This was a binary classification problem where we had to predict whether the melanoma condition in a provided image is benign or malignant. Our combined effort got us 253rd position (top 8%) on the Private Leaderboard among 3,314 teams. This was my first bronze medal in any Kaggle competition and second medal overall. My detailed analysis and codes are available here in my GitHub repo. A lot of credit goes to my amazing team partners Mohammad Innat and Uday Kamal. Innat's well-explained solution is available here.

Intracranial Hemorrhage (ICH), which refers to bleeding inside the cranium, is a serious health problem requiring rapid and intensive medical intervention. In case of an emergency, it is necessary to urgently diagnose the subtype of IC from brain Computed Tomography (CT) scans. However, in hospital emergency rooms in low-resource countries, a skilled radiologist is almost non-existent, and an emergency medical officer has to make this urgent and critical decision. This may lead to missed diagnosis, especially in the case of subarachnoid hemorrhage which often shows subtle changes in the CT image. In this work, we have developed a web-based AI tool for generating colorful heatmaps over the grayscale anatomical 2D CT image slices identifying the possible regions of the subtype of acute IC. The tool can be used by radiologists for AI-based assistance for making a more accurate and faster diagnosis. Our Web Application for Intracranial Hemorrhage Detection got accepted at the Society for Imaging Informatics in Medicine (SIIM) 2020 Conference. Currently, the project is being supported by BrainStation 23. They have deployed a web tool called RadAssist using our deep learning model. This project was supervised by Dr. Taufiq Hasan and Dr. Paul Nagy.

mHealth Lab is working towards developing biomedical signal processing and machine learning algorithms for remote health monitoring, and prototyping innovative mHealth devices for improving people’s health, overall well-being, and quality of life. This lab is founded and maintained by the Biomedical Engineering Department of BUET. I worked as a Research Assistant (RA) there.

Diabetic retinopathy is a diabetes complication that affects eyes. It's caused by damage to the blood vessels of the light-sensitive tissue at the back of the eye (retina). It can be cured if treated at an early stage; otherwise, it could lead to permanent blindness. Detection of the stage of diabetic retinopathy is a challenging task even for doctors. Aravind Eye Hospital in India collected thousands of retinopathy images and hosted this competition for the detection of the level of DR in these images. In this competition, we secured the position 37 on the Private and 55 among 2,943 teams on the Public leaderboard. Our model scored 0.829 on the public LB and 0.926 on the Private LB. Codes are available here.

I got myself admitted to the Masters Program of EEE Department, BUET in June 2019. I took courses on Modern Power System, Optical Fiber, Biomedical Signal Processing, and Digital Image Processing.

Extraction of relevant pathological terms from radiology reports is important for correct image label generation and disease population studies. We published a paper titled Pathology Extraction from Chest X-Ray Radiology Reports: A Performance Study where we compared the performance of some known application program interfaces (APIs) for the task of thoracic abnormality extraction from radiology reports. We explored several medical domain-specific annotation tools like Medical Text Indexer (MTI) with Non-MEDLINE and MeSH On Demand (MOD) options and generic Natural Language Understanding (NLU) API provided by the IBM cloud. Our results showed that although MTI and MOD are intended for extracting medical terms, their performance is worse compared to generic extraction APIs like IBM NLU. Finally, we trained a DNN-based Named Entity Recognition (NER) model to extract the key concept words from radiology reports. Our model outperformed the medical-specific and generic API performance by a large margin. Our results demonstrated the inadequacy of generic APIs for pathology extraction tasks and established the importance of domain-specific model training for improved results. The paper is available here.

Semion Limited is a startup founded by Dr. Khalid Ashraf. The primary goal of this company is to create and promote services in the health sector through digitization of patient and their associated data management. The digitization process is backed up by tremendous technological advancement and customization through constant research and development. They also develop artificial intelligence-based algorithms for automating abnormality detection and assisting doctors. I worked as a Machine Learning Researcher there.

I worked as a non-paid Writer and Editor at Zero2Infinity, a monthly science magazine, from March 2013 to June 2015.