LungLens TB

Empowering Tuberculosis detection in low- and middle-income countries (LMICs) with AI-driven chest X-ray analysis. Our platform aims to reduce radiologists' workload while delivering patient-centered care through precise lesion localization and personalized, multilingual summaries, revolutionizing healthcare efficiency.

LungLens TB is a comprehensive solution designed to address the challenges of tuberculosis (TB) diagnosis, particularly in low to middle-income countries (LMICs), by leveraging cutting-edge artificial intelligence (AI) technologies. At its core, our platform employs deep convolutional neural networks (CNNs) such as ResNet, EfficientNet, and DenseNet for chest X-ray analysis. These models are first trained on a dataset of 112,120 chest X-rays from the National Institutes of Health (NIH) for a binary classification task, distinguishing between various findings like an infiltration, atelectasis, emphysema, and no finding. This initial training phase ensures robust performance in identifying abnormalities indicative of TB. Following this, transfer learning is applied to fine-tune the models using chest X-ray datasets sourced specifically from LMICs. This step is crucial as it allows the models to adapt to the unique characteristics and variations present in images from these regions, thereby enhancing their ability to accurately detect TB cases. Furthermore, our platform incorporates gradient-based visualization techniques such as Grad-CAM (Gradient-weighted Class Activation Mapping) to provide interpretable insights into the model's decision-making process. By generating saliency maps, we enable radiologists to visualize which regions of the chest X-ray are most influential in the model's classification, aiding in lesion localization and diagnosis. On the patient communication front, we integrate GPT-4, a state-of-the-art natural language processing (NLP) model, to generate patient-friendly summaries of radiology reports. This involves transforming complex medical jargon into clear and understandable language tailored to the patient's comprehension level. Additionally, our system supports multilingual capabilities, ensuring that these summaries are accessible to patients regardless of their native language. In summary, LungLens TB represents a significant advancement in the field of AI-driven healthcare. By combining deep learning techniques with interpretable AI and personalized patient communication, our platform not only enhances TB diagnosis efficiency but also promotes patient-centered care in LMICs, ultimately contributing to improved healthcare outcomes on a global scale.

Flask, Python, PyTorch, HTML, CSS, JavaScript, Ajax,

https://github.com/isamu-isozaki/LungLens-TB/tree/main

Satvik Tripathi (st3263@drexel.edu), Isamu Isozaki (imi25@drexel.edu), Manil Shrestha (ms5267@drexel.edu), Dubem Okoye (do427@drexele.edu)