Deep Learning in Medical Imaging (DLMI) final project. Trained a CNN and GAN to upscale and denoise ultrasound images of liver cancer, improving diagnostic clarity while preserving critical biological structures.
Overview
The project addresses noise and low resolution in ultrasound images caused by machine or patient variability. A deep learning pipeline was built to reduce noise and enhance resolution, making it easier for clinicians to interpret images without obscuring important anatomical detail. The work focused on ultrasound images of liver cancer as the application domain.
Tech stack
CNN and GAN-based models for image enhancement, implemented in Python with Jupyter notebooks for experimentation. The repository includes preprocessing, training (including an Attention U-Net variant), and evaluation workflows, plus utilities for adding and visualizing noise in the source images.
Gallery
