OSTEODEEP: AI-Driven Bone Health Assessment with Chatbot Integration Using X-Rays

Abstract

Low bone mass conditions, encompassing a spectrum from osteopenia to the more severe osteoporosis, represent an escalating global health challenge due to their silent progression and high fracture risk. This paper introduces OSTEODEEP, a novel platform for bone health assessment from X-ray images, integrating a deep learning model with a chatbot for enhanced user interaction. The diagnostic module employs a hybrid ensemble approach. Features are extracted from knee X-ray images using pre-trained EfficientNetB2 and VGG-19 models. These features are then fused, their dimensionality is reduced via Principal Component Analysis, and they are classified into normal, osteopenia, and osteoporosis categories using a stacking ensemble classifier composed of SVM and XGBoost as base learners and LR as a meta-learner. Evaluated on a public dataset of 1,947 X-ray images, the proposed model achieved a value of 90.10% for test accuracy and 90.14% weighted F1-score. The model demonstrated excellent discriminative power with a macro-average ROC AUC of 0.96 and showed particular strength in identifying the challenging osteopenia class (Recall: <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{9 1. 2 0 \%}$</tex>). The OSTEODEEP platform demonstrates that combining deep feature extraction with an ensemble classifier offers a robust and highly accurate solution for bone density screening using standard X-rays. By integrating a diagnostic tool with a ChatGPT-powered chatbot, this work presents a practical and scalable solution to improve early detection, bridge diagnostic gaps in primary care, and enhance patient engagement.

Ähnliche Arbeiten