Supervised Classification of Respiratory Ciliary Motion from Video Microscopy

Reliable assessment of respiratory ciliary motion is essential for evaluating mucociliary clearance and supporting the diagnosis of respiratory dysfunctions such as Primary Ciliary Dyskinesia (PCD). Current analysis relies on expert visual inspection of high-speed videos, which is subjective and difficult to scale. This study proposes a supervised, phenotype-driven framework for classifying ciliary motion from video microscopy, comparing two spatial preprocessing strategies: restricting feature extraction to the masked ciliated region and using the entire image, as well as different motion descriptors. Basal motility versus hypomotility was best classified with masked frequency descriptors and Support Vector Machine with Radial Basis Function (SVM-RBF), reaching 78.06% balanced accuracy. Hypomotility versus secondary ciliary alteration required combined frequency and autoregressive descriptors, reaching 78.81% balanced accuracy with logistic regression, while Random Forest achieved the highest macro F1-score of 79.24%. Thus, the optimal pipeline is phenotype-specific: masked frequency descriptors best separate basal motility from hypomotility, whereas combined spectral and temporal descriptors better distinguish abnormal phenotypes.