Use Cases - Radiology (Proof-of-Concept): Provides a tool for radiologists to quickly reference visually similar X-ray images, potentially aiding in diagnosis or identifying subtle patterns. - Research (Exploration): Allows researchers to investigate potential relationships between X-ray images based on visual similarity. - Education (Demonstration): Serves as a learning aid for understanding image representations and similarity search in a medical context.

System Architecture - Deep Learning Model: ResNet-50 (pre-trained on ImageNet): This model extracts meaningful feature vectors representing the visual content of X-ray images. We leverage a pre-trained model for efficiency and to avoid extensive training for this prototype. - Vector Similarity Search Database: Milvus: Optimized for fast and efficient similarity searches on high-dimensional vector data.

Workflow 1. Image Preprocessing: Incoming X-rays are resized and normalized for consistency.

2. Feature Extraction: The pre-trained ResNet-50 model (without the final classification layer) transforms each X-ray image into a high-dimensional feature vector.

3. Vector Storage: Milvus stores the extracted feature vectors, enabling fast similarity comparisons.

4. Query Image: The user uploads a query X-ray image.

Search: - The query image's feature vector is generated. - Milvus performs a vector similarity search, finding the most visually similar images within the dataset.

Results: MedSearch successfully demonstrates the core principles of vector similarity search for medical image retrieval. Users can upload a query image and see visually similar X-ray images retrieved from the dataset.