Background Nailfold capillary density is a useful measure in systemic sclerosis (SSc) classification and evaluation. Its manual detection may be time-consuming, hampering its use in largescale investigations. We evaluated a new automated system to assess the absolute nailfold capillary number.
Objectives To attest the instrumental reliability of automatic counting in SSc patients using nailfold video capillaroscopy (NVC) images.
Methods 75 NVC random images, from SSc patients, were blindly analyzed by four raters (2 less and 2 more experienced; raters: 1,2,3,4) from two European centers. Each rater was asked to define the region of interest (ROI) on the NVC images and to manually count the number of capillaries, according to the following instructions: upper bound placed on top of the longest capillary head and lower bound placed on half of the length of that longest capillary (figure 1); if the common branch of an abnormal shape (neoangiogenesis) is in ROI it is counted as being one; if the common branch is out of ROI it is counted as separate capillaries; if the capillary is on the edge of the vertical line of ROI, it is only counted when the head of the capillary is half in ROI; if the capillary head is on the edge of lower bound, it is counted as soon as the “head” part is in the ROI; all “heads” in the ROI are counted (not only distal row). The dedicated automated system (AUTOCAPI-ds medica, IT) also counted the number of capillaries in the same ROI (figure 1). Reliability between the manual and automatic counting was investigated per rater through intraclass correlation coefficient (ICC) and reported with 95% confidence interval (CI). External validation was obtained by multi-rating of the same set of images. Average difference between automated and manual counting per rater was calculated.
Results The ICC (95% CI) of manual versus automatic counting in ROI was 0.77 (0.61–0.86) for rater 1 (p<0.0001), 0.81 (0.71–0.88) for rater 2 (p<0.0001), 0.65 (0.50–0.76) for rater 3 (p<0.0001) and 0.81 (0.71–0.87) for rater 4 (p<0.0001). The average difference was -0.69 for rater 1, 0.04 for rater 2, -0.03 for rater 3 and 0.16 for rater 4.
Conclusions This is a first study to attest the reliability of a new automated system to calculate the absolute number of capillaries in a ROI arising from SSc NVC images. High performance of the new automated counting system was confirmed in pathological conditions (SSc).
Disclosure of Interest None declared