Computer aided diagnostic model for knee osteoarthritis: A multi-modal feature regression approach

Background

X-ray imaging is crucial for diagnosing knee osteoarthritis (KOA) in clinical treatment. Computer-assisted diagnostic models reduce the impact of physicians' subjective factors on the accuracy of X-ray diagnoses. Therefore, continuous improvement of these models is necessary.

Methods

This study introduces a novel computer-aided diagnostic model for KOA, based on multi-modal feature regression using X-ray images. Specifically, two different modalities of diagnostic features are extracted. Firstly, by analyzing the diagnostic indicators of KOA in X-ray images, we explore the diagnostic basis of osteoarthritis from the perspective of image content and design several image content-based features for measuring bone gap, bone skin thickness, and bone mass. Meanwhile, medical information-based features such as age, gender, and surgical history are also integrated as diagnostic features. Then, the mapping relationship between the diagnostic features and the severity of osteoarthritis (denoted by K-L classification) is established using support vector regression to build the knee osteoarthritis diagnostic model.

Results

To validate the efficacy of our diagnostic model, we curated the NDKY-N2H knee X-ray image database, encompassing 1200 knee joint X-ray images, each paired with its corresponding K-L classification. By incorporating image preprocessing and a module for basic patient information, the accuracy of the model's KOA diagnosis has been improved. The final X-ray image KOA diagnostic model, developed based on multimodal feature regression, achieved an accuracy of over 98.42 % ± 0.11 % in identifying KOA. Additionally, for diagnosing KOA severity using K-L grading, the accuracy reached 85.06 % ± 0.49 %. These findings indicate that the proposed model performs exceptionally well in diagnosing knee osteoarthritis.

Conclusion

In conclusion, our research highlights the critical role of accurate KOA diagnosis in enhancing patient health. The innovative approach using multi-modal feature regression, which combines image content and medical information features, shows significant promise for providing reliable and efficient KOA diagnoses through X-ray imaging.