I am a junior in high school at the Academy of Science and Technology with a passion for computer science, engineering, and applications to real-world problems. This website displays my projects, research, and future work.
I have created several Augmented Reality Applications with C++ and Unity C#. Notably, I have developed a surgical aid for brain tumor removal incorporating AR.
I have significant experience in machine learning, having developed predictive algorithms including a distress detector and training surgical robots with reinforcement learning.
Over the past years, I have developed Python computer vision models to aid self-driving cars in bad weather, perform pose and marker estimation, create 3D reconstructions, and detect facial emotion.
Utilizing Javascript, HTML, and CSS, I have developed interactive and responsive websites for several organizations. I am especially skilled in three.js, incorporating 3D rendering and interactions through WebGL.
I have experience in Arduino and Raspberry Pi for creating robotic devices and cicuits. These include a humanoid robot with 17 motor control and other various Arduino projects.
I recently attended a workshop hosted by Johns Hopkins, conducting research on natural language processing, including chatbots, LLMs, and machine translation. I have coded applications to recognize speech, speech emotion, and perform voice control.
In order to develop my medical applications, I conducted significant research into analyzing patient scan data, using softwares such as 3DSlicer, Paraview, and TrackViz, and reconstructing the data in various 3D formats.
To faciliate my projects in augmented reality and robotics, I have gained extensive experience in CAD and 3D printing using Fusion360 and Blender.
Brain cancer poses a formidable challenge as tumors infiltrate the fragile and critical structures of the brain. Currently during surgery, vital information and scans are displayed on separate 2D monitors, diverting focus, compromising precision, and hampering the surgeon’s ability to navigate brain anatomy. To address these challenges, NeuroLens presents a novel augmented reality (AR) based surgical aid for the visualization of patient-specific tumor, vasculature, and white matter tract models directly on the patient with millimeter accuracy.
First, several deep learning models were trained on Magnetic Resonance Images (MRI) from more than 1000 patients for the delineation of these critical structures in 3D. Second, a comprehensive AR application was programmed in Unity and deployed to the Microsoft Hololens 2. This AR application featured an unprecedented 6 DOF tracking algorithm able to superimpose the 3D surgical guide models from any orientation in real-time onto a patient phantom head with less than 2 millimeters of error.
Mild Traumatic Brain Injury (mTBI) is a type of brain injury that can lead to acute cognitive impairment, increased risk of subsequent injuries, and lower quality of life. Yet, it is often very difficult to detect because there are a lack of visible abnormalities and standard symptoms, and rather, damage is done at microscopic scales This research utilized deep learning to recognize abnormalities like disconnects and shifts in white matter from diffusion MRIs (dMRI).
Since little labeled data is present for mTBI, my research uses an unsupervised learning approach - a custom autencoder was trained on over 400 unlabeled healthy dMRIs. As a result, the model could recognize anomalies in patient white matter. In collaboration with two local brain clinics, these anomalies were visualized in augmented reality (AR) for proving the extent of mTBI to a jury in insurance and personal injury cases.
Working with the Johns Hopkins Neurology Department, I have developed a virtual reality app for students to use to explore the brain in a detail never seen before. This app allows students to view all the regions of the brain pull regions out for further analysis. They can also pull volumetric MRI slices from the model to see the brain from both 2D and 3D perspectives.
Moreover, they can practice dissections, which are generally expensive to conduct in real life. The app provides all the necessary utlities to conduct a dissection, including removing cadaver slices and exploring cross sections, but through virtual reality
The field of robotics stands to revolutionize how surgery is performed by removing hujman error from the process. The goal of this research is to develop an autonomous robot that could navigate and perform basic tasks in minimally invasive neurosurgery. Notably, this research will focus on using Sim-to-Real, where the robot will be trained in a simulation environment and then transferred to an physical surgical robot.
Using medical scans, like MRIs and CTs, a digital twin will be created through training segmentation algorithms. Then, the digital twin will be configured in the simulation environment, such that the movement of brain tissue and bone will match real-life physics. The robot, the da Vinci Research Kit (dVRK), will be trained on the digital twin using a reinforcement learning policy, to learn specific tasks.