Web-based semi-automated retinal fluid segmentation tool for OCT images. Combines traditional image processing with deep learning models (U-Net, MedSAM) to help medical professionals efficiently annotate and segment retinal fluid regions. Features interactive annotation tools, multiple segmentation algorithms, and user-friendly web interface.

Fine-tuning MedSAM and SAM for medical image segmentation with custom prompt strategies

Parameter-efficient brain lesion segmentation using MedSAM and LoRA on the MICCAI WMH Challenge dataset.

Anatomy-guided Cross-modal Fusion for automated radiology report generation from DICOM CT scans. Built with CT-CLIP, MedSAM, GatorTron, and LLaMA-3 (LoRA).

Benchmarking SAM and MedSAM on ISIC 2018 melanoma segmentation — with a focus on the deployment gap: published SAM papers measure performance using ground-truth-derived prompts that are unavailable in real clinical settings. This project quantifies that gap.

Monorepo containing the frontend (Next.js), backend (Node.js + Express.js), and GPU inference service (TensorRT-optimized Python engine) for a cardiac segmentation application, streamlining local deployment.

This is the official repository of Team Anastasia, which achieved 1st place in ImageCLEF 2025 Dermatological Segmentation Challenge (Subtask 1)..

Machine Learning for Medical Image Processing. Project done for Charité, the university hospital of Berlin. The aim was to segment coronary arteries and then extract a graph from it, in order to aid detection of coronary artery disease (CAD).

A comprehensive benchmark study evaluating the robustness of Segment Anything Model (SAM) and its medical domain adaptation (MedSAM) under realistic noisy medical imaging conditions.