Fig. 3.

Examples for misclassification: (red arrows represent features of interest) a Correct grading by the radiologist, misinterpretation by the AI. This subject was graded as normal by the radiologist; however, related to a mildly oblique projection, there appears to be a prominent bony structure in the intercondylar notch that could be interpreted as an osteophyte which would result in a mild OA grade. b Correct grading by the radiologist, misinterpretation by the AI. This was graded as mild OA given small osteophytes on either side of the tibial plateau. These osteophytes appear insignificant and could have therefore potentially also been graded as no OA. c Incorrect grading by the radiologist, correct grading by the AI. This was graded as mild OA by the radiologist; however, there is lateral femoro-tibial joint space narrowing, marked by the fact that the lateral joint space should be more narrow than the medial joint space, consistent with moderate OA, which is what the algorithm graded it as moderate OA