Chetana Krishnan

Ph.D

Biomedical Science

Dept of Radiology UAB

Graduate Research Assistant

- Developed a convolutional neural network for automatic liver lesion classification using dual-phase CT scans. Used techniques like 2D and 3D - Created MRI brain extraction and segmentation pipelines with multi-atlas registration and voxel-based morphometry analyses to quantify - Led project applying diffusion tensor imaging and tractography algorithms on neonatal brain MRIs to map early brain connectivity patterns and their disruptions in developmental disorders - Implemented active shape and appearance models for cardiac MR image segmentation to assist in automating ventricular volume quantification and ejection fraction measurement - Proposed a radiomics approach extracting shape, intensity, and texture features from breast DCE-MRI for improving diagnostic accuracy in breast cancer subtyping - Developed variational autoencoders for robust anomaly detection in chest X-rays and experimented with generative adversarial networks for realistic CT scan synthesis - Automated traditional region growing method and modified CLAHE filters for 3D images for lung and airway segmentation - Authored and co-authored six papers in peer-reviewed journals and conference proceedings, summarizing the results - Developed machine learning algorithms to automate prostate cancer detection from multiparametric MRI scans, achieving over 90% accuracy - Collaborated with the Mayo Clinic and UCLA for automatic cyst counting and kidney segmentation from MR images - Created quantitative imaging biomarkers to characterize liver lesions from CT scans and correlate findings with clinical outcomes. Increased - Led ultrasound tissue characterization projects extracting spectral parameters via attenuation estimation and Nakagami modeling for classifying - Developed variational Bayesian nonnegative matrix factorization methods for multi-parametric MRI tissue pattern analysis to assist in Alzheimer's diagnosis - Experimented with the application of Siamese networks for one-shot learning classification tasks in medical imaging to mitigate limited - Developed a system for automated chest x-ray report generation using PaLM, a state-of-the-art LLM with over 500 billion parameters. Finetuned the BioLM LLM on chest X-rays and echocardiogram videos to generate free-text impressions for cardiac disease screening. Evaluated - Created a Bayesian optimization system leveraging Codex for prompt programming and automatic hyperparameter tuning of medical image - Developed a disease prediction model using CLIP-MoE to correlate DCE MRI, lab tests, and clinical notes over time - ImageBot - Developed an interactive AI tool for lung nodule segmentation refinement by integrating human-in-the-loop feedback with active