Xray-Vision is a multimodal machine learning pipeline designed to assist in the interpretation of chest X-rays by combining unsupervised image clustering with natural language processing (NLP) techniques. This approach aims to make radiological insights more accessible, especially in settings with limited medical expertise.
Image Clustering: Utilizes a pre-trained VGG16 model to extract features from chest X-ray images, followed by KMeans clustering to group similar images based on visual patterns.
Report Summarization: Applies transformer-based models (e.g., BART) to summarize associated radiology reports, providing concise descriptions of common findings within each image cluster.
Clinical Term Analysis: Implements co-occurrence analysis and similarity assessments of clinical terms in reports to understand language patterns and their diagnostic relevance.
Unsupervised Learning: Enables pattern discovery without the need for labeled data, facilitating scalable analysis of large datasets.
Multimodal Integration: Combines visual and textual data to provide a comprehensive understanding of chest X-rays.
Accessibility: Aims to support healthcare professionals by offering preliminary insights, potentially reducing diagnostic workload.
Capstone_Final_Project.ipynb: Main Jupyter notebook containing the implementation of the pipeline.
USML_Project_Report.pdf: Detailed report outlining the project's methodology and findings.
X-ray vision.pptx: Presentation summarizing the project's objectives and results.