Title

Automated Knee Osteoarthritis Diagnosis: A Stacked Ensemble Deep Learning Approach with Explainable AI Techniques

Abstract

Knee osteoarthritis (KOA) is a prevalent degenerative joint disease that affects people all over the world. Because diagnosis and treatment are often delayed, KOA frequently causes severe disability. While a lot of research has been done to predict KOA therapy, most of the suggested ways are unreliable because they are not linked to explanatory AI methodology, strong preprocessing techniques, or appropriate hyperparameter tuning. This research provides a deep learning framework for KOA classification that can solve binary(diagnosis) and multi-class (severity prediction) classification problems using the Osteoarthritis Initiative (OAI) dataset. Our proposed stacked ensemble model, which incorporates Xception, EfficientNetB5, and InceptionV3, outperformed the individual models with an accuracy of 86.29% in KOA diagnosis and 96.93% in KOA severity prediction. By standardizing image quality and highlighting important features for classification, a thorough image pre-treatment pipeline that includes scaling, sharpening, denoising, histogram equalization, and contrast enhancement boosts our method even more. Gradient-weighted Class Activation Mapping (Grad-CAM), Faster ScoreCAM, and Local Interpretable Model-agnostic Explanations (LIME) are three sophisticated explainability tools that we incorporated into our model workflow to guarantee transparency and interpretability. These tools offer distinct visual insights into the model’s decision-making protocol. Our findings offer a strategy that balances transparency with performance, perhaps leading to an earlier and more accurate diagnosis of KOA.

Authors

Mohammad Azad, Tanvir Rahman Anik