M. Neubauer (Krems, AT)
Center for Regenerative Medicine and OrthopedicsPresenter Of 1 Presentation
P153 - Artificial Intelligence in Radiographic Diagnostics in Correlation with Clinical Scores in Knee Osteoarthritis Patients
Abstract
Purpose
Assessing knee osteoarthritis (OA) is predominantly done radiographically via the Kellgren-Lawrence score (KLS).
Artificial Intelligence (AI) aid during analyzing xrays has been shown to lead to more accurate KL diagnostics and thus improved prognostic values of x-rays.
However, it is unclear wheter those AI-aided diagnostic results more likely correlate with clinical and biomechanical scores.
The aims of this study are to investigate (i) a correlation of clinical scores (KOOS, IPAQ) and (ii) biomechanical values in comparison of the AI-aided versus the AI-unaided x-ray diagnostics.
Subjects were sub-analyzed from a larger, physiotherapy trial (MLKOA).
Methods and Materials
63 patients were analyzed in this study (MLKOA) with symptomatic knee OA ranging from KL0-3.
At baseline x-rays and clinical scores (KOOS, IPAQ) as well as biomechanical values were taken.
X-ray readings at baseline were performed twice by experienced physicians (2 orthopedic surgeons and a radiologist) AI-unaided and again after 3 weeks AI-aided.
Aided and un-aided readings have been performed twice by each reader to minimize intra-reader variability. The order of x-rays was randomly assigned at each reading.
Read subscores in a semi-quantitaive manner included deformity, joint space narrowing, osteophytosis and sclerosis resulting in a numerical value that defines the KL score.
Outcome measures included: ICC = (intra-class correlation/agreement rate in between readers with/without AI); accuracy (surgeons versus ground truth oft he radiologist), Receiver operating Curve (individual learning curve with/without AI) and the correlation with clinical scores.
AI analysis was performed with a software from „Imaging Biopsy Lab“ (Vienna, Austria).
Results
In total 63 x-rays were included after analyzing 72 DICOM pictures (only knees included into MLKOA were analyzed even tough some subjects received an x-ray oft he contralateral knee).
A trend of an increased correlation with AI-aid and clinical scores was detected.
Conclusion
The use of AI in digital imaging analysis increases reproducability and accuracy in diagnosing Osteoarthritis
Presenter Of 1 Presentation
P153 - Artificial Intelligence in Radiographic Diagnostics in Correlation with Clinical Scores in Knee Osteoarthritis Patients
Abstract
Purpose
Assessing knee osteoarthritis (OA) is predominantly done radiographically via the Kellgren-Lawrence score (KLS).
Artificial Intelligence (AI) aid during analyzing xrays has been shown to lead to more accurate KL diagnostics and thus improved prognostic values of x-rays.
However, it is unclear wheter those AI-aided diagnostic results more likely correlate with clinical and biomechanical scores.
The aims of this study are to investigate (i) a correlation of clinical scores (KOOS, IPAQ) and (ii) biomechanical values in comparison of the AI-aided versus the AI-unaided x-ray diagnostics.
Subjects were sub-analyzed from a larger, physiotherapy trial (MLKOA).
Methods and Materials
63 patients were analyzed in this study (MLKOA) with symptomatic knee OA ranging from KL0-3.
At baseline x-rays and clinical scores (KOOS, IPAQ) as well as biomechanical values were taken.
X-ray readings at baseline were performed twice by experienced physicians (2 orthopedic surgeons and a radiologist) AI-unaided and again after 3 weeks AI-aided.
Aided and un-aided readings have been performed twice by each reader to minimize intra-reader variability. The order of x-rays was randomly assigned at each reading.
Read subscores in a semi-quantitaive manner included deformity, joint space narrowing, osteophytosis and sclerosis resulting in a numerical value that defines the KL score.
Outcome measures included: ICC = (intra-class correlation/agreement rate in between readers with/without AI); accuracy (surgeons versus ground truth oft he radiologist), Receiver operating Curve (individual learning curve with/without AI) and the correlation with clinical scores.
AI analysis was performed with a software from „Imaging Biopsy Lab“ (Vienna, Austria).
Results
In total 63 x-rays were included after analyzing 72 DICOM pictures (only knees included into MLKOA were analyzed even tough some subjects received an x-ray oft he contralateral knee).
A trend of an increased correlation with AI-aid and clinical scores was detected.
Conclusion
The use of AI in digital imaging analysis increases reproducability and accuracy in diagnosing Osteoarthritis