Rupasri Chalavadi

Leveraging Natural Language Processing to explore Medical Charts!

Radiology and AI Laboratory (RAIL) at Johns Hopkins University Malone Center for Engineering in Healthcare. • Machine Learning in Python, with a focus on PyTorch, NumPy, Pandas, SciPy and SciKitLearn, to lead stroke damage prediction projects driven by deep learning approaches. • Pioneer the development of a deconvolution-free method using 4D CT perfusion (CTP) data. This approach predicts core and penumbra infarct volumes on MRI through an end-to-end neural-network-based methodology, addressing challenges in data-intensive deconvolution-based methods. • Analyze and process 4D CT perfusion data, extracting features using NumPy and Pandas. Utilize OpenCV for intricate image processing tasks ensuring robust handling of medical imaging data. • Implement and fine-tune neural network model architectures, largely CNNs, enhancing accuracy and efficiency of stroke damage predictions. Utilized Matplotlib and Seaborn for data visualization and model evaluation.

• Artificial Intelligence - Fall 2024 • Computer Vision - Spring 2024 • Cognitive Neuroscience - Spring 2023, Spring 2024 • General Physics 2 for Engineering - Spring 2024 • General Physics 1 for Engineering - Fall 2023

• Patient education volunteer as part of the Violet Project from the Johns Hopkins Hospital Ob/Gyn department working to discuss with and educate patients regarding menstrual, sexual, and reproductive health in the Pediatric Emergency Department, Harriet Lane Clinic. • Developed and conducted 3-month-long educational workshops on menstrual, sexual, and reproductive health at Baltimore City Schools.

Dept: Research and Exploratory Department (REDD) Human and Machine Intelligence Research Group Project: Cohort Research Investigating Micro-Scale Optimized Networks • Engineered a novel, computationally efficient approach to cell segmentation using virtual agents. • Published results in "A Novel Semi-Automated Proofreading and Mesh Error Detection Pipeline for Neuron Extension." DOI: https://doi.org/10.1101/2023.10.20.563359. • Worked primarily in Python using NumPy, Pandas, SciKit Learn, PyTorch, OpenCV, and Trimesh for robust automation of meshes representing cells and synaptic connections for automating cell segmentation extension. • Designed and deployed an advanced error detection pipeline for forward and reverse neuron extension on the MICrONS dataset. Successfully recovered 2.8 million synaptic sites for 4138 unique neurons with an 80% accuracy rate. Utilized Matplotlib and Seaborn for data analysis and visualization.

Dept: Research and Exploratory Department (REDD) Human and Machine Intelligence Research Group • Developed a Proofreader Evaluation and Analysis ToolKit (PEAK) in Python to analyze the performance of manual proofreaders in contributing edits to the cell segmentation of the Minnie65 MICrONS dataset. • This toolkit assesses proofreader performance using metrics (historical accuracy, time taken, perceived difficulty, etc.) to evaluate edits proposed across multiple proofreaders, predicting a singular edit per error (using the Bayesian paradigm) with an accuracy rate of 93.3% as compared to 88% accuracy for conventional methods of combining edits from multiple proofreaders. Statistically proved significance of data. • Poster presentations: NIH BRAIN initiative's 9th Annual conference (June 2023); Johns Hopkins School of Engineering poster session (May 2023) - won the Deans’ Design Day award on behalf of the Applied Mathematics and Statistics Department; Johns Hopkins Undergraduate Research Symposium (April 2023).

• Worked as an undergraduate researcher at the Fuchs' Lab @Johns Hopkins Medical Institute to quantitatively analyze the density and distribution of 'ribbons' (biological structures) in the ear's inner hair cells in mice. • Analyzed the relationship between these ribbons and AMPA receptors in order to investigate the causes of age-related hearing loss. • Worked extensively with ImageJ in Fiji to analyze synapse densities, and the 3D visualization software IMARIS to analyze synaptic connections between inner hair cells and type II afferent neurons, arguing for a role in the pathway for noxious sound.

• Supervisor: Dr. Deva PriyaKumar, Dr. Deepalatha Subramanian. • Worked with the computational chemistry software Gaussian09 to analyze the effects of isoelectronic substitutions of N+ and B- in place of C in the buckybowl Corannulene on its structural properties (electronic nature, size, position, and HOMO-LUMO gap energies) • Detailed research procedure and findings in a summary paper - attached below.

Double Major in Computer Science and Neuroscience

Completed the International Baccalaureate Diploma Program (IBDP) with a 43/45 final score. Took IB Mathematics, Chemistry, Biology at the higher level, and Economics, Spanish, English at the standard level. Prior to this, completed the IB Middle Years Program (IB MYP).