The Knee OsteoArthritis Labeling Assistant (KOALA) software provides metric measurements of the joint space width and indicators for presence or absence of radiographic features of osteoarthritis (OA) on posterior-anterior or anterior-posterior (PA/AP) knee X-ray images. The outputs aid clinical professionals who are interested in the analysis of knee OA in adult patients, either suffering from knee OA or having an elevated risk of developing the disease.
Koala was trained on 35,000 digital XR images of 4,500 patients from study centers across the United States. Each image was consensus-graded by board certified radiologists according to OARSI and KL. Outputs are summarized in a KOALA report that can be viewed on any FDA approved DICOM viewer workstation.
Osteoarthritis is a paralyzing joint disease that can lead to joint replacement. Early detection as well as therapy can prevent unnecessary surgeries. Our first FDA cleared and CE-marked module KOALA supports physicians in detecting signs of knee osteoarthritis based on standard joint parameters and OARSI criteria of standing radiographs of the knee. KOALA addresses subjectivity in the longitudinal assessment of knee OA by providing a standardized, objective reading methodology, thereby enabling medical experts to monitor disease progression more confidently. KOALA can change the lives of over a billion patients living with bone diseases for the better.
KOALA receives radiographs from a PACS, analyzes the images and stores the analysis results in the PACS. As soon as KOALA receives the image, the analysis and the generation of reports are fully automated and entail no user interaction besides the user viewing the reports. The user can send images to KOALA via standardized DICOM commands or the file interface and receive reports over the same interface. In this sense, the user does not “operate” the device but simply reviews the reports and can accept or reject them. The IB Lab KOALA software does not provide a graphical user interface that uses the DICOM-viewers of existing workflows by attaching reports to the original study. All relevant images can be pre-analyzed by KOALA automatically by appropriate filter rules in the PACS. With this the original workflow is seamlessly augmented by KOALA.
IB Lab KOALA is a radiological fully-automated image processing software device of either computed (CR) or directly digital (DX) images intended to aid medical professionals in the measurement of minimum joint space width; the assessment of the presence or absence of sclerosis, joint space narrowing, and osteophytes based OARSI criteria for these parameters; and, the presence or absence of radiographic knee OA based on Kellgren & Lawrence Grading of standing, fixed-flexion radiographs of the knee. It should not be used in-lieu of full patient evaluation or solely relied upon to make or confirm a diagnosis. The system is to be used by trained professionals including, but not limited to, radiologists, orthopedics, physicians and medical technicians.
It should not be used in-lieu of full patient evaluation or solely relied upon to make or confirm a diagnosis. The system is to be used by trained professionals including, but not limited to, radiologists, orthopedics, physicians and medical technicians.
Model performance metrics
Performance of KOALA was tested on a standalone dataset containing over 6000 images from over 1000 patients from the OAI dataset, and including over 12000 individual knee readings. Testing data was not included in the training of KOALA.
Each image in the testing dataset was consensus-graded by board certified radiologists according to OARSI and KL. When tested on the ability to differentiate between the absence or presence of OARSI-defined joint space narrowing (JSN), sclerosis (SC) and osteophytes (OS), as well as Kellgren-Lawrence (KL) score > 1, KOALA achieves the following sensitivities and specificities: JSN = (0.83, 0.80), SC = (0.82, 0.82), OS = (0.86, 0.79), KL = (0.87, 0.83).
KOALA also produces measurements of joint space width (JSW) in the medial and lateral compartment of each knee. JSW was also tested on the same dataset using a bootstrapping of orthogonal linear regression to compare KOALA measurements with ground truth readings. Resulting slopes and intercepts (with 95% confidence intervals) were 1.02 (0.99; 1.05) and -0.08 (-0.22; 0.03) on the medial side, and 0.97 (0.93; 1.00) and 0.08 (-0.15; 0.30) on the lateral side.