A non-parametric segmentation methodology for oral videocapillaroscopic images

Area of Science:

• Biomedical Engineering
• Medical Imaging Analysis
• Microcirculation Research

Background:

• Oral videocapillaroscopy is crucial for assessing peripheral microcirculation.
• Accurate segmentation of capillaries and background in microphotographs is essential for diagnostic support.
• Current methods may require manual segmentation, which can be time-consuming and subjective.

Purpose of the Study:

• To develop and validate a non-parametric methodology for automated segmentation in oral videocapillaroscopy.
• To enhance diagnostic support for clinicians evaluating peripheral microcirculation.
• To objectively assess the performance of the proposed segmentation technique.

Main Methods:

• Utilized wavelet analysis and mathematical morphology for image preprocessing.
• Implemented an image segmentation algorithm minimizing luminosity variance between capillaries and background.
• Validated the methodology using real microphotographs and physician-provided segmentations.

Main Results:

• The methodology achieved high robustness, evidenced by precision-recall criteria (average precision 0.924, recall 0.923).
• The segmentation performance, measured by the Jaccard index, closely matched physician performance (mean 0.858, std dev 0.064).
• A leave-one-patient-out cross-validation approach confirmed the method's reliability.

Conclusions:

• The proposed non-parametric methodology offers a robust and automated solution for oral videocapillaroscopy image analysis.
• This technique can significantly aid clinicians in the diagnostic process of peripheral microcirculation evaluation.
• The findings suggest the potential for improved efficiency and objectivity in microcirculation diagnostics.