Deep Learning with 3D CNN for Knee Osteoarthritis Detection and Classification

Of all joints, the knee is the most commonly afflicted with osteoarthritis (OA), which is the most common form of arthritis. Even though CNNs are seriously being utilized in medical imaging, 3D CNNs are yet to be applied as a standard for diagnosing knee OA. In this, the 3D CNN model was applied on the knee MRI sequences to classify the presence of osteoarthritis (OA). Probably the strongest point of 3D CNNs over 2D CNNs is that 3D CNNs can process the whole MRI sequence in one go, while 2D CNNs can only analyze a single photograph. In this, the KL grading ranking the intensity of OA from 0 to 4 was used for each one of the knees based on MRI data taken. The result showed that the 3D CNN model could differentiate the OA and non-OA knees well both in test and validation phases. Another model used for comparison is a 2D CNN model that uses X-ray images. Although much less accurate than the 3D CNN on MRI data, this model was also able to be used quite well. Thus, the diagnosis of knee OA is much more accurate by the method combining MRI and 3D CNN technology as compared to just with X-ray imaging, although both of these models have some clinical values. The work concludes by depicting how 3D CNN can be useful in the detection of OA through complex sequences of MRI images compared to traditional 2D methods.