FRI0478 Dynamic automated synovial imaging (DASI) For differentiating between rheumatoid arthritis and other forms of arthritis: automated versus manual interpretation in contrast-enhanced ultrasound

Background Rheumatoid arthritis (RA) is the most aggressive chronic arthritis and affects about 1% of population, impacting severely both the individual wellbeing and the health care system. Early diagnosis and effective treatment can avoid the devastating outcome of RA1, but the differential diagnosis is especially difficult at its onset. The diagnosis relies on conventional methods (clinical parameters, autoantibodies), even if distinct vascularization patterns have been identified in biopsy specimens2. Contrast-enhanced ultrasound (CEUS) allows a non-invasive dynamic study of synovial vascularisation and perfusion, although its capacity in differentiating among different arthritis forms has not yet been evaluated3,4. Objectives to investigate the performance of quantitative analysis of CEUS data versus manual semiquantitative assessment in differentiating RA from other arthritis (non-RA). Methods 78 outclinic patients with finger joints arthritis were recruited, 33 with RA and 45 with other arthritis. The most active joint was chosen for CEUS examination as previously described3, using a US device (MyLab25, Esaote) equipped with Contrast tuned Imaging (CnTI, Esaote), and as contrast agent sulfur hexafluoride microbubbles (SonoVue; Bracco International). Both the anatomical B-mode image and the CnTI cineloop video were digitally stored for subsequent quantiative analysis or manual review. Two in arthritis experienced radiologists manually assessed the examinations as consistent with RA or not. Quantitative image analysis was performed firstly applying a semi-automatic detection of synovial boundaries5. Then, the contrast time-activity curve of all pixels belonging to the synovial and perisynovial region was analysed fitting a gamma curve f(t)=A(t-t0)a×e(t-t0)/b on the data. The statistics summarizing the distribution of the estimated kinetics parameters in the synovial and in the perisynovial tissue were computed and their difference between the two groups (RA and non-RA) analyzed, so to study the existance of different vascularization or flow patterns. Finally, a supervised classifier (random forest) was trained to classify each patient through its CEUS-derived parameters, validating the classifier diagnostic power using a leave-one-out strategy. Results Manual assessment of CEUS examination consistent with RA or non-RA performed by radiologists showed high sensitivity (0,9), but indeed low specificity and accuracy (0,46 and 0,69, respectively). On the