Repository for all biomedical image processing projects, specifically in classification, localization, and segmentation. This repository is designed to replicate and build upon state-of-the-art (SOTA) results in these different domains.
This project uses a Swin Transformer-based model for the multi-class and multi-label classification of chest X-rays from the NIH ChestX-ray dataset. The model aims to detect various thoracic diseases by analyzing the X-ray images.
This project employs the DINO (Self-Distillation with No Labels) approach for image localization tasks on the VinDr dataset. The bounding boxes are provided to use as reference for the end results.
This notebook provides test results and evaluation metrics for the localization models applied on medical imaging datasets. It includes performance analysis and visualization of the model's predictions.
This project focuses on segmentation tasks, specifically targeting the detection and segmentation of pneumothorax conditions in chest X-ray images. The model is trained and tested on a relevant dataset to identify and segment pneumothorax regions.
This notebook explores both 2D and 3D medical image classification using the MedMNIST dataset. The notebook aims to replicate and build upon SOTA techniques for handling both 2D and 3D medical images, with a focus on diverse datasets and deep learning methods.