Premium
Preliminary Clinical Evaluation of Semi‐automated Nailfold Capillaroscopy in the Assessment of Patients with Raynaud’s Phenomenon
Author(s) -
MURRAY ANDREA K.,
FENG KAIYAN,
MOORE TONIA L.,
ALLEN PHILLIP D.,
TAYLOR CHRISTOPHER J.,
HERRICK ARIANE L.
Publication year - 2011
Publication title -
microcirculation
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.793
H-Index - 83
eISSN - 1549-8719
pISSN - 1073-9688
DOI - 10.1111/j.1549-8719.2011.00104.x
Subject(s) - tortuosity , medicine , automated method , raynaud disease , pattern recognition (psychology) , artificial intelligence , nuclear medicine , computer science , surgery , materials science , porosity , composite material
Please cite this paper as: Murray, Feng, Moore, Allen, Taylor, and Herrick (2011). Preliminary Clinical Evaluation of Semi‐automated Nailfold Capillaroscopy in the Assessment of Patients with Raynaud’s Phenomenon. Microcirculation 18 ( 6 ), 440–447. Abstract Objectives: Nailfold capillaroscopy is well established in screening patients with Raynaud’s phenomenon for underlying SSc‐spectrum disorders, by identifying abnormal capillaries. Our aim was to compare semi‐automatic feature measurement from newly developed software with manual measurements, and determine the degree to which semi‐automated data allows disease group classification. Methods: Images from 46 healthy controls, 21 patients with PRP and 49 with SSc were preprocessed, and semi‐automated measurements of intercapillary distance and capillary width, tortuosity, and derangement were performed. These were compared with manual measurements. Features were used to classify images into the three subject groups. Results: Comparison of automatic and manual measures for distance, width, tortuosity, and derangement had correlations of r = 0.583, 0.624, 0.495 ( p < 0.001), and 0.195 ( p = 0.040). For automatic measures, correlations were found between width and intercapillary distance, r = 0.374, and width and tortuosity, r = 0.573 ( p < 0.001). Significant differences between subject groups were found for all features ( p < 0.002). Overall, 75% of images correctly matched clinical classification using semi‐automated features, compared with 71% for manual measurements. Conclusions: Semi‐automatic and manual measurements of distance, width, and tortuosity showed moderate (but statistically significant) correlations. Correlation for derangement was weaker. Semi‐automatic measurements are faster than manual measurements. Semi‐automatic parameters identify differences between groups, and are as good as manual measurements for between‐group classification.