Attention-based CNN for KL-Grade Classification


  • Authors: Bofei Zhang, Jimin Tan, Kyunghyun Cho, Gregory Chang, Cem M. Deniz
  • Published at ISBI 2020

Code available here

Knee osteoarthritis (OA) is a chronic degenerative disorder of joints and it is the most common reason leading to total knee joint replacement. Diagnosis of OA involves subjective judgment on symptoms, medical history, and radiographic readings using Kellgren-Lawrence grade (KL-grade).

In this study, we applied ResNet to first detect knee joint from radiographs and later combine ResNet witg CBAM to make a prediction of the KL-grade automatically.

Fig.1 Model Architecture for Knee Joint Detection and KL Grade Classification

The proposed model achieved the current state-of-the-art on OAI dataset with a multi-class average accuracy of 74.81%, mean squared error of 0.36, and quadratic Kappa score of 0.88. We also study the model by applying GradCAM on the model prediction.

Fig.2 Model output a distribution on KL Grade. GradCAM suggests how model makes a decision