AbstractNailfold capillaroscopy is a useful tool to diagnose endocrine, cardiovascular, neuropsychiatric, rheumatic and other diseases. Diagnoses are made on the presence or absence of certain types of capillary loops which are classified according to their shape. We have surveyed six clinicians, asking them to classify 217 capillary loops, in order to establish quantitative nailfold capillary loop classification criteria. The participating clinicians were not unanimous about the classification of any capillary, and there was no consensus about the class of 17% of the capillary loops. Some of the clinicians classified two occurrences of the same shape differently. This clearly demonstrates the need for well established classification criteria.
Nailfold capillary loop classes can be divided into two major groups: Descriptive Classes (DC); "cuticulis", "open", "tortuous", "crossed", "bizarre" and "bushy", and Label Classes; "enlarged", "elongated" and "giant". Furthermore label classes can be divided into two groups, Width Anomaly Classes (WAC); "enlarged" and "giant", and Length Anomaly Class (LAC), "elongated". While descriptive classes give information about the shape of a loop, label classes emphasise dimensional anomalities of a loop. Assignment of a loop with one the descriptive or label classes causes information loss about the dimensions or shape, respectively. In order to preserve as much information as possible within a class, we propose a new class system that contains 17 classes which are the combination of WAC, LAC and DC.
We propose quantitative classification criteria for commonly used classes: "cuticulis", "open", "tortuous", "elongated" and "giant". Although the class "enlarged" can be expressed quantitatively, inappropriate assignments of "enlarged" by the paricipating clinicians have not allowed us to set quantitative classification criteria. While definition of the class "crossed" is purely qualitative, a classification mechanism that is neither qualitative nor quantitative is proposed for "bizarre" and "bushy" loops.
We propose the use of pattern recognition algorithms that are based on the evaluation of the capillary shape parameters such as loop length, loop width, limb width, the curvature, orientation, etc., to classify capillaries. By the use of mathematical morphology, skeletonization, topological relations, feature vectors, in an hierarchical structure, the software TANCCAS (The Automated Nailfold Capillary Classification and Analysis System) has been developed to calculate the shape parameters and to classify the capillary loops.
These algorithms have been implemented on a Pentium PC and have resulted in an 88% accuracy level which is compared to participating clinicians' overall classifications of the test images.
|Date of Award||Feb 1998|
- Nailfold capillaroscopy
- classification criteria
- diagnosis tool
- image processing