Quantitative ultrasound for diagnosis and assessment of rheumatoid arthritis
Quantitative analysis of contrast-enhanced ultrasound enables fast, 97% accurate diagnosis and assessment of rheumatoid arthritis for personalized therapy.
Rheumatoid arthritis (RA) is a chronic systemic disease whose main characteristic is persistent articular inflammation. This results in joint destruction and loss of function. It is one of the major causes of disability in industrialized countries, affecting 1–2% of the population world-wide, and it leads to inability to work and increased mortality.1 RA is the worst outcome among the different forms of arthritis, but an early diagnosis and effective treatment can avoid its most devastating effects. Unfortunately, it is difficult to distinguish between different types of arthritis, especially at their onset. In fact, even if the inflammatory process has already begun at a microvascular level (creating new vessels in the joints), there may be an absence of changes in both the clinical manifestations and in the radiological appearance.2
Ultrasound imaging of microbubble contrast agents (contrast- enhanced ultrasound, or CEUS) is an imaging technique that can detect changes in the microvasculature, assisting early diagnosis of RA with less discomfort for the patient, reduced cost, shorter examination time, and more general availability compared to the standard MRI approach.3 Here, we show that a quantitative analysis of synovial perfusion (blood flow to the joint) using CEUS enables identification of the different arthritis forms.4
In practice, clinicians generally examine CEUS data (see Figure 1) considering the whole inflamed joint. This region-of-interest (ROI) approach is fast, but only allows for description of the area of vascularization and its perfusion at a macroscopic level. In contrast, our approach enables a greater level of detail, considering each pixel of the image separately. The method can be more susceptible to noise and patient movement, but it enables a measure of localized perfusion variations. Moreover, this approach provides information about the presence of specific perfusion patterns (see Figures 2 and 3) and their spatial distribution. Figure 3 shows how perfusion kinetics vary widely within the synovia, making it difficult to describe them by a single value, as with conventional ROI analysis.
We recruited 115 patients between the ages of 29 and 78 years affected by either RA (57 subjects) or other types of arthritis (non-RA, 58 subjects) from the University of Padua Hospital. All patients showed signs of inflammation in finger joints but similar clinical and serological (serum) values. Two rheumatologists chose the most active joint for each patient and examined it with CEUS5 and performed the clinical examination (grade and diagnosis). We quantified CEUS data by modeling the perfusion curves with a gamma-variate model6 both at the ROI and pixel level. We described the curves in terms of peak value, time to peak, raise and washout time (the time needed to raise the intensity from the baseline value to half-maximum and from the peak value to half-maximum, respectively), regional blood flow, and regional blood volume.
We compared the diagnostic power of the CEUS quantification results at ROI and pixel levels, and the semiquantitative grade provided by the radiologists through visual inspection. To this end, we trained three different supervised random forest classifiers to assess the diagnostic accuracy of the following possible clinical setups: semiquantitative CEUS grade with serological marker values; ROI-based perfusion parameters; and pixel-based perfusion parameters.
When we considered only serological marker values and semiquantitative grades, the classifier achieved a specificity of 93% but only a sensitivity of 64%, with an overall accuracy of 84%. Interestingly, the classifier trained with the regional perfusion parameters had worse performance than that using manual assessment, achieving a specificity of 60%, a sensitivity of 64%, and an accuracy of 62%. This indicates that the ROI-based description of the CEUS data is not sufficient to derive a comprehensive description of the complex heterogeneity of perfusion patterns within the synovia. By contrast, when we considered the pixel-based perfusion analysis, the classifier achieved a sensitivity of 95%, a specificity of 100%, and an overall accuracy of 97%. We could also provide maps of perfusion parameters (see Figure 4) for time of appearance and mean transit time.
In summary, we have compared the analysis of CEUS data at ROI and pixel level and showed that a more detailed description of the vascularization can distinguish different arthritis types more effectively than using only ROI-based analysis or the radiologist's visual assessment of the images. In particular, we showed that pixel-based analysis of CEUS data can discriminate RA from non-RA arthritis even when there are mild or no clinical differences.
In future work, we will investigate whether perfusion patterns identified in the synovia by our method can be linked to different disease phenotypes. If this is the case, quantitative analysis might provide more personalized and phenotype-specific approaches to therapies and patient management.
Enrico Grisan is an assistant professor in the Department of Information Engineering. His main interests are the processing and analysis of biomedical images, in particular of optical endoscopy and confocal microendoscopy, ultrasound, and MRI.
Gaia Rizzo is a senior postdoctoral researcher in the Department of Information Engineering. Her research activity covers the quantification of positron emission tomography (PET) images, arterial input function modeling, correlation of genomic data with PET images of protein density, and development of methods for the kinetic analysis of ultrasound images.
Roberto Stramare is associate professor in the Radiology Department, and since 2014 has been director of the School of Specialization in Radiodiagnostics. He has authored more than 110 publications.
General Hospital of Bolzano
Bernd Raffeiner graduated in medicine from the University of Innsbruck in 2005, and then specialized in rheumatology at the University of Padua, where he obtained a PhD in rheumotological sciences in 2013. He has authored 70 publications in national and international journals and congresses.