Introduction
Nailfold videocapillaroscopy is a very useful non-invasive diagnostic and monitoring tool in rheumatic and autoimmune diseases. Its aim is to observe the evolution of structural and morphological changes occurring in blood capillaries, allowing, among other things, the identification of disease patterns associated with systemic sclerosis or Raynaud’s phenomenon.
Despite being a simple technique, it has a significant limitation: inter-observer variability. The acquired images must be examined by healthcare professionals to correctly identify and describe abnormalities in capillary structure, which is subject to their own experience.
Automated software-based systems can be of great help in overcoming this limitation. Capillary.io has developed the CAPI-Score algorithm, which reduces variability in capillaroscopy interpretation. CAPI-Score identifies disease patterns by eliminating observer subjectivity. Its accuracy has been tested in a cohort of samples from patients with systemic sclerosis and Raynaud’s phenomenon, and the results have been successfully published in the journal Rheumatology. Read on to learn more!
What Is the CAPI-Score Algorithm
The Fast-Track algorithm, proposed by expert members of the European Alliance of Associations for Rheumatology (EULAR), allows capillaroscopists of any level to differentiate, with some reliability, a scleroderma pattern from a non-scleroderma one. CAPI-Score goes even further and classifies systemic sclerosis samples into early, active, or late scleroderma patterns, following EULAR criteria, and distinguishes between normal and nonspecific patterns.
CAPI-Score is inspired by the Fast-Track algorithm, which has limitations in its objectivity by not specifying entirely clear cutoff points. To eliminate the subjectivity of Fast-Track, CAPI-Score uses quantitative criteria to assign or discard patterns in a completely objective manner with a simple algorithm.
To achieve this, CAPI-Score is based on a decision tree that allows assigning each capillaroscopy to one of the following categories: normal, nonspecific, SSc-early, SSc-active, and SSc-late. The algorithm combines 4 objectively and quantitatively defined rules to classify patterns, analyzing different variables that establish a criterion to divide the data dichotomously into one category or another. The selected variables are: density, proportion of capillaries with abnormal shape (formerly known as branching, arborified or ramified capillaries), giant capillaries, tortuosities, and hemorrhages. These variables are grouped according to consensus patterns, using expert opinion as a reference to identify distinctive features and define a cutoff point.
Nine capillaroscopists participated in the study and analyzed 1,040 capillaroscopies from patients with systemic sclerosis or Raynaud’s phenomenon. Each of these capillaroscopies was randomly assigned to three of the nine experts, who manually and blindly studied them, only viewing the capillaroscopy images and no further patient details. Capillaroscopies where consensus was reached among two or more capillaroscopists in the classification of the disease pattern were considered the gold standard. Finally, 851 capillaroscopies that obtained consensus were used for the analysis.
A Decision Tree Based on 4 Rules That Allows Classifying Capillaroscopies
The four rules of the decision tree used by CAPI-Score to classify the different patterns are as follows:
Rule 1: Distinction between scleroderma patterns and non-scleroderma patterns. A capillaroscopy is indicative of a scleroderma pattern whenever the capillary density is ≤6 capillaries per mm and/or giant capillaries are detected and/or the proportion of capillaries with abnormal shapes is >10%.
Rules 2 and 3: Classification of scleroderma patterns (early vs. active vs. late). When the density is ≥5 capillaries per mm, the pattern is SSc-early, unless giant capillaries are ≥10% or abnormal shapes are ≥5%. In such cases, the pattern becomes SSc-active.
On the other hand, when the density is <5 capillaries per mm, the pattern is initially SSc-late. However, if giant capillaries are ≥33% or abnormal shapes are ≤7%, the pattern becomes SSc-active. Nevertheless, regardless of these conditions, if ≤7% of the capillaries are giant or ≥15% of the capillaries are abnormal, the pattern remains SSc-late.
Rule 4: Classification of non-scleroderma patterns (normal vs. nonspecific). When the density is >6 capillaries per mm, the pattern is nonspecific when one or more of the following conditions are met: percentage of tortuosities ≥20%; presence of hemorrhages; percentage of capillaries with abnormal shape ≥2%. Otherwise, the pattern is normal.
Capillary.io uses artificial intelligence to extract quantitative variables from capillaroscopies and apply the CAPI-Score algorithm to suggest a pattern for the capillaroscopy. The software analyzes capillary density and the proportion of giant capillaries, abnormal shapes, hemorrhages, and tortuosities, allowing a pattern to be assigned to any capillaroscopy quantitatively.
The Accuracy of CAPI-Score
Experimental results show that the probability that the algorithm and an expert capillaroscopist differ when classifying a pattern is significantly reduced.
To calculate the algorithm’s accuracy in classifying the different disease patterns, the reliability of each step was evaluated by analyzing the concordance between the opinion of expert capillaroscopists, considered as reference, and the algorithm’s prediction. 851 capillaroscopies were analyzed using the software, and it was observed that the accuracy of the classification between scleroderma patterns and non-scleroderma patterns applying the first rule was 0.88, and the agreements were always above 85%.
Within the stratification of the different scleroderma patterns, the accuracy was 0.82, and agreements were always above 80%, except for the late pattern.
In the last step of the algorithm, the classification of non-scleroderma patterns into normal or nonspecific, the accuracy was 0.73, and agreements were above 70%. In all cases, if the consensus among the three capillaroscopists is unanimous, the accuracy and agreements increase considerably.
Advantages of Using Capillary.io
Capillary.io is a platform for capillaroscopy that is agile, simple, and, as has been demonstrated, objective. This innovative tool, in addition to easily capturing and organizing capillaroscopy photos with automatic calibration, analyzes the detected capillaries and assigns a pattern to the different samples thanks to its algorithms and models based on deep learning.
It is scientifically validated and backed by reproducible results: it reduces inter-observer variability, eliminating the main limitation of nailfold videocapillaroscopy. Additionally, it is capable of accurately detecting disease patterns.
Capillary.io emerges as a solution to the lack of homogenization in capillaroscopy analysis, allowing the full potential of the technique to be exploited. As if that weren’t enough, it reduces the time dedicated to analysis and decreases its complexity.
Want to know more about Capillary.io? Explore our website or signup here!