Abstract
Knee osteoarthritis (KOA) is a common degenerative joint disease that
results in disability due to joint dysfunction and pain. Almost one-fifth of early KOA
cases are missed during the routine practice resulting in the progression of the disease.
This narrative review aimed to explore and analyze various literatures that proposed
Convoluted Neural Network (CNN) model in detecting KOA and its severity based on
Kellgren Lawrence grading classification. At first, 221 publications were retrieved
using the search term “artificial intelligence” and Knee osteoarthritis”. Only studies
that used CNN and radiographic images were included in this study in which only 14
studies fitted our inclusion criteria. Each paper was thoroughly investigated for the
input data and CNN model adopted as well as the performance and limitation of that
study. Lastly, the conclusion was made and discussed using these results. Object
detection and Classification models were among the most popular techniques adopted.
Our results showed that object detection models were overall superior regarding the
accuracy in the detection of KOA and its severity. The application of CNN for the
detection of KOA from radiographic images has shown great promise where each
technique has its own advantage. In the foreseeable future, the combination of object
detection and classification detection may provide excellent potential as a merit tool to
help orthopedists and related physicians for the proper diagnosis and treatment of
KOA.