TBdetect

TBDetect: Accelerating Tuberculosis Diagnosis with AI

About the Project

Tuberculosis (TB) remains one of the most persistent global health challenges, especially in underserved areas where timely diagnosis and treatment are essential but often unavailable. Inspired by the urgency to address this issue, we embarked on a mission to develop TBDetect—an AI-powered tool to deliver fast, accessible, and reliable TB diagnoses. Our goal is to support healthcare providers by accelerating the diagnosis process, ultimately reducing transmission rates and enabling timely treatment.

What Inspired Us

The World Health Organization reports that millions of people suffer from TB each year, yet diagnostic delays are common due to limited access to healthcare infrastructure. Witnessing the immense impact of delayed diagnosis on patients and communities, we saw an opportunity to bridge the gap using advanced AI technology. Our vision was clear: make TB detection faster and more accessible, particularly in areas with limited medical resources.

What We Learned

Throughout this project, we gained in-depth insights into AI and machine learning for healthcare, along with hands-on experience in handling complex datasets. Specifically, we learned:

The importance of data preprocessing, particularly with medical images, to ensure reliable AI model performance. How to use deep learning algorithms, such as DenseNet and Inception V3, for image analysis tasks in a clinical context. The significance of explainable AI in healthcare, which led us to integrate interpretive explanations and confidence scores to enhance user understanding of the results. Our journey also highlighted the ethical considerations involved in developing healthcare technologies, ensuring that our model's predictions are accurate, unbiased, and respectful of user privacy.

Building the Project

Our solution, TBDetect, is built on a deep learning model trained to analyze X-ray images for TB-related anomalies. Here’s a quick breakdown of our approach:

Data Collection: We sourced X-ray images from the TBX11K dataset, which contains annotated images across various TB categories. Data Processing: The images underwent transformations, including resizing, flipping, rotation, and normalization, to enhance the model's training process. Model Training: We compared DenseNet-121 and Inception V3 architectures. After evaluation, DenseNet-121 emerged as the optimal choice due to its dense connectivity and efficiency in capturing TB features. This model allowed us to build a robust tool capable of fast, real-time analysis. Website Development: A user-friendly website serves as the interface, allowing individuals to upload X-ray images and clinical data for instant diagnosis. Users receive results within 10 minutes, complete with explanations, confidence scores, and visualizations. Challenges We Faced

Our journey wasn’t without obstacles. Some of the key challenges we encountered included:

Data Quality and Availability: Medical imaging data for TB can be challenging to source, especially with balanced classes across TB categories. We had to carefully preprocess and augment our data to ensure quality and consistency. Interpretability and Explainability: Ensuring that our model’s predictions were interpretable was crucial for user trust. Developing meaningful visualizations and confidence scores demanded considerable effort and collaboration with healthcare professionals. System Integration: Building a seamless pipeline from data input to output, with real-time results on a web interface, involved overcoming technical complexities related to web integration and server processing. Looking Ahead

We believe TBDetect has the potential to be a game-changer in TB diagnostics, particularly for remote and underserved communities. Moving forward, we plan to:

Expand to a mobile platform, increasing accessibility for users in resource-limited areas. Collaborate closely with healthcare professionals to validate our model against clinical standards. Integrate the system with Electronic Health Records (EHR) to streamline diagnosis and ensure accuracy.