Predict And Classify Knee Osteoarthritis From X-Ray Imagery Using Deep Learning

Abstract

Knee Osteoarthritis (KOA) is a progressive degenerative joint disease affecting approximately 365 million people
globally, with prevalence rates of 9.6% in men and 18% in women over 60. This paper presents a deep learning-based
web system for automated 5-class Kellgren-Lawrence (KL) grade classification of knee X-ray images. Three models
are developed and compared: MobileNetV2 with transfer learning (85.6% accuracy), a Custom 4-layer CNN (83.2%
accuracy), and Random Forest with 100 estimators (89.2% accuracy). MobileNetV2 freezes the first 14 of 19 inverted
residual blocks and fine-tunes the last 5 with a custom Dropout(0.5)→Linear(1280,5) classifier head. The system is
deployed as a Flask web application with SQLite persistence, Bootstrap 5 dark-themed UI (#0a0a1a / #14b8a6), dragand-
drop upload, five-class confidence probability bars, Chart.js analytics dashboard, and Docker containerization
on port 5011.