Projects
Everything I've worked on, from the past to the present.
Presurgical Risk Stratification using ECGs
Surgical decision-making hinges on precise risk assessments, yet the role of
electrocardiography (ECG) in evaluating surgical risks is often debated, despite its
widespread use. In our study, we analyzed 12-lead ECGs from 28,661 adult patients
before major noncardiac surgeries, utilizing the MIMIC-IV dataset. We developed and
trained convolutional neural network (CNN)-based models to predict significant
post-surgical complications such as acute myocardial infarction (MI), in-hospital
mortality (IHM), and a composite of MI, 30-day mortality, and stroke.
Simulating Intracortical Microstimulation in MATLAB
This MATLAB simulation project aims to replicate the decision-making process for
intracortical microstimulation in the human somatosensory cortex. It focuses on
emulating real-life choices required for effective stimulation during a pinch task
involving the index finger and thumb. The simulation considers various stimulation
parameters to optimize tactile sensations' quality and intensity. This project is
based on detailed protocols and processes, using MATLAB and Arduino, to simulate
electrode placement, optimal electrode configuration, data collection, and pulse
generation, enhancing the understanding of somatosensory feedback in prosthetics.
Pneumatactor Arrays for Vibrotactile Feedback
Soft and conformable pneumatic tactile interfaces have the potential to
complement
traditional rigid systems by offering more natural and subtle feedback. This
presents the need for economical, scalable miniaturized systems adept in
providing
high-frequency vibrotactile stimuli. We present the development of a novel
silicone-based pneumatic vibrotactor (pneumatactor) array controlled by a
sequence
of high-frequency air valves engineered to deliver vibrotactile feedback.
ContROLLcraft - Minecraft Controller for the Differently Abled
Introducing our innovative Minecraft contROLLer! Crafted to emulate a mouse and
keyboard experience, this unique device is exclusively foot-operated. With one
box dedicated to movement control—walking, running, and direction via a
precision roller—and another for jumping, crouching, and block interactions,
this controller brings a new level of immersion to your Minecraft adventures.
Electrophysiological Correlates of Prospective Sense of Agency
The feeling of being in control of our actions, known as the Sense of Agency, is
a fascinating aspect of human experience. This sensation, rooted in the brain's
response to the consequences of our actions, underpins our perception of
autonomy and active engagement in the world. Neuroscience seeks to unravel the
complexities behind this phenomenon, offering insights that span philosophy,
psychology, robotics, and AI. In this exploration, we delve into how the brain
generates this sense of control, examining the interplay of cognitive processes
and sensory feedback.
Deep Learning based MRI Processing and Skull Stripping
In the ever-evolving field of neuroimaging, skull stripping stands as a crucial
process, ensuring that brain scans are free from any surrounding tissue
interference. This meticulous task enhances the accuracy of brain analysis,
facilitating everything from segmentation to 3D modeling. While traditional
methods remain a gold standard, they are impractical for handling large datasets
efficiently. Enter deep learning techniques: by leveraging powerful U-Net
architectures, these advanced methods promise to revolutionize skull stripping,
making it faster and more reliable than ever before.
Wild Blueberry Harvester Automation
Wild blueberries, known for their bold, untamed flavor and rich nutrient
content, grow low to the ground in rocky areas. I developed a camera-based
system to automate the measurement of berry volume and bin switching in the
harvesting process. This project was supported by Mitacs Globalink, supporting
top undergraduate students worldwide for a completely funded summer in Canada.
Hand Rehabilitation Device for Acute Burn Patients
The Hand Rehabilitation Device specifically targets the MCP, DIP, and PIP joints
of the fingers, providing focused therapy for acute burn patients. This
affordable solution facilitates essential exercises to restore mobility and
function in these critical joints. Its low-cost design ensures accessibility for
patients requiring effective rehabilitation.
TinyMLevator - Smart Elevator using Embedded ML
Embedded ML at it's current pace is finding applications in every nook and
cranny of the world. TinyMLevator brings it to the elevator industry. It
consists of two main parts: detecting people and recognizing speech to enable a
user standing in front of the elevator to simply speak where he intends to go.
As easy as that!
PicoBoo - Thermal Security System with Motion Tracking
PicoBoo utilizes thermal imaging and lightweight CNNs as the foundation for its
security system, operating on a Raspberry Pi. Using data from a FLIR
T420bx Thermal Imaging Camera, we created a human image training dataset,
implementing an optimized EfficientNet-Lite neural network for high-accuracy
human tracking in thermal camera video streams. The detections are displayed
through a web app for easy browser access.
CovPrev - App-enabled COVID-19 Symptom Testing Unit
During the COVID-19 pandemic, the need for safety measures like
social distancing and sanitization could not have been understated. Enter
CovPrev: an affordable unit targeting
crucial symptoms—high body temperature and low SpO₂ levels—with specialized
sensors. Deployable in bustling areas, CovPrev integrates a sanitization spray,
thermometer, pulse oximeter sensors, and a data analysis system accessible
through a cross-platform app.
