Team Leader
Mak Rahman
Texas A&M University
Computer Science
maklachur@tamu.edu
Project Type
Research
Who Can Join
Graduate Students, Masters Students, Undergraduate Students
Project Description
This project investigates how deep learning can improve medical image analysis in an efficient and accessible manner. Our main objectives are:
– Create lightweight yet precise models for tasks like skin lesion, polyp, and organ segmentation and analysis. We aim to focus on architectures that find a balance between accuracy and efficiency, such as state space models (SSMs/Mamba), CNN-Transformer hybrids, and adaptive fusion modules.
– Incorporate generative models to enhance segmentation by producing meaningful insights directly from medical images.
– Conduct transparent experiments using public datasets, employ standardized evaluation metrics (Dice, IoU, HD95), and develop reproducible PyTorch pipelines.
– The expected outputs include a reproducible codebase, dataset preparation scripts, comprehensive ablation studies (covering parameters, FLOPs, inference time), and a concise or full-length paper or technical report suitable for workshops.
Team Needs
We welcome motivated students from diverse backgrounds. Helpful skills include:
– Deep Learning & CV: PyTorch, model training, testing, fine-tuning, transfer learning, and LoRA.
– Medical Imaging: Familiarity with medical image data; interest in anatomy and clinical context.
– Reproducibility & Systems: Conda/venv, CUDA, experiment logging (TensorBoard/Weights & Biases).
– MLOps: Git/GitHub, clean coding practices, and documentation.
Special Opportunities
– Contribute to workshop or conference papers; co-authorship available for significant contributions.
– Gain hands-on experience with efficient medical image analysis pipelines.
– Receive mentorship for skill development, including regular feedback, coding support, and opportunities to lead sub-modules.
– Help release a reproducible, lightweight baseline for the community.