Fracture Vision AI

Bone fractures are among the most common medical conditions requiring immediate and accurate diagnosis. Conventional manual interpretation of radiographs is time-consuming and subject to variability among radiologists, often leading to diagnostic delays or inconsistencies. To overcome these limitations, this project develops an automated bone-fracture detection system that integrates deep-learning models with an intuitive web interface. The system employs a customized InceptionV3 architecture enhanced with a Bottleneck Attention Module (BAM), enabling the model to focus on clinically relevant features such as subtle discontinuities and fine structural patterns in bone images. Images are preprocessed by resizing to 224×224 pixels and normalized to match the training distribution, ensuring consistency between training and inference. The model is trained on acurated dataset of fractured and non- fractured X-ray images, with augmentation techniques applied to improve generalization and reduce over fitting. Performance evaluation is carried out using validation accuracy, loss curves, and a confusion matrix, with additional metrics such as ROC-AUC and PR-AUC for binary classification robustness. The web application, built using Flask, provides a user-friendly interface that allows clinicians and researchers to upload X-ray images, receive predictions with confidence scores, and view supporting analytics such as class distribution and prediction history. Authentication mechanisms, secure file handling, and integrated visualization charts enhance usability and reliability. The proposed system demonstrates the potential of combining attention-based deep learning with interactive web deployment for clinical decision support. This project lays the groundwork for scalable, real-time diagnostic tools that can complement radiological expertise in resource-constrained healthcare environments.